Cover page

Image Name: AI for the Planet (Image courtesy)

While AI could become an extra dimension for human creativity and welfare, it could also become a threat if

misused by human greed and fanatism. The picture depicts the hope that we will use AI to beneﬁt all life forms

and conserve our ecosystems and biodiversity.

Managing Editor Chief Editor Editorial Board Correspondence

Ninan Sajeeth Philip Abraham Mulamoottil K Babu Joseph The Chief Editor

Ajit K Kembhavi airis4D

Geetha Paul Thelliyoor - 689544

Arun Kumar Aniyan India

Sindhu G

Journal Publisher Details

Publisher : airis4D, Thelliyoor 689544, India

Website : www.airis4d.com

Email : nsp@airis4d.com

Phone : +919497552476

Editorial

by Fr Dr Abraham Mulamoottil

airis4D, Vol.2, No.11, 2024

www.airis4d.com

This edition also introduces the News Desk high-

lighting airis4D outreach initiative ”Igniting Young

Minds” led by Geetha Paul and Ninan Sajeeth Philip

through the Vigyan Jyothi program at Jawahar Navo-

daya Vidyalaya School in Pathanamthitta and St. Thomas

College, Kozhenchery. The event aimed to inspire

students toward science and technology, with support

from airis4D team members Sindhu G. and Jinsu Ann

Mathew.

This edition gives two articles on the 2024 No-

bel Prize in Physics. In ”The Legacy of Hopﬁeld

and Hinton,” Arun Aniyan explores the transformative

contributions of John Hopﬁeld and Geoﬀrey Hinton

to artiﬁcial intelligence, computational neuroscience,

and machine learning. Originally a physicist, Hop-

ﬁeld pioneered the Hopﬁeld network, using concepts

from statistical mechanics to explain neural memory

storage, bridging physics with computational neuro-

science. Drawing from a rich scientiﬁc lineage, Hin-

ton revolutionised neural networks with breakthroughs

like backpropagation and capsule networks, advancing

AI’s capacity for complex data representations. Their

academic ancestries reveal a legacy of interdisciplinary

inﬂuence, connecting physics, biology, and chemistry

through contributions from renowned scientists, high-

lighting the collaborative nature of scientiﬁc advance-

ment. Today, students of Hopﬁeld and Hinton continue

to lead in AI, exemplifying how mentorship, knowledge

transfer, and collaborative networks fuel progress and

sustain innovation across generations.

Blesson George introduces the signiﬁcance of the

2024 Nobel Prize for Physics. The 2024 Nobel Prize

in Physics awarded John Hopﬁeld and Geoﬀrey Hinton

for their foundational discoveries that enabled machine

learning with artiﬁcial neural networks, marking a shift

in how physics-driven methodologies impact AI. In-

spired by neurobiology and atomic spin physics, Hop-

ﬁeld’s neural network laid the groundwork for energy-

based models, while Hinton’s Boltzmann machine ex-

panded machine learning using principles from statis-

tical physics. This award stirred debate about AI’s

ﬁt within physics, as critics argue it diverges from

traditional boundaries and experimental evidence re-

quirements. However, the Nobel Committee defended

their choice, highlighting the role of physics in shaping

transformative technologies. This recognition reﬂects

a broader view of physics as an interdisciplinary force,

fueling advancements across various ﬁelds.

The article ”Will AI Bring an End to Humanity?”

by Ninan Sajeeth Philip discusses the potential dan-

gers of AI, especially regarding inequality, ethics, and

misuse in warfare. It emphasises that the threat isn’t

AI itself but how it’s used, particularly by powerful

entities that could exploit technology for surveillance,

autonomous weapons, and manipulation. The author

warns that unequal access to AI might deepen social

divides and fuel geopolitical conﬂicts. The article sug-

gests collaborative frameworks for responsible AI use,

equitable technology access, and ethical standards in its

development and deployment to address these issues.

In the Black Hole Stories-13, More About Black

Hole Spin Measurements, Ajit Kembhavi explores the

complex techniques for measuring black hole spins,

focusing on X-ray reﬂection spectra emitted from ac-

cretion disks surrounding supermassive black holes.

Building on the analysis of galaxy MCG-6-30-15, he

discusses how the spin impacts the shape of speciﬁc

X-ray emission lines, especially the 6.4 keV iron line.

Modiﬁcations in the reﬂection spectrum due to Comp-

ton scattering, ﬂuorescence, and relativistic eﬀects are

key to spin determination. Observations, especially

using high-resolution X-ray satellites, have shown that

many supermassive black holes have high spin val-

ues, often clustering above 0.5, indicating rapid rota-

tion. Further research and technological advancements

aim to reﬁne these measurements, potentially provid-

ing more insights into black hole characteristics and

behaviours.

In ”Learning is Better than Programming,” Linn

Abraham highlights the transformative contributions

of Geoﬀrey Hinton to artiﬁcial intelligence, speciﬁ-

cally through machine learning with neural networks.

The article discusses the 2024 Nobel Prize in Physics

awarded to Hinton and John J. Hopﬁeld for foun-

dational discoveries in neural networks, emphasising

Hinton’s role in advancing AI with innovations like

backpropagation, Restricted Boltzmann Machines (RBMs),

and Convolutional Neural Networks (CNNs). The ar-

ticle details AlexNet, a CNN model developed by Hin-

ton’s team, which achieved groundbreaking results in

the 2012 ImageNet Challenge, marking a turning point

in computer vision. By allowing the model to learn

features from data rather than relying on manual pro-

gramming, Hinton’s work showcased the power of deep

learning, emphasising the value of data and computa-

tional power over manual coding in solving complex

tasks.

The article ”Applications of AI in Astronomy and

Astrophysics” by Sindhu G highlights the transforma-

tive role of artiﬁcial intelligence in modern astron-

omy. AI’s integration with machine learning (ML)

and deep learning (DL) algorithms is enhancing data

analysis, classiﬁcation, and real-time detection, allow-

ing astronomers to analyse vast datasets generated by

advanced telescopes like the James Webb Space Tele-

scope. By automating processes such as star and galaxy

classiﬁcation, AI enables researchers to focus on sig-

niﬁcant discoveries, with predictive models aiding in

forecasting cosmic events. For instance, convolutional

neural networks (CNNs) eﬃciently classify stars and

galaxies, while recurrent neural networks (RNNs) im-

prove the accuracy of exoplanet detection and charac-

terisation. AI models help assess exoplanet habitabil-

ity by analysing transit and radial velocity data, paving

the way for groundbreaking discoveries about potential

life-supporting planets.

Moreover, AI enhances studies of variable stars,

X-ray binaries, and gravitational waves by automating

light curve analysis and improving signal detection ac-

curacy. This is especially crucial for capturing weak

gravitational wave signals amidst noise. AI also ad-

vances the detection of Fast Radio Bursts (FRBs) and

aids in multi-messenger astronomy, where AI frame-

works process data from diﬀerent wavelengths to cap-

ture associated phenomena. In cosmology, AI-powered

simulations allow researchers to model dark matter

and understand its role in cosmic evolution. AI-based

image processing techniques such as deblurring, de-

noising, and super-resolution improve telescope data

quality, enhancing our ability to explore the universe.

Through these diverse applications, AI is reshaping as-

tronomy and astrophysics by enabling discoveries once

considered impossible.

In the article ”Harnessing Artiﬁcial Intelligence

for Biodiversity Conservation: A New Era of Eco-

logical Understanding,” Geetha Paul explores how AI

revolutionises conservation eﬀorts. AI’s integration

into biodiversity research enhances data collection,

species identiﬁcation, and habitat monitoring, provid-

ing real-time insights that were previously challenging

to obtain. Technologies such as acoustic monitoring,

drones, and camera traps powered by deep learning al-

gorithms allow for precise wildlife tracking and proac-

tive conservation actions. Predictive analytics is vi-

tal in foreseeing risks like habitat loss due to climate

change, helping conservationists devise timely solu-

tions. Ethical considerations surrounding AI use, such

as data privacy and employment impacts, underscore

the need for responsible implementation. Ultimately,

AI’s transformative potential in conservation enables

more eﬀective, data-driven strategies that can secure

biodiversity for future generations.

The article ”Bronchopulmonary Dysplasia in Preterm

Infants: Addressing the Challenges and Advancements

iii

in Management” by Kalyani Bagri highlights the com-

plexities of Bronchopulmonary Dysplasia (BPD), a

chronic lung disease in preterm infants, especially those

born before 28 weeks. BPD can lead to severe respira-

tory issues, prolonged hospital stays, and neurodevel-

opmental delays, impacting long-term quality of life.

While advancements like surfactant therapy and less

invasive ventilation methods have improved outcomes,

preventing BPD remains challenging. Treatment op-

tions, such as postnatal corticosteroids, are cautiously

used due to potential adverse eﬀects. Additionally,

the article discusses the development of India-speciﬁc

predictive models to assess BPD risks better. Emerg-

ing treatments, including stem cell therapy and genetic

research, show promise for more personalised and ef-

fective management strategies for this high-risk group.

The article ”Ethical Implications of Using AI in

Scientiﬁc Research” by Jinsu Ann Mathew discusses

the transformative role AI plays in accelerating scien-

tiﬁc discoveries, while emphasizing key ethical chal-

lenges. Key concerns include ensuring accuracy and

trustworthiness, as AI models can produce mislead-

ing outcomes if trained on biased or incomplete data,

especially in critical ﬁelds like drug discovery and en-

vironmental science. Transparency in AI, or explain-

able AI (XAI), is also essential for fostering trust by

making AI’s decision-making processes interpretable.

Additionally, biases within AI systems can perpetu-

ate social inequalities if not managed responsibly. To

address these issues, the article suggests a framework

for responsible AI in research, advocating for rigor-

ous testing, transparency, bias mitigation, interdisci-

plinary collaboration, and continuous oversight to en-

hance AI’s ethical and reliable contribution to scientiﬁc

progress.

Igniting Young Minds

by News Desk

airis4D, Vol.2, No.11, 2024

www.airis4d.com

Geetha Paul and Ninan Sajeeth Philip with students after an outreach activity for students at Vigyan Jyothi

program at Jawahar Navodaya Vidyalaya School, Pathanamthitta and at St Thomas College, Kozhencheri

where Sindhu G and Jinsu Ann Mathew from airis4D also were present.

Contents

Editorial ii

Igniting Young Minds vi

I Artiﬁcial Intelligence and Machine Learning 1

1 The Legacy of Hopﬁeld and Hinton 2

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 John Hopﬁeld . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Geoﬀrey Hinton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 What does the Academic Ancestry Teach Us ? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Physics in the 2024 Nobel Prize for Physics 7

2.1 Contributions of John Hopﬁeld and Geoﬀrey Hinton . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Physics Employed in Hopﬁeld Network and Boltzmann Machine . . . . . . . . . . . . . . . . . . 8

2.3 Spin System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Will AI Bring an End to Humanity? 11

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.2 Inequality and Access to AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.3 Ethical Dilemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.4 Misuse of AI in Warfare . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.5 Mitigation Methods for Reducing the AI Divide . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

II Astronomy and Astrophysics 15

1 Black Hole Stories-13

More About Black Hole Spin Measurements 16

1.1 The Reﬂection Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.2 Modiﬁcations of the Reﬂection Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.3 The Distribution of the Spin Parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Learning is Better than Programming 20

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.2 ImageNet Challenge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.3 AlexNet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.4 The Legacy of AlexNet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Applications of AI in Astronomy and Astrophysics 23

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

CONTENTS

3.2 Classiﬁcation and Identiﬁcation of Celestial Objects . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.3 Exoplanet Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.4 Variable Stars and X-ray Binary Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.5 Gravitational Wave Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.6 Fast Radio Burst (FRB) Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.7 Cosmology and Dark Matter Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.8 Image Processing and Enhancement for Telescope Data . . . . . . . . . . . . . . . . . . . . . . . 27

3.9 Astroparticle Physics and Cosmic Ray Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

III Biosciences 30

1 Harnessing Artiﬁcial Intelligence for Biodiversity Conservation: A New Era of Ecolog-

ical Understanding 31

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.2 The Challenge of Biodiversity Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.3 The Role of AI in Biodiversity Conservation and Data Collection Eﬃciency . . . . . . . . . . . 32

1.4 Wildlife Monitoring and Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

1.5 Wildlife Conservation Tools and Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

1.6 Wildlife Conservation Tools and Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

1.7 Case Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

1.8 Behavioural Studies Through Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.9 Ethical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.11 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2 Bronchopulmonary Dysplasia in Preterm Infants: Addressing the Challenges and Ad-

vancements in Management 37

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.2 Mortality and Morbidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.3 Short-Term and Long-Term Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.4 Pathogenesis of BPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.5 Prevention and Treatment Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.6 Postnatal Corticosteroids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.7 Predictive Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.8 Emerging Treatments and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3 Ethical Implications of Using AI in Scientiﬁc Research 42

3.1 Ensuring Accuracy and Trustworthiness in AI Findings . . . . . . . . . . . . . . . . . . . . . . . 42

3.2 The Transparency Challenge: Opening the “Black Box” . . . . . . . . . . . . . . . . . . . . . . . 43

3.3 The Risk of Bias in AI Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.4 Moving Forward: A Framework for Responsible AI in Research . . . . . . . . . . . . . . . . . . 44

viii

Part I

Artiﬁcial Intelligence and Machine Learning

The Legacy of Hopﬁeld and Hinton

by Arun Aniyan

airis4D, Vol.2, No.11, 2024

www.airis4d.com

1.1 Introduction

When the Nobel Prize for Physics in 2024 was

declared, the major scientiﬁc journal Nature published

an article “Physics Nobel scooped by machine-learning

pioneers”. Whether it was scooped by people who

are not pure physicists is another debate. Many even

among the scientiﬁc community are not fully aware of

the background and history of both John Hopﬁeld and

Geoﬀrey Hinton. The community of young researchers

who have grown under both of them has already made

a huge impact, not only in the area of machine learning

but also in areas like chemistry, biology, medicine, and

so on.

It is interesting and important to understand the

ancestry as well as the legacy of both Hopﬁeld and

Hinton. Suppose one looks at the connection tree of

previous Nobel prize winners. In that case, it is intrigu-

ing to note that the people connected to the winners

come from the same group and have also won similar

prestigious awards.

1.2 John Hopﬁeld

John Joseph Hopﬁeld stands as one of the most

inﬂuential ﬁgures at the intersection of physics, neuro-

science, and computer science. Born on July 15, 1933,

in Chicago, Illinois, Hopﬁeld emerged from a family

deeply rooted in science – his father was a physicist

and his mother a professional chemist, setting the stage

for his future interdisciplinary achievements.

Hopﬁeld’s academic journey began at Swarth-

more College, where he earned his undergraduate de-

gree in physics. He went on to complete his Ph.D. in

physics at Cornell University in 1958 under the super-

vision of Nobel laureate Albert Overhauser. His early

career focused on solid-state physics, but his intellec-

tual curiosity would eventually lead him to revolution-

ize our understanding of neural networks.

Throughout his career, Hopﬁeld held prestigious

positions at multiple institutions. He served on the

faculty at the University of California, Berkeley (1961-

1964), Princeton University (1964-1980), and the Cal-

ifornia Institute of Technology (1980-1997), before re-

turning to Princeton. This academic journey reﬂects

not just institutional aﬃliations, but a remarkable intel-

lectual evolution from pure physics to the foundations

of computational neuroscience.

His most transformative work came in 1982 when

he introduced what became known as the Hopﬁeld net-

work – a breakthrough that fundamentally changed our

understanding of how neural networks can process in-

formation and store memories. This work brilliantly

bridged the gap between physics and biology, apply-

ing concepts from statistical physics to explain how

networks of neurons could exhibit collective computa-

tional properties. This contribution won him the Nobel

Prize in Physics.

The genius of the Hopﬁeld model lay in its con-

nection to physics. He showed that neural networks

could be understood through the lens of statistical me-

chanics, particularly using the concept of energy land-

scapes. By mapping neural activity to energy states, he

demonstrated how networks could converge to stable

states, representing memories or solutions to computa-

tional problems. This physical interpretation provided

1.3 Geoﬀrey Hinton

a powerful framework for understanding both artiﬁcial

and biological neural networks.

Before his neural network breakthroughs, Hop-

ﬁeld made signiﬁcant contributions to molecular biol-

ogy. His work on kinetic proofreading explained how

biological systems achieve high accuracy in molecu-

lar processes like DNA replication and protein syn-

thesis. He proposed that cells use energy-driven non-

equilibrium processes to enhance the accuracy of molec-

ular recognition, a concept that became fundamental to

our understanding of cellular processes.

In the 1990s and 2000s, Hopﬁeld turned his at-

tention to understanding how biochemical networks in

cells make decisions. He developed theoretical frame-

works for understanding how cellular circuits process

information and make reliable decisions despite the in-

herent noise in biological systems. His work helped

explain how cells can achieve reliable behavior from

unreliable components.

Figure 1 shows the ancestral tree of Hopkins and

one can see many prominent ﬁgures in Physics in his

ancestry. One of the most prominent ancestors close

to Hopkins is Paul Ehrenfest, an interesting scientiﬁc

ﬁgure who won the ﬁrst Nobel Memorial Prize in Eco-

nomics. Ehrenfest was originally trained in Chemistry

and his higher studies shifted toward Physics. His con-

tributions were mainly in thermodynamics, statistical

mechanics, and the foundations of quantum mechanics.

Going a level up, a major ﬁgure in ancestry is

Ludwig Boltzmann who put forward the deﬁnition of

entropy. Entropy is one of the key principles used by

Hopﬁeld in coming up with a neuron learning model

that was put forward. Most of all the major contri-

butions to the area of thermodynamics were done by

Boltzmann. It is worth noting that Boltzmann’s contri-

butions span both physics and chemistry.

Other prominent ﬁgures in the ancestral tree in-

clude people like Hermann van Helmotz, Max Planck,

A.M. Lyapunov, Theodore Lyman, Karl Schwarzschild,

Carl Runge, David Hilbert, and J.J. Thompson. They

are the major pillars of modern Physics and Mathemat-

ics.

Figure 1: Academic ancestral tree of John Hopkins

1.3 Geoﬀrey Hinton

Geoﬀrey Everest Hinton was born on December

6, 1947, in Wimbledon, London. He comes from a dis-

tinguished lineage of scientists – he is the great-great-

grandson of George Boole, whose Boolean algebra be-

came the foundation of computer science, and the son

of Howard Hinton, an entomologist. This rich scien-

tiﬁc heritage would later inﬂuence his groundbreaking

work in artiﬁcial intelligence.

Hinton’s academic journey began at King’s Col-

lege, Cambridge, where he studied experimental psy-

chology and physiology, graduating in 1970. He went

on to complete his Ph.D. in artiﬁcial intelligence from

the University of Edinburgh in 1978. During this pe-

riod, he developed a deep interest in understanding how

the human brain processes information and how these

principles could be applied to machines.

In the 1970s and 80s, Hinton pioneered new ap-

proaches to training neural networks, such as the back-

propagation algorithm, which allowed multi-layer neu-

ral networks to learn complex representations from

data. This was a signiﬁcant advancement, as earlier

neural network models were limited to shallow, single-

layer architectures. However, during this period, neural

networks fell out of favor with the majority of AI re-

searchers, who favored symbolic, rule-based systems.

Hinton persisted in his belief that neural networks held

the key to achieving human-like intelligence in ma-

chines, continuing to reﬁne his ideas and publish sem-

inal papers on the topic.

Hinton’s big breakthrough came in the late 2000s,

when he and his students demonstrated the power of

deep neural networks - neural networks with multi-

ple hidden layers. By leveraging increased computing

power and larger datasets, deep neural networks were

able to learn rich, hierarchical representations of data,

far surpassing the capabilities of shallow models.

Hinton’s work on deep belief networks, convo-

lutional neural networks, and related deep learning

techniques laid the foundations for the current AI rev-

olution. These methods have enabled transformative

applications in computer vision, natural language pro-

cessing, speech recognition, and beyond.

Throughout his career, Hinton has made signif-

icant theoretical contributions to understanding how

neural networks learn and represent information. His

work on topics like the ”credit assignment problem,”

neural network optimization, and the role of unsuper-

vised pre-training has advanced the fundamental sci-

ence behind deep learning. For example, his research

on ”capsule networks” proposed a new neural network

architecture that better captures the hierarchical, part-

whole relationships in data, an important step towards

more human-like perception and reasoning.

Hinton has held various prestigious positions. He

worked at the University of California San Diego, Carnegie

Mellon University, and notably at the University of

Toronto, where he has been a professor since 1987.

In 2013, he joined Google (now Alphabet) as part of

their acquisition of DNNresearch Inc., while maintain-

ing his position at the University of Toronto. He also

served as the Chief Scientiﬁc Advisor for the Vector

Institute for Artiﬁcial Intelligence.

The academic ancestral tree of Hinton is very in-

teresting in many respects. Hinton’s basic degree is

in Experimental Psychology with some background

in Chemistry, his ancestors span in a large area of sci-

ence. His doctoral advisor Hugh Christopher Longuet-

Higgins was a Professor of Chemistry who later in the

year of his career shifted to Cognitive Sciences which

was a relatively new area in the late 1960s.

One of Higgins’ advisors, Robert S. Mulliken, is

a Nobel Prize winner in Chemistry and a pioneer of

molecular orbital theory. Another ancestor of Hinton,

Charles Alfred Coulson, was both a mathematician and

a chemist. His doctoral students included prominent

ﬁgures like David Eisenberg and Peter Higgs, who also

won a Nobel Prize.

Higher level up another ancestor of Hinton, Fritz

Haber is another Nobel Laureate in Chemistry known

for the Haber-Bosch process. Interestingly he is known

as the “Father of Chemical warfare” because he came

up with deadly chemical methods used during the First

World War.

Apart from Hinton’s ancestors, his students are

modern pioneers in the area of artiﬁcial intelligence.

The top AI leaders today who are running large AI

1.4 What does the Academic Ancestry Teach Us ?

Figure 2: Academic ancestral tree of Geﬀ Hinton

companies are mostly students or postdocs who worked

under Hinton. Yan LeCun the pioneer of Convolutional

Neural Networks, was Hinton’s postdoc. Another post-

doc Peter Dayan is known for Q-Learning which is

the most important formulation used in reinforcement

learning.

Ilya Sutskever and Alex Krizhesky are two of

Hinton’s most famous students who came up with the

AlexNet architecture which drastically changed the fu-

ture of computer vision applications using deep learn-

ing. Ilya Sutskever is also the co-founder of OpenAI

and now leading Safe Superintelligence Inc.

This year the Nobel Prize for Chemistry was awarded

to DeepMind’s Demis Hassabis. A large fraction of

DeepMind’s researchers and engineers have worked in

Hinton’s lab. So the young generation that Hinton has

brought up is leading the path for AI today.

1.4 What does the Academic

Ancestry Teach Us ?

There is one fundamental pattern in the academic

ancestry of Nobel Prize winners - one or more direct or

indirect ancestors have always won Nobel Prizes. Even

though the correlation exists it is never a cause for a

researcher to receive a Nobel prize. But interestingly

enough this is a direct representation of a quote from

Issac Newton - “If I have seen further, it is by standing

on the shoulders of giants”. Every Nobel Prize winner

has stood on the shoulders of his ancestors who were

giants in their subjects.

Academic ancestry also shows that certain good

traits for research are passed on from earlier generations

to younger ones which has always advanced scientiﬁc

understanding. People who have won Nobel prizes

have qualities that are of extreme quality and naturally,

they get passed down to their students. This is one

factor that nurtures bright students who are extremely

successful.

The academic ancestry tree also shows how dif-

ferent areas of science are interconnected and how dis-

coveries in one area advance another. For example,

concepts of Entropy in Physics which is also rooted in

1.4 What does the Academic Ancestry Teach Us ?

Chemistry have been used to advance the development

of algorithms in AI. There are researchers whose ba-

sic training is in one subject and who later switch to a

diﬀerent subject to win Nobel Prizes. This is a perfect

example of skill and knowledge transfer which is a key

factor for a successful career.

Another key lesson from the ancestral tree is the

power of networking and collaboration. Many of the

early researchers who laid the foundations of basic sci-

ence made most of the major discoveries while in-

teracting with other researchers or even while having

debates with others. Sharing and collaborative ideas

always triumph and help advance science.

Reference

Academic Tree

The Encyclopedia of Computer Science, 4th Edi-

tion, Wiley Publications.

About the Author

Dr.Arun Aniyan is leading the R&D for Arti-

ﬁcial intelligence at DeepAlert Ltd,UK. He comes from

an academic background and has experience in design-

ing machine learning products for diﬀerent domains.

His major interest is knowledge representation and com-

puter vision.

Physics in the 2024 Nobel Prize for Physics

by Blesson George

airis4D, Vol.2, No.11, 2024

www.airis4d.com

The 2024 Nobel Prize in Physics was awarded to

Hopﬁeld and Hinton “for foundational discoveries and

inventions that enable machine learning with artiﬁcial

neural networks.” This decision has ignited extensive

discussions and stirred apprehensions across various

spheres. Social media and public forums have seen

heated debates questioning whether the Nobel Com-

mittee’s choice appropriately aligns with the traditional

boundaries of Physics. While most agree the award

honors groundbreaking and deserving contributions in

artiﬁcial intelligence—a ﬁeld rapidly transforming our

world—concerns remain about the suitability of award-

ing such work under Physics.

Critics argue that the Nobel Committee may have

succumbed to the current AI hype, especially since this

year’s Chemistry Nobel also recognized an AI-based

advancement, AlphaFold. Although undeniably revo-

lutionary, they contend that awarding a Physics Nobel

for AI deviates from past traditions, which typically

required experimental evidence for recognition. Fur-

thermore, questions have arisen about whether AI is

merely a tool within Physics rather than Physics itself,

especially since one of the laureates isn’t traditionally

a physicist. Despite these reservations, however, many

strongly believe that honoring Hopﬁeld and Hinton was

a well-considered and forward-thinking choice.

In its popular press release, the Nobel Committee

emphasized that ”they used physics to ﬁnd patterns in

information,” presenting a perspective that counters the

criticism of AI merely being a tool for Physics. While

some argue that AI serves only as an auxiliary tool in

Physics, the Committee recognized that the pioneers

of artiﬁcial intelligence leveraged core principles of

Physics to develop foundational methods that underpin

today’s powerful machine learning. This viewpoint

reﬂects the Committee’s broader perspective on how

physics-driven methodologies can lead to transforma-

tive discoveries well beyond traditional domains.

2.1 Contributions of John Hopﬁeld

and Geoﬀrey Hinton

John Hopﬁeld made a signiﬁcant breakthrough in

1982 with the development of the Hopﬁeld network,

one of the earliest forms of artiﬁcial neural networks.

Inspired by neurobiology and molecular physics, this

network demonstrated how a system of interconnected

nodes could store and retrieve information. The Hop-

ﬁeld network is based on principles of atomic spin

physics, where each atom behaves like a tiny mag-

net, contributing to a material’s overall properties.

Hopﬁeld used this analogy to design a network that

operates similarly to an energy-based spin system in

physics, where training involves adjusting the connec-

tions between nodes so that speciﬁc images or patterns

represent low-energy states. When presented with a

distorted or incomplete image, the network iteratively

updates node values, minimizing energy until it closely

matches a stored image, thus enabling pattern recogni-

tion and recall.

Geoﬀrey Hinton expanded upon Hopﬁeld’s con-

cepts in 1984 by creating the Boltzmann machine, a

model that allowed computers to learn from data us-

ing principles from statistical physics. Unlike tradi-

tional programming, the Boltzmann machine learns

directly from examples, drawing on the science of sys-

2.2 Physics Employed in Hopﬁeld Network and Boltzmann Machine

tems made up of numerous, similar elements. This

machine is trained by being exposed to typical exam-

ples that reinforce the patterns within the data. As

a result, the Boltzmann machine can classify images

and generate new examples that reﬂect the patterns on

which it was trained, marking a signiﬁcant step forward

in machine learning and pattern recognition.

Energy-Based Models:

The Hopﬁeld network and Boltzmann machine are

both energy-based models that rely on the concept of

low-energy states to function. In the Hopﬁeld network,

an energy function is used to identify the stable or low-

energy state that corresponds to a stored pattern. This

stable state represents the memory or pattern that the

network recalls when given an incomplete or noisy in-

put. The illustration provided eﬀectively demonstrates

how the network ”ﬁnds” a stored pattern by minimizing

energy.

Similarly, in the Boltzmann machine, the energy

function is central to learning. It enables the network

to model data distributions by adjusting the weights

of connections between nodes. By minimizing energy,

the Boltzmann machine can learn and generate patterns

that reﬂect the underlying structure of the data. In

both models, the concept of a low-energy state is key,

as it allows the networks to stabilize into meaningful

representations or memories.

2.2 Physics Employed in Hopﬁeld

Network and Boltzmann Machine

2.2.1 Statistical Mechanics

In the Hopﬁeld network, statistical mechanics is

applied through an energy function similar to the Ising

model, a foundational concept in statistical mechanics

for modeling ferromagnetic interactions. The network

consists of binary neurons x

∈ {−1, 1}, each of which

interacts with others through symmetric weights w

The energy function E of the network is deﬁned as:

E = −

∑

i,j

−

∑

where b

represents the bias terms acting as exter-

nal ﬁelds. This energy function describes the stability

of patterns in the network, where lower energy states

correspond to stable or ”remembered” patterns. When

the network receives an input pattern, it iteratively up-

dates the states x

of neurons to reduce the energy E,

guiding the system towards a minimum energy state

associated with a stored pattern. This energy mini-

mization process mirrors physical systems in statisti-

cal mechanics, where systems evolve towards states of

minimum energy, thus achieving stability.

In Boltzmann machines, statistical mechanics prin-

ciples are fundamental, particularly through the use of

the Boltzmann distribution and energy minimization

concepts from thermodynamics. Boltzmann machines

are designed to model the probability distribution of

data by adjusting weights to minimize an energy func-

tion, allowing them to ”learn” from examples.

The energy function E of a Boltzmann machine

for a conﬁguration of visible and hidden units x is given

by:

E(x) = −

∑

i,j

−

∑

where w

are the symmetric weights between

units, and b

are bias terms. Using this energy func-

tion, the probability of a given state x in a Boltzmann

machine is modeled by the Boltzmann distribution:

P (x) =

−E(x)/T

where T represents the ”temperature” parame-

ter, controlling the system’s randomness, and Z is the

partition function, which normalizes the distribution.

The training process involves adjusting weights w

to minimize the Kullback–Leibler divergence between

the model’s distribution and the data distribution, ef-

fectively learning the data’s structure. The Boltzmann

2.3 Spin System

machine uses stochastic updates where each unit’s state

is updated probabilistically, following the sigmoid ac-

tivation function:

P (x

= 1∣x

−i

) =

1 + e

−∆E/T

where ∆E is the change in energy if unit x

were

ﬂipped. This probabilistic updating allows the Boltz-

mann machine to escape local energy minima, enabling

it to approximate global minima over time and reach

thermal equilibrium, thereby learning the distribution

of input data .

2.3 Spin System

Spin systems are models used in statistical me-

chanics and condensed matter physics to represent mag-

netic properties of materials. In these systems, ”spins”

represent magnetic moments of particles (such as elec-

trons) and are typically modeled as binary variables

with values +1 or −1, representing ”up” or ”down”

states. These spins interact with each other, forming

complex patterns based on temperature and interaction

strength.

1. Ising Model

The Ising model is a fundamental spin system

that considers a lattice where each site (or par-

ticle) has a spin s

that can take values +1 or

−1. The Hamiltonian (energy function) for the

Ising model, considering only nearest-neighbor

interactions, is given by:

H = −J

∑

⟨i,j⟩

−

∑

where J is the interaction strength, ⟨i, j⟩ denotes

pairs of neighboring spins, and h

represents an

external magnetic ﬁeld at site i. For ferromag-

netic interactions (J > 0), neighboring spins pre-

fer to align, while for antiferromagnetic interac-

tions (J < 0), they prefer to be anti-aligned.

2. Energy Minimization

In spin systems, conﬁgurations with lower en-

ergy are more stable, particularly at low tem-

peratures. At thermal equilibrium, spins tend

to settle into conﬁgurations that minimize the

system’s total energy.

3. Phase Transitions

Spin systems exhibit phase transitions, such as

a change from a disordered (paramagnetic) to

an ordered (ferromagnetic or antiferromagnetic)

phase as temperature changes. For instance, in

the 2D Ising model without an external ﬁeld, a

critical temperature T

exists at which the system

spontaneously transitions from a disordered state

to an ordered state.

4. Relation to Neural Networks

Spin systems, particularly the Ising model, are

closely related to Hopﬁeld networks and Boltz-

mann machines in machine learning. In these

neural networks, binary neuron states and their

interactions mimic spins and their interactions

in a spin system. Energy minimization in neural

networks is analogous to ﬁnding stable conﬁg-

urations in spin systems, with low-energy states

corresponding to stable patterns or memories.

5. Boltzmann Distribution

Spin conﬁgurations at thermal equilibrium fol-

low the Boltzmann distribution:

P (conﬁguration) =

−H/(k

T )

where H is the system’s Hamiltonian, T is the

temperature, k

is the Boltzmann constant, and

Z is the partition function that normalizes the

probability distribution.

2.4 Conclusion

The 2024 Nobel Prize in Physics marks a signif-

icant milestone, underscoring the ongoing dissolution

of traditional boundaries between scientiﬁc disciplines.

In an era increasingly deﬁned by the AI revolution, this

award highlights the pivotal role that Physics plays in

the development of artiﬁcial intelligence and neural

networks. Such recognition broadens our understand-

ing of Physics, extending its foundational principles

into transformative technologies that impact us all.

The prize serves as a reminder that interdisciplinary

2.4 Conclusion

advancements can fuel innovation, with beneﬁts that

reach across ﬁelds and into everyday life.

References

1. Lucas Bottcher and Gregory Wheeler, ”Statisti-

cal Mechanics and Artiﬁcial Neural Networks:Principles,

Models, and Applications” arXiv:2405.10957v1

[cond-mat.dis-nn] 2024

2. Why is Nobel Prize in Physics awarded to Geoﬀ

Hinton and John Hopﬁeld?

3. Press release-The nobel prize

About the Author

Dr. Blesson George presently serves as an

Assistant Professor of Physics at CMS College Kot-

tayam, Kerala. His research pursuits encompass the

development of machine learning algorithms, along

with the utilization of machine learning techniques

across diverse domains.

Will AI Bring an End to Humanity?

by Ninan Sajeeth Philip

airis4D, Vol.2, No.11, 2024

www.airis4d.com

While Artiﬁcial Intelligence (AI) has the potential

to revolutionize various sectors positively, it also poses

signiﬁcant threats to humanity. This paper explores the

negative aspects of AI, particularly focusing on issues

of inequality, ethical dilemmas, and the potential for

misuse in warfare. The discussion emphasizes that the

real danger lies not in AI itself but in how it is wielded

by those with access to advanced technologies. Ad-

ditionally, this paper outlines strategies to mitigate the

AI divide and promote equitable access to AI technolo-

gies.

3.1 Introduction

As AI continues to evolve and integrate into var-

ious aspects of society, concerns regarding its poten-

tial negative impacts have gained prominence. Al-

though AI can enhance productivity and eﬃciency, it

also raises critical questions about ethics, inequality,

and control. This paper discusses how AI can become

a threat to humanity, particularly in contexts where

disparities exist between those who possess advanced

technologies and those who do not. The implications

of these disparities are profound, as they can lead to

social unrest, exacerbate existing inequalities, and cre-

ate new forms of conﬂict. Therefore, it is essential to

critically examine the multifaceted threats posed by AI.

3.2 Inequality and Access to AI

One of the most pressing concerns regarding AI is

its potential to exacerbate existing inequalities. Access

to AI technologies is often limited to aﬄuent individ-

uals or nations, creating a divide that could lead to

signiﬁcant social unrest. As industries increasingly

adopt AI for automation and eﬃciency, low-skilled

workers are at risk of displacement while high-skilled

positions become more lucrative [7]. For example,

during the COVID-19 pandemic, many businesses ac-

celerated their adoption of AI tools for automation,

leading to signiﬁcant job losses in sectors such as retail

and hospitality [2]. This economic shift may result in

an ”M-shaped” wealth distribution where wealth be-

comes concentrated among a small elite while a sig-

niﬁcant portion of the population faces unemployment

or underemployment. The disparity in access to tech-

nology can foster resentment and conﬂict within soci-

eties, potentially leading to protests or civil unrest as

marginalized groups demand equitable opportunities.

Countries that lead in AI development may gain

substantial geopolitical advantages, potentially leading

to a new form of colonialism where technologically

advanced nations exploit less developed regions [6].

For instance, during conﬂicts like the recent Israel-

Palestine escalation in 2023, advanced surveillance and

drone technologies were employed by Israel to mon-

itor movements and conduct targeted operations with

precision [11]. Such technological superiority raises

ethical questions about power dynamics and the conse-

quences for less technologically advanced adversaries.

The ability of one nation to leverage superior technol-

ogy against another not only shifts the balance of power

but also sets a dangerous precedent for future conﬂicts

where technology becomes a primary weapon. This

dynamic can create a cycle of dependency where less

developed nations are forced to rely on more advanced

3.3 Ethical Dilemmas

powers for security and stability.

Moreover, this inequality extends beyond national

borders; within countries, marginalized communities

often lack access to cutting-edge technology. Dispari-

ties in education and resources mean that only a select

few can harness the beneﬁts of AI advancements. For

instance, rural areas may lack internet access or tech-

nological infrastructure necessary for participating in

an increasingly digital economy. This digital divide

can perpetuate cycles of poverty and limit opportuni-

ties for upward mobility. As access becomes increas-

ingly tied to economic power, those without means

may ﬁnd themselves further excluded from societal ad-

vancements driven by AI.

3.3 Ethical Dilemmas

AI raises numerous ethical questions that society

must address as its capabilities expand. One major con-

cern is the autonomy granted to AI systems in decision-

making processes. There are fears that as these systems

become more autonomous, humans may lose control

over critical decisions that aﬀect lives. The possibil-

ity of autonomous weapons systems operating without

human intervention presents a dire scenario where ma-

chines could make life-and-death decisions based on

algorithms rather than ethical considerations. For ex-

ample, during military operations in Ukraine, there

have been reports of autonomous drones being used

for targeted strikes without direct human oversight [8].

This raises profound moral questions about account-

ability: if an autonomous system makes a mistake that

results in civilian casualties, who is responsible?

Furthermore, AI technologies enable unprecedented

levels of surveillance that can infringe on individual

privacy rights. Governments may utilize AI for mon-

itoring citizens under the guise of security, leading to

authoritarian practices reminiscent of dystopian soci-

eties [7]. The implementation of social credit sys-

tems in countries like China exempliﬁes how AI can be

used to enforce conformity and suppress dissent [12].

Such systems leverage data analytics to monitor be-

havior continuously; those who deviate from accepted

norms may face penalties or restrictions on their free-

doms. This creates a chilling eﬀect on free speech

and personal expression as individuals become wary

of surveillance.

The ethical implications extend into algorithmic

bias as well; if AI systems are trained on biased data

sets, they can perpetuate existing prejudices or cre-

ate new forms of discrimination. For instance, facial

recognition technologies have been shown to misiden-

tify individuals from minority groups at higher rates

than their white counterparts. Such biases can have

real-world consequences when these technologies are

deployed in law enforcement or hiring practices. Thus,

addressing these ethical dilemmas requires robust frame-

works that ensure transparency and accountability in

AI development.

Additionally, there is an urgent need for interdis-

ciplinary collaboration among technologists, ethicists,

policymakers, and civil society organizations to create

guidelines that govern AI’s use responsibly. Engaging

diverse stakeholders can help mitigate risks associated

with biased algorithms and ensure that technological

advancements beneﬁt all segments of society rather

than exacerbating existing inequalities.

3.4 Misuse of AI in Warfare

The application of AI in military contexts raises

signiﬁcant ethical and moral concerns that cannot be

overlooked. The development of autonomous weapons

systems poses unique threats as these machines can op-

erate without human oversight. These systems have the

potential to engage in combat autonomously based on

pre-deﬁned algorithms or real-time data analysis [10].

If these machines are programmed with ﬂawed algo-

rithms or biased data, they could make erroneous deci-

sions leading to unintended casualties. Recent reports

indicate that during conﬂicts like those in Syria and

Libya, autonomous drones have been deployed with

minimal human intervention [10]. Such developments

highlight the urgent need for international regulations

governing the use of autonomous weapons.

Moreover, AI’s ability to analyze vast amounts of

data quickly can lead to rapid decision-making pro-

cesses that may escalate conﬂicts without adequate hu-

3.5 Mitigation Methods for Reducing the AI Divide

man judgment. The speed at which decisions are made

could outpace diplomatic eﬀorts aimed at de-escalation

[6]. For instance, during the recent Israel-Hamas con-

ﬂict in 2024, rapid deployment of drone strikes based

on real-time data led to signiﬁcant civilian casualties

and raised international concerns about proportionality

in warfare [1]. As military strategies increasingly rely

on automated systems for targeting and engagement,

there is a growing risk that conﬂicts could spiral out

of control due to miscalculations or overreliance on

technology.

The potential for cyber warfare further compli-

cates this landscape; adversaries could deploy AI-driven

cyber-attacks against critical infrastructure such as power

grids or communication networks. Such attacks could

destabilize nations without traditional military engage-

ment while causing widespread chaos and disruption

[9]. Additionally, misinformation campaigns powered

by AI algorithms can manipulate public perception

during conﬂicts by spreading false narratives rapidly

across social media platforms [13]. These tactics not

only undermine trust but also complicate diplomatic

resolutions.

To mitigate these risks associated with military

applications of AI technology requires global coop-

eration among nations aimed at establishing treaties

regulating autonomous weapons systems similar to ex-

isting agreements on chemical and biological weapons.

Furthermore, fostering dialogue between military lead-

ers and ethicists could help ensure that technological

advancements align with humanitarian principles.

3.5 Mitigation Methods for Reducing

the AI Divide

Addressing the disparities created by AI requires

targeted strategies aimed at promoting equitable access

and responsible use:

Enhancing Data Quality and Diversity: Im-

proving the quality and diversity of datasets used

in training algorithms helps prevent bias and

ensures fairer outcomes across diﬀerent demo-

graphic groups [6]. Organizations should focus

on collecting comprehensive datasets that repre-

sent various populations accurately; this includes

obtaining additional data points from underrep-

resented groups which can signiﬁcantly reduce

algorithmic bias.

Implementing Fairness and Transparency Stan-

dards: Establishing standards for fairness en-

sures accountability in decision-making processes

involving AI systems [7]. Companies should

adopt frameworks requiring transparency so users

understand how outcomes are derived from algo-

rithms; this could involve using explainable ar-

tiﬁcial intelligence (XAI) techniques which pro-

vide insights into how decisions are made.

Promoting Inclusive Education and Training:

Enhancing access to education focused on data

science is essential for reducing skill gaps related

to technology use [5]. Initiatives like coding boot

camps speciﬁcally targeting underserved com-

munities can empower individuals with neces-

sary skills while fostering a more diverse work-

force capable of contributing meaningfully within

tech sectors.

Encouraging Diverse Teams in Development:

Diverse teams bring varied perspectives which

help identify biases early during development

phases [10]. Companies should prioritize hir-

ing practices promoting diversity across gender

identity/race/socioeconomic backgrounds within

their engineering teams; this diversity fosters

creativity while ensuring products cater eﬀec-

tively towards broader audiences.

Responsible Deployment: Engaging stakehold-

ers before deploying technologies ensures con-

sideration for potential societal impacts [4]. Or-

ganizations should develop policies protecting

vulnerable populations from adverse eﬀects while

maximizing beneﬁts derived from their tech-

nologies; this might include setting up advisory

boards comprised of community representatives

who provide insights into local needs/concerns

regarding new deployments.

Public Policy Frameworks: Governments should

establish regulations mandating fairness/account-

3.6 Conclusion

ability standards within AI applications [4]. Poli-

cies might include requirements for regular au-

dits assessing impacts across diﬀerent demo-

graphic groups; additionally fostering interna-

tional cooperation on regulatory frameworks can

help address cross-border challenges posed by

these technologies.

Leveraging Technology for Social Good: Fo-

cusing speciﬁcally on applications addressing

societal issues helps bridge gaps created by tech-

nological advancements [3]. For instance utiliz-

ing machine learning algorithms analyzing health

data from underserved populations allows pub-

lic health interventions tailored towards speciﬁc

community needs thereby improving overall out-

comes signiﬁcantly.

3.6 Conclusion

Addressing the AI divide requires a multifaceted

approach involving improvements in data quality stan-

dards, transparency initiatives, inclusive education, di-

verse teams responsible deployment practices, support-

ive public policies, leveraging technology for social

good, etc. By implementing these strategies collec-

tively we can work towards future where artiﬁcial in-

telligence serves tool enhancing equality rather than

exacerbating existing disparities.

References

[1] Various Authors. Casualties of the israel–hamas

war. Wikipedia, 2024.

[2] Bikash Chatterjee. Precision medicine: Promis-

ing future as ai, biomarkers and technology ...

Clinical Trials Arena, 2023.

[3] Anil Kumar et al. Machine learning approaches

for early diagnosis of neurodevelopmental disor-

ders. Journal of Neural Engineering, 2023.

[4] Priya Sinha et al. Ai-based biomarkers for autism

spectrum disorder: A review. Frontiers in Psy-

chology, 2023.

[5] Richard Sutton et al. Cognitive modeling using

artiﬁcial intelligence: Applications in education.

Educational Technology Research and Develop-

ment, 2023.

[6] Marcus Gullberg. Exploring the impact of arti-

ﬁcial intelligence (ai) on society. Graphite Note,

2021.

[7] Bernard Marr. What is the impact of artiﬁcial

intelligence (ai) on society? Forbes, 2021.

[8] BBC News. Ukraine war : How autonomous

drones are changing warfare, 2023. News Article.

[9] James Robinson. Cyber warfare : A new

paradigm. International Security Review, 2019.

[10] John Sullivan. The role of autonomous weapons

in modern warfare. Military Review, 2023.

[11] Noah Sylvia. The israel defense forces’ use of

ai in gaza: A case of misplaced purpose. Rusi,

2024.

[12] Michael Cheng-Tek Tai. The impact of artiﬁ-

cial intelligence on human society and bioethics.

PMC Journal, 2020.

[13] Clint Tucker. Social media manipulation : A new

threat landscape. Journalism Studies, 2017.

About the Author

Professor Ninan Sajeeth Philip is a Vis-

iting Professor at the Inter-University Centre for As-

tronomy and Astrophysics (IUCAA), Pune. He is also

an Adjunct Professor of AI in Applied Medical Sci-

ences [BCMCH, Thiruvalla] and a Senior Advisor for

the Pune Knowledge Cluster (PKC). He is the Dean

and Director of airis4D and has a teaching experience

of 33+ years in Physics. His area of specialisation is

AI and ML.

Part II

Astronomy and Astrophysics

Black Hole Stories-13

More About Black Hole Spin Measurements

by Ajit Kembhavi

airis4D, Vol.2, No.11, 2024

www.airis4d.com

In BHS12 we described how the spin of a black

hole can aﬀect the width of X-ray emission lines emit-

ted by reﬂection from the inner region of an accretion

disc around the black hole. We considered the speciﬁc

case of the active galaxy MCG-6-30-15 and showed

how the spin of the black hole at the centre of the galaxy

can be estimated from the shape of a line emitted by

iron. In this story we will consider further examples

of spin determination for supermassive black holes.

Much of the material here is based on two excellent

reviews, Brenneman (2013) and Reynolds (2013).

1.1 The Reﬂection Spectrum

We have seen in BHS12 that the X-ray emission

arises from a thin accretion disc around the black hole

and a corona which is some distance from the disc. The

disc is geometrically thin in the sense that its thickness

is very much smaller than its radius. For a supermas-

sive massive black hole (SMBH), the temperature is

about 10

K, which means that the thermal radiation

from the hot disc has a peak in the ultraviolet or soft

X-ray region of the spectrum. The disc gets progres-

sively colder at increasing radii, and the net emission

is the sum of the emission from annular regions each

of which emits at a ﬁxed temperature. Some of the

radiation reaches a much hotter corona, which is at a

temperature of about 10

K, located at some distance

from disc. The thermal photons reaching the corona

undergo inverse Compton scattering by the very hot

electrons in it, which leads to increase in energy of

the spectrum. The result of the scattering is that the

photons acquire a power-law spectrum of the form I(ν)

˜ ν

−α

, where ν is the frequency, I(ν) is the intesity of

the radiation, and α is a constant, typically in the range

0 - 1.5. Some of the power-law photons leave the sys-

tem, while the rest are emitted towards the disc. The

photons reaching the disc undergo Compton scattering,

which lowers their energy; after one or more scatter-

ings the photons can emerge from the disk. Some of

the photons incident on the disk are absorbed through

the photoelectric eﬀect by various ions present in the

disc gas, which then produce emission lines through

ﬂuorescence, as described in BHS 12. We are repeat-

ing here, for convenience of understanding, a ﬁgure

from BHS 12:

Figure 1: A sketch showing an accretion disc and

a corona and the various components of the radiation

which are explained in the text. The temperatures are

indicated in kilo electron volts (keV), with 1 KeV being

equivalent to 11.6 million K. The inner disc is indicated

to be at the innermost stable circular orbit.

Shown in Figure 2 is the spectrum produced by

1.2 Modiﬁcations of the Reﬂection Spectrum

the combination of ﬂuorescence and Compton scat-

tering from a power-law spectrum incident on a thin

accretion disc. This spectrum is modiﬁed by various

eﬀects which are discussed in the next section. It is the

modiﬁed spectrum which would be seen by a distant

observe.

Figure 2: Simulation of a reﬂection spectrum pro-

duced when power-law spectrum, shown as a dashed

line, is incident on the inner region of a thin accretion

disc. The relativistic broadening is not taken into ac-

count. It is seen that that the iron (Fe) line is dominant.

Fluorescence lines produced by other elements are also

shown. All the lines are seen to be very narrow. The

broad hump beyond ˜7 keV is produced by absorption

of the reﬂected continuum by iron and the Compton

scattering of high energy photons to lower energies.

The ﬁgure is from A. C. Fabian and G. Miniutti (2005).

1.2 Modiﬁcations of the Reﬂection

Spectrum

The shape of the features in the reﬂection spec-

trum is changed due to relativistic eﬀects as explained

in BHS12. These have to be taken into account to

generate a model of the broadened spectrum, and such

a model has to be ﬁtted to an observed X-ray spec-

trum to derive the parameters which go into the model.

These include properties of the gas in the disc, like

its chemical composition, the inclination of the disc

to the observer’s line of sight and so forth. But most

importantly for the spin determination, the inner radius

of the accretion disc has to be determined as a model

parameter. This inner disc radius is identiﬁed with the

inner most stable circular orbit r

ISCO

of the black hole,

which allows the spin parameter a to be determined, as

described in BHS12.

For the modelling to be correctly carried out, it is

necessary to include absorption and emission processes

which can aﬀect the reﬂection spectrum. Absorption

can take place due to a cold gas which, due to its low

temperature, is largely neutral, i.e. not ionised. The

gas is located at a distance of about 10

- 10

from

the black hole at the centre, where r

is the grav-

itational radius. The gas can be distributed in clumps

so that it does not cover the inner region uniformly.

Radiation reaching these cold clumps can be absorbed

due to the photoelectric eﬀect, modifying the lower

energy part of the spectrum. Another component is a

warm absorber, which is at a temperature of about 10

K – 10

K, and is partially ionised. Such a gas can

again absorb the spectrum through the photoelectric

eﬀect right up to the Fe line energies, depending on

the stage of ionisation. The outer parts of the accre-

tion disk also produces a reﬂection spectrum, which is

not broadened due to relativistic eﬀects, because of the

large distance from the black hole. This spectrum has

a narrow iron emission line at 6.4 keV, which has to

be taken into account while modelling the spectrum.

The outer accretion disc also contributes to the Comp-

ton hump. There is another excess over the power-law

emission at lower energy, known as the soft excess,

which has to be included in the modelling.

Figure 3 shows the X-ray spectrum of the Seyfert

galaxy MGC 3783 observed by the Suzaku satellite.

The prominent iron line emission and the X-ray bump

are seen. The iron line has prominent broad and nar-

row components, with the latter produced by the outer

regions of the accretion disc. Absorption is seen at

X-ray energies of a few keV, and a soft excess below 1

keV.

A model ﬁt to the spectrum, taking into account

various components, relativistic eﬀects etc. led to a

spin parameter a > 0.98.

In BHS12 we found that the broadening of the

6.4 keV line of iron in MCG-6-13-15, with the low

energy tail reaching ˜3keV immediately leads to the

expectation that the black hole has high spin. That

is borne out by detailed modelling which leads to the

1.3 The Distribution of the Spin Parameter

value of the angular momentum parameter a > 0.97.

Another interesting case is the active galaxy Fairall 9,

which is bare Seyfert galaxy, in the sense that it lacks

signiﬁcant absorption, i.e. the intervening absorbing

matter is absent. The soft excess too is signiﬁcantly

weaker in this galaxy. These factors make the ﬁtting

process easier than in the other cases discussed, and

with Suzaku data, a spin parameter value of a = 0.65

(+0.5, -0.5 as the one sigma errors) is obtained.

Figure 3: X-ray spectrum of the Seyfert 1 galaxy

NGC 3783. The features in the spectrum are described

in the text. The ﬁgure is taken from Brenneman (2013).

It should be mentioned here that while all the dis-

cussion above seems to show that the observed spec-

tra are fully consistent with their production by a this

accretion disc (with a corona) around a spinning black

hole, it is possible to model the features in the spectrum

with multiple absorbers that partially cover the source,

without any need of relativistic eﬀects to broaden the

lines (Miller, Turner, Reeves 2008). However, more

recent data make the model less plausible, which we

will describe in a later story.

1.3 The Distribution of the Spin

Parameter

The determination of the spin requires X-ray data

with adequate signal-to-noise ratio, that is with a sig-

nal good enough for the parameters of the model to

be determined with reasonable errors. It is also neces-

sary to use a physical model which takes into account

various processes that can contribute to and aﬀect the

spectrum, as brieﬂy described above. Keeping these

and other related factors in mind, Brenneman (2013)

has reported 22 spin measurements made for 22 AGN

by various groups. The distribution of spin values is

shown in Figure 3. It is seen that that the measured

spins seem to cluster at higher values, with no spin

measurement consistent with a = 0, which corresponds

to the Schwarzschild case. The total number of spin

measurements in the distribution is of course too small

for any ﬁrm conclusions to be drawn about the distribu-

tion. Also, the ﬁgure is based on spin values reported

before 2013. There would be further measurements in

subsequent years of course, and the use of data from

NUSTAR and other satellites would allow the spec-

trum to be ﬁtted over a broader energy range, which

would help to take into account various eﬀects and dis-

tinguish between possible models. These maters will

be discussed in future stories.

Figure 4: The distribution of measured spin pa-

rameter a values for 22 supermassive black holes. The

ﬁgure is taken from Brenneman (2013).

The spin measures reported above are all for su-

permassive black holes. Spin measurements for stellar

mass black holes will be discussed in future stories.

1.3 The Distribution of the Spin Parameter

About the Author

Professor Ajit Kembhavi is an emeritus

Professor at Inter University Centre for Astronomy

and Astrophysics and is also the Principal Investiga-

tor of the Pune Knowledge Cluster. He was the former

director of Inter University Centre for Astronomy and

Astrophysics (IUCAA), Pune, and the International

Astronomical Union vice president. In collaboration

with IUCAA, he pioneered astronomy outreach activ-

ities from the late 80s to promote astronomy research

in Indian universities.

Learning is Better than Programming

by Linn Abraham

airis4D, Vol.2, No.11, 2024

www.airis4d.com

2.1 Introduction

The 2024 Nobel Prize in Physics was awarded to

John J. Hopﬁeld and Geoﬀrey E. Hinton for “for foun-

dational discoveries and inventions that enable machine

learning with artiﬁcial neural networks”. The aim of

this article is to understand a few concepts in the ﬁeld

by turning our attention to the contributions of Geof-

frey Hinton. Hinton is often called the “The Godfather

of A.I.”, whereas the title of “Father of A.I.” is often

credited to either John McCarthy or Alan Turing. What

are his most signiﬁcant contributions to the ﬁeld? In

the 1980s he showed that the backpropogation algo-

rithm discovered by Rumelhart (and independently by

others before him) could be used for learning repre-

sentations from data. In the work that was published

in “Letters to Nature” he applied backpropogation to

enable a machine learn the distributed representation

of words. However the hype that ensued ﬁzzled down

over the next twenty years or so years as researchers felt

dissillusioned. In 2012, he again showed the power of

a deep Convolutional Neural Network (CNN) namely

AlexNet by winning the ILSVRC challenge by a large

margin. Hinton was also the ﬁrst to describe the Re-

stricted Boltzmann Machines (RBM) which forms the

basis of Deep Belief Networks (DBN). In the rest of the

article, let us try to understand how AlexNet changed

computer vision and AI in general.

2.2 ImageNet Challenge

In a special edition of the original AlexNet paper,

the authors give credit to the ImageNet dataset for the

Figure 1: Image showing two of the subtrees and

associated images in the ImageNet dataset.

deep learning revolution. The ImageNet dataset took

about two years to be put together and was the basis

of the ImageNet Large Scale Visual Recognition Chal-

lenge (ILSVRC) which began in 2010. It is important

to note that the ImageNet datastet is built upon the hi-

erarchical structure provided by WordNet. Each mean-

ingful concept in WordNet was called a synonym set or

‘sysnet’. WordNet has 80,000 such sysnets and the goal

of ImageNet was to have approximately 500 to 1000

images per sysnet. These sysnets are part of the many

subtrees such as mammal, bird, ﬁsh, reptile, amphib-

ian, vehicle, furniture, musical instrument, geological

formation, tool, ﬂower, fruit. Figure 1 shows images

from two of those subtrees. The year 2012 probably

marked the the point when the ice began to melt after a

long AI winter. The winners of the ﬁrst two years made

use of intelligent features derived from images that re-

searchers painstakingly came up with such as SIFT,

Fisher vectors etc. This they combined with shallow

Machine Learning algorithms like Support Vector Ma-

chines (SVM), Principal Component Analysis (PCA)

and others. But in 2012 the team SuperVision consist-

ing of Alex Krizhevsky, his Ph.D. supervisor Geoﬀrey

2.3 AlexNet

Figure 2: Diagram showing the architecture of the

AlexNet model.

E. Hinton and Ilya Sutskever put forward the AlexNet

model. It obliterated the competition, beating the run-

ner up in top-5 error rates by more than 10 percentage

points. For each sample in the test set, each entry

model provides a probability of being the right answer.

A top-5 error only considers an error when the actual

label is not amongst the top-5 most probable answers.

2.3 AlexNet

By winning the ILSVRC challenge with a huge

margin, the authors proved without doubt that deeper

networks were the answer. Before thinking about the

implications of their discovery let us ﬁrst try to un-

derstand what their model is all about. The AlexNet

model can be shortened as follows.

(CNN → RN → MP)² → (CNN³ → MP) → (FC

→ DO)² → Linear → softmax

It consists of eight layers, the ﬁrst ﬁve being con-

volutional layers (CNN) and followed by three fully

connected (FC) layers. Figure 2 shows a representative

diagram of this architecture. The CNNs had a ReLU

activation which they found to be faster to train than

tanh neurons. They also implement local response nor-

malization (RN) on the output of the CNN. Some of

the convolutional layers are followed by max-pooling

layers (MP). The ﬁnal layers consists of a 1000 way

softmax layer. All in all, it had 60 million parame-

ters and 650,000 neurons. The network was trained

on the large ImageNet dataset. Since the images from

the dataset had arbitrary rectangular shapes, these were

ﬁrst downsampled to a resolution of 256 × 256. The

authors also incorporated data augmentation by us-

ing image translation and horizontal reﬂections with

Figure 3: On the left, Alex Krizhevsky and on the

right, Geoﬀrey Hinton.

patches of 227 × 227. This increased their training set

by a factor of 2048 although the generated images are

highly correlated. This means that during testing they

would extract a similar patch from all four corners and

a centre patch and pool the decisions of the network on

all these ﬁve patches and their horizontal reﬂections.

2.4 The Legacy of AlexNet

By 1986, partly due to the work of Hinton and

others it was known that the Multi-Layer Perceptron

was capable of learning any decision surface. And

the backpropogation algorithm was what made it pos-

sible to train networks with many layers. However

how would this fare when applied to tasks as complex

as human vision? For certain visual recognition tasks

such as hand written digit recognition using the MNIST

dataset it might be feasible to think of an MLP solu-

tion. However in a much more general classiﬁcation

or detection task such as that put out by the ILSVRC,

it became obvious that what was important was not

just the pixel intensities in the images. It was known

that humans vision involved a hierarchy of features

computed from the intensities like shapes, edges and

textures. Thus researchers came up with the convo-

lutional neural network that would allow the machine

to learn such features without exactly specifying what

features where to be learnt.

The researchers behind AlexNet showed that less

was more, by not hard coding intelligent features into

their model but still allowing for the model to learn

such features from data. They showed that learning

was better than programming in solving such complex

REFERENCES

tasks. They also showed that the depth of the network

was critical to the result that they had achieved. They

showed this by removing a convolutional block from

their network and showing the performance degrade

substantially. Finally to practically train such deep

networks they made use of the GPU by writing imple-

mentations of the 2D convolution operation and other

associated operation which could make use of GPUs.

They also made use of two GPUs by cleverly distribut-

ing the network between those. Ever since GPUs have

come to dominate the world of deep learning. Ded-

icated frameworks like tensorﬂow, pytorch etc. were

developed to make it easier than ever for deep learning

researchers to make use of GPUs without getting their

hands dirty. In summary they showed that what was

lacking was better and more data and more computa-

tion and not the algorithms.

References

[Russakovsky et al.(2015)] Olga Russakovsky, Jia

Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh,

Sean Ma, Zhiheng Huang, Andrej Karpathy,

Aditya Khosla, Michael Bernstein, Alexander C.

Berg, and Li Fei-Fei. ImageNet Large Scale

Visual Recognition Challenge. arXiv:1409.0575

[cs], January 2015.

[ros(2017)] Deep Learning for Computer Vision with

Python: Practitioner Bundle. PyImageSearch,

United States, 1st edition 1.3 edition, 2017. ISBN

978-1-72248-783-6.

[Ford(2018)] Martin R. Ford. Architects of Intelli-

gence: The Truth about AI from the People Build-

ing It. Packt Publishing, Birmingham, UK, 2018.

ISBN 978-1-78913-151-2.

[Marsland(2014)] Stephen Marsland. Machine Learn-

ing: An Algorithmic Perspective. Chapman and

Hall/CRC, 2 edition, October 2014. ISBN 978-0-

429-10250-9. doi: 10.1201/b17476.

[Krizhevsky et al.(2017) Krizhevsky, Sutskever, and Hinton]

Alex Krizhevsky, Ilya Sutskever, and Geoﬀrey E.

Hinton. ImageNet classiﬁcation with deep convo-

lutional neural networks. Communications of the

ACM, 60(6):84–90, May 2017. ISSN 0001-0782,

1557-7317. doi: 10.1145/3065386.

[Deng et al.(2009)Deng, Dong, Socher, Li, Kai Li, and Li Fei -Fei]

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li,

Kai Li, and Li Fei-Fei. ImageNet: A large-scale

hierarchical image database. In 2009 IEEE

Conference on Computer Vision and Pattern

Recognition, pages 248–255, Miami, FL, June

2009. IEEE. ISBN 978-1-4244-3992-8. doi:

10.1109/CVPR.2009.5206848.

About the Author

Linn Abraham is a researcher in Physics,

specializing in A.I. applications to astronomy. He is

currently involved in the development of CNN based

Computer Vision tools for prediction of solar ﬂares

from images of the Sun, morphological classiﬁca-

tions of galaxies from optical images surveys and ra-

dio galaxy source extraction from radio observations.

Applications of AI in Astronomy and

Astrophysics

by Sindhu G

airis4D, Vol.2, No.11, 2024

www.airis4d.com

Figure 1: Image Credit: Dr. David Ragland and

Medium

3.1 Introduction

The integration of Artiﬁcial Intelligence (AI) into

astronomy and astrophysics marks a transformative

leap in our capacity to explore and understand the

universe. As technological advancements drive the

production of vast amounts of data from ground-based

telescopes and space missions, processing and inter-

preting this complex information presents signiﬁcant

challenges. Traditional data analysis methods are of-

ten insuﬃcient for handling the volume and intricacy

of astronomical data, creating a need for innovative so-

lutions. AI, especially through machine learning (ML)

and deep learning (DL) algorithms, has emerged as

a powerful tool for eﬃciently analyzing large datasets,

identifying patterns, and even predicting celestial events.

AI’s applications in astronomy are extensive, from

automating the detection and classiﬁcation of celestial

objects to enhancing observational techniques and op-

timizing telescope operations. For instance, machine

learning models can swiftly classify stars, galaxies, and

transient phenomena such as supernovae with high ac-

curacy, freeing astronomers to concentrate on signiﬁ-

cant discoveries rather than becoming entangled in data

processing. Predictive AI models assist researchers in

forecasting cosmic events, thereby improving planning

and observation strategies.

The ongoing advancements in telescopes, such as

the James Webb Space Telescope and Vera C. Rubin

Observatory, are expected to generate unprecedented

quantities of data, making AI essential for modern

astronomical research. Instruments like the Hubble

Space Telescope and Sloan Digital Sky Survey (SDSS)

have already demonstrated how AI-driven automation

can advance our understanding of the cosmos by im-

proving data analysis, object classiﬁcation, and real-

time detection capabilities.

This article explores AI’s crucial contributions

to data analysis, exoplanet discovery, predictive mod-

eling, and more in astronomy and astrophysics. By

examining these advancements, we aim to showcase

how AI is reshaping our understanding of the universe,

enabling discoveries once thought impossible, and ad-

dressing challenges like data biases and the balance

between AI-driven insights and human creativity in

scientiﬁc inquiry.

3.2 Classiﬁcation and Identiﬁcation of Celestial Objects

3.2 Classiﬁcation and Identiﬁcation

of Celestial Objects

AI has transformed the classiﬁcation of celestial

objects by processing vast datasets that contain bil-

lions of stars, galaxies, and transient phenomena. Un-

like traditional labor-intensive methods, AI models can

rapidly identify and categorize these objects based on

deﬁning characteristics such as shape, brightness, and

spectra. Convolutional neural networks (CNNs), for

example, enable astronomers to classify galaxies into

morphological types—such as spiral, elliptical, and ir-

regular—by training on surveys like the Sloan Digital

Sky Survey. This capability helps scientists uncover

correlations between galaxy structure and evolution.

In stellar classiﬁcation, machine learning algo-

rithms analyze spectral data and properties like tem-

perature, luminosity, and age, allowing for precise cat-

egorization essential to studies of star formation and

evolution. AI also aids in classifying supernovae by

analyzing light curves to distinguish between types,

advancing our understanding of stellar death and en-

abling supernovae to be used as standard candles for

distance measurement.

AI’s impact extends to transient phenomena, where

machine learning quickly classiﬁes events such as gamma-

ray bursts and supernovae, essential for timely follow-

up observations. Furthermore, unsupervised cluster-

ing techniques allow AI models to uncover patterns in

data without prior assumptions, leading to discover-

ies of new relationships and phenomena. An example

is the AstroAI program, where unsupervised learning

was used to catalog over 14,000 X-ray source detec-

tions, highlighting AI’s role in unveiling novel insights

within astronomical datasets.

3.3 Exoplanet Detection

AI is transforming the ﬁeld of exoplanet detection

by streamlining the analysis of massive datasets from

space telescopes like Kepler and TESS. These tele-

scopes observe light curves, tracking changes in stellar

brightness over time to detect the characteristic dips

caused by planets transiting across their host stars. Tra-

ditional methods for analyzing light curves are labor-

intensive and often miss subtle signals. AI, particu-

larly deep learning models like recurrent neural net-

works (RNNs) and long short-term memory (LSTM)

networks, has revolutionized this process by eﬃciently

identifying these signals and ﬁltering out false posi-

tives, signiﬁcantly increasing the speed and accuracy

of exoplanet discovery.

Beyond detection, AI plays a critical role in char-

acterizing exoplanets by analyzing transit and radial

velocity data to infer essential details about a planet’s

size, orbit, and atmospheric composition. These in-

sights are vital for determining habitability and prior-

itizing candidates for further study. For instance, re-

searchers at the University of Georgia have developed

models that analyze synthetic data of protoplanetary

environments, which can then be applied to real ob-

servations, enhancing the likelihood of ﬁnding faint,

previously undetectable exoplanets.

AI models have achieved impressive success rates,

with studies showing up to 96% accuracy in predict-

ing exoplanet presence by detecting subtle patterns in

brightness data. This capability is especially crucial

in light of the vast number of potential planets in our

galaxy alone. Additionally, AI aids in the search for

potentially habitable planets. Researchers have created

algorithms that recognize Earth-like anomalies among

non-habitable exoplanets, allowing scientists to stream-

line their search for planets that could support life.

As missions like the James Webb Space Tele-

scope begin to provide unprecedented data, AI’s role

in exoplanet detection is poised to expand further. By

advancing detection methods, enhancing characteriza-

tion, and enabling the targeted search for habitable

worlds, AI is paving the way for discoveries that may

one day reveal the existence of life beyond Earth.

3.4 Variable Stars and X-ray Binary

Analysis

AI is playing a crucial role in the study of variable

stars and X-ray binaries, enabling researchers to un-

cover new patterns and classiﬁcations. Variable stars,

3.5 Gravitational Wave Detection

known for their ﬂuctuating brightness, are analyzed

through light curve modeling, which reveals insights

into their size, age, and other properties. This is partic-

ularly signiﬁcant for stars like Cepheids and RR Lyrae,

which serve as important distance markers in the uni-

verse.

In the analysis of variable stars, machine learn-

ing algorithms automate classiﬁcation by examining

light curves using techniques such as k-nearest neigh-

bors (kNN), random forests (RF), and neural networks

(NN). For example, research utilizing data from the Op-

tical Gravitational Lensing Experiment (OGLE) suc-

cessfully classiﬁed 31,798 light curves and identiﬁed

178 new variable stars. Unsupervised learning meth-

ods, like Principal Component Analysis (PCA) and

Independent Component Analysis (ICA), help iden-

tify patterns in light curves without prior labeling,

with ICA proving particularly eﬀective for classifying

Cepheid variables. The development of hybrid neural

networks that combine convolutional neural networks

and long short-term memory (LSTM) networks fur-

ther enhances the classiﬁcation process, achieving up

to 85% accuracy in studies using light curves from

OGLE and the Catalina Real-time Transient Survey

(CRTS).

Similarly, AI transforms the analysis of X-ray

binaries, where understanding variability is essential

for studying accretion processes and orbital dynam-

ics. Machine learning techniques analyze X-ray light

curves to detect patterns corresponding to diﬀerent

states of these systems, such as quiescent, outburst, or

transition states. AI can also perform anomaly detec-

tion by establishing baseline behaviors through super-

vised learning, allowing researchers to identify unusual

events that may indicate new physical phenomena or

changes in system dynamics.

In both areas, AI algorithms facilitate automated

light curve analysis, enabling the identiﬁcation of pat-

terns and anomalies indicative of speciﬁc types of vari-

ability. They accurately classify variable stars based on

light curve characteristics and determine the periods of

periodic variable stars, improving our understanding of

their pulsation mechanisms. Additionally, AI aids in

exoplanet detection by analyzing subtle dips in stellar

light curves caused by transiting planets.

In the realm of X-ray binaries, AI enhances source

detection and classiﬁcation in large-scale surveys, an-

alyzes time-series data to study variability patterns,

and conducts spectral analysis to gain insights into the

physical conditions of accreting matter. Moreover, AI

contributes to estimating the masses and spin parame-

ters of black holes by examining their X-ray emissions.

Through these applications, AI is revolutionizing the

study of variable stars and X-ray binaries, opening new

avenues for discovery in astrophysics.

3.5 Gravitational Wave Detection

Artiﬁcial Intelligence is revolutionizing the detec-

tion and analysis of gravitational waves, the ripples in

spacetime produced by catastrophic cosmic events like

black hole or neutron star mergers. Since LIGO (Laser

Interferometer Gravitational-Wave Observatory) made

its ﬁrst detection in 2015, researchers have been explor-

ing innovative methods to enhance detection sensitivity

and accuracy, with AI techniques proving invaluable

for processing vast amounts of data and improving sig-

nal detection amidst noise.

Detecting gravitational waves is challenging due

to their inherently weak signals, often buried in noisy

data. Traditional detection methods can struggle to dif-

ferentiate genuine signals from various environmental

noises. However, AI models, particularly those based

on deep learning, have been developed to eﬀectively

address this issue. For example, CNNs have been uti-

lized to classify signals from LIGO’s data, achieving

high accuracy in identifying gravitational wave events

while ﬁltering out noise. Notably, a team at Argonne

National Laboratory created an AI framework that pro-

cessed an entire month of LIGO data in under seven

minutes, demonstrating that AI can match the sensitiv-

ity of traditional template matching algorithms while

operating more eﬃciently.

AI’s rapid processing capabilities enable real-time

analysis of gravitational wave signals, which is crucial

for multi-messenger astronomy, where gravitational

wave data can be combined with electromagnetic sig-

nals from events like gamma-ray bursts. Integrating

3.6 Fast Radio Burst (FRB) Detection

AI into this workﬂow allows astronomers to respond

swiftly to transient events, increasing the chances of

capturing associated phenomena across diﬀerent wave-

lengths. Additionally, AI is employed for noise reduc-

tion in gravitational wave data analysis, with machine

learning models trained to recognize patterns associ-

ated with noise and eﬀectively ﬁlter these out, enhanc-

ing the clarity of the data for more accurate signal

extraction.

AI also contributes to parameter estimation through

Bayesian inference techniques, allowing for the estima-

tion of the physical parameters of gravitational wave

sources, such as their masses, spins, and distances.

Furthermore, AI facilitates model selection to iden-

tify the most suitable physical models for observed

gravitational wave signals. In the realm of multi-

messenger astronomy, AI correlates gravitational wave

detections with observations from other astronomical

instruments, such as telescopes and neutrino detectors,

enabling joint analysis that provides a comprehensive

understanding of cosmic events. Through these appli-

cations, AI is transforming gravitational wave detec-

tion, paving the way for groundbreaking discoveries in

our understanding of the universe.

3.6 Fast Radio Burst (FRB) Detection

Fast Radio Bursts (FRBs) are intense, millisecond-

long ﬂashes of radio waves originating from distant

galaxies, and their origins remain one of modern as-

tronomy’s greatest mysteries. The detection of these

enigmatic signals poses signiﬁcant challenges due to

their brief duration and the vast amounts of data gener-

ated by radio telescopes. Traditional methods of data

analysis are often time-consuming and ineﬃcient, mak-

ing it diﬃcult to identify these ﬂeeting signals amid

the noise. However, advancements in Artiﬁcial Intelli-

gence are revolutionizing the detection and analysis of

FRBs.

AI models, particularly convolutional neural net-

works, are essential for real-time detection, enabling

astronomers to quickly respond to FRBs as they occur.

Machine learning algorithms can identify distinctive

patterns associated with FRBs, signiﬁcantly reducing

the likelihood of false positives by accurately distin-

guishing genuine signals from radio interference. For

instance, an AI-based detection system developed by

Ph.D. student Wael Farah at the Molonglo Radio Ob-

servatory can identify FRBs within seconds of their

arrival, which was previously unattainable with con-

ventional methods. This system has already detected

multiple bursts, including some of the most energetic

recorded, allowing researchers to capture high-quality

data for detailed analysis.

The application of AI extends beyond simple iden-

tiﬁcation to sophisticated data processing techniques.

For example, FETCH (Fast Extragalactic Transient

Candidate Hunter) is a machine-learning software de-

veloped at West Virginia University that sifts through

massive datasets to distinguish genuine FRB signals

from noise and interference. This signiﬁcantly reduces

the time researchers spend on manual data analysis and

enables the classiﬁcation of FRBs into diﬀerent cate-

gories based on properties such as dispersion measures

and polarization. By analyzing larger datasets with AI

tools, astronomers can discern patterns and potentially

link speciﬁc characteristics of FRBs to their sources.

The integration of AI with advanced observational

techniques further enhances our ability to study FRBs.

The Canadian Hydrogen Intensity Mapping Experi-

ment (CHIME) telescope utilizes digital signal pro-

cessing methods to reconstruct signals from multiple

directions simultaneously, allowing it to detect FRBs

at a signiﬁcantly higher rate than traditional telescopes.

In its ﬁrst year of operation, CHIME discovered over

500 new bursts. Moreover, AI-driven systems assist in

multi-wavelength follow-up observations, where data

from optical and X-ray telescopes are analyzed along-

side radio signals, providing a comprehensive view

of the environments surrounding FRB sources. This

holistic approach enhances our understanding of the

physical conditions that give rise to these mysterious

bursts.

3.7 Cosmology and Dark Matter Studies

3.7 Cosmology and Dark Matter

Studies

Artiﬁcial Intelligence is revolutionizing cosmo-

logical research, particularly in the study of dark mat-

ter and dark energy. By analyzing large-scale struc-

tures and aiding in the interpretation of vast datasets,

AI is transforming how researchers explore the uni-

verse’s evolution. As astronomical surveys and obser-

vational technologies improve, machine learning and

deep learning techniques are increasingly crucial for

addressing the complexities of cosmological data.

AI enhances data analysis by eﬃciently process-

ing the immense volumes generated by large-scale sur-

veys that map galaxy positions and velocities. This

capability allows astronomers to sift through petabytes

of information, identifying subtle signals that help de-

tect phenomena like gravitational lensing and super-

novae, which are vital for understanding dark matter

and the universe’s expansion. Machine learning mod-

els, particularly convolutional neural networks, excel

at analyzing weak lensing data, enabling researchers

to extract detailed cosmological parameters that oﬀer

deeper insights into dark matter distribution and its

inﬂuence on galaxy formation.

The characterization of dark matter remains a sig-

niﬁcant challenge in cosmology. AI techniques are

employed to improve our understanding of its prop-

erties by modeling its gravitational eﬀects on visible

matter. For example, researchers use machine learning

to compare theoretical predictions with observational

data, reﬁning knowledge of how dark matter interacts

with ordinary matter and its role in cosmic structure

formation. Additionally, AI has proven instrumental in

analyzing data from gravitational wave observatories

like LIGO and Virgo, where it helps identify and clas-

sify signals from black hole mergers and neutron star

collisions, providing insights into the mass distribution

of dark matter.

Moreover, AI accelerates cosmological simula-

tions, which are essential for understanding the uni-

verse’s evolution. Traditional simulations are often

computationally intensive and time-consuming, but gen-

erative AI models can speed up this process by creat-

ing realistic simulations based on existing data. This

allows researchers to explore a broader range of sce-

narios concerning dark matter properties and cosmic

evolution without the heavy computational burden of

extensive simulations. By training AI models on these

simulations, scientists can quickly infer cosmological

parameters from observational data, streamlining the

analysis process.

3.8 Image Processing and

Enhancement for Telescope Data

Artiﬁcial Intelligence is signiﬁcantly enhancing

image processing and enhancement techniques for tele-

scope data, addressing challenges such as atmospheric

distortions, noise, and other factors that degrade the

quality of astronomical images. These challenges in-

clude blurring from atmospheric turbulence, electronic

noise obscuring faint details, cosmic rays creating bright

streaks, and the limited dynamic range of captured im-

ages.

To overcome these challenges, AI-driven image

processing techniques are being employed. One pri-

mary application is deblurring, where AI algorithms

utilize deep learning models, including convolutional

neural networks, to reverse the blurring process and

recover the original image. Researchers from North-

western University and Tsinghua University have de-

veloped an AI algorithm that eﬀectively removes blur

from ground-based telescope images, resulting in im-

ages with up to 38.6% fewer errors compared to tra-

ditional methods. This improvement is crucial for ac-

curately measuring the shapes of galaxies and under-

standing gravitational eﬀects in the universe.

In addition to deblurring, AI techniques are also

used for noise reduction, employing ﬁltering and wavelet

denoising methods to smooth out noise while preserv-

ing important features. Deep learning models trained

on large datasets can eﬀectively denoise images, en-

hancing their clarity. Furthermore, algorithms can de-

tect and remove cosmic ray artifacts through detection

and interpolation, and machine learning can be applied

3.9 Astroparticle Physics and Cosmic Ray Detection

for more accurate identiﬁcation of these artifacts.

AI-based super-resolution algorithms play a sig-

niﬁcant role in increasing the eﬀective resolution of

telescope images, allowing astronomers to observe ﬁner

details of cosmic structures and distant galaxies. This

capability has applications in both ground-based obser-

vatories and space telescopes like Hubble. To optimize

performance, researchers train AI models using sim-

ulated data that mimics future telescope conditions,

preparing them for real-world applications.

The development of open-source AI tools has de-

mocratized access to these algorithms, enabling as-

tronomers worldwide to adapt them for speciﬁc needs

and encouraging collaboration within the astronomical

community. As telescopes become more sophisticated,

integrating AI into their operational frameworks is es-

sential. For instance, the James Webb Space Telescope

(JWST) employs various algorithms for image opti-

mization and noise correction, enhancing its ability to

capture detailed images across diﬀerent wavelengths.

3.9 Astroparticle Physics and Cosmic

Ray Detection

AI is making signiﬁcant strides in astroparticle

physics, particularly in the detection and analysis of

cosmic rays and neutrinos. Cosmic rays, which are

high-energy particles primarily consisting of protons,

originate from outer space and interact with the Earth’s

atmosphere. Their study is crucial for understanding

fundamental astrophysical processes. AI techniques

enhance detection capabilities, improve data analysis,

and facilitate real-time monitoring of cosmic ray activ-

ity.

In observatories like IceCube, AI models pro-

cess massive data volumes to distinguish rare neutrino

events from background noise, enabling insights into

high-energy cosmic processes. Similarly, AI aids in

identifying and classifying cosmic ray events by ana-

lyzing their trajectories and interaction patterns, con-

tributing to a deeper understanding of the sources and

propagation mechanisms of high-energy particles in

the universe.

Recent initiatives, such as the Ray Deep Project,

utilize deep learning techniques for real-time cosmic

ray detection by identifying ionizing particles like muons

and electrons through advanced image processing meth-

ods. This project employs convolutional neural net-

works on FPGA (Field-Programmable Gate Array) de-

vices to achieve high-speed analysis, allowing detec-

tion times of less than 50 milliseconds and generating

alerts regarding increases in cosmic ray activity. An-

other noteworthy advancement is the AstroTeq.ai de-

tector, which accurately detects muons from cosmic

rays using scintillators connected to silicon photomul-

tiplier (SiPM) diodes in a TOP-BOTTOM coincidence

mode for enhanced accuracy.

AI’s role extends beyond detection to encompass

sophisticated data analysis techniques. For instance,

deep learning algorithms are applied to remove back-

ground noise from cosmic ray data collected in neutrino

experiments. In liquid argon time projection chambers

(LArTPCs), where cosmic muons often dominate, deep

neural networks classify pixels to ﬁlter out irrelevant

data, improving the quality of scientiﬁc results from

neutrino experiments.

Furthermore, AI technologies enable real-time

monitoring of cosmic ray activity, which is vital for

applications like muon tomography, an imaging tech-

nique that utilizes muons generated by cosmic rays to

create detailed images of structures such as volcanoes

or nuclear facilities. By integrating AI systems capable

of rapid data processing, researchers can generate alerts

for increased cosmic ray activity, allowing timely re-

sponses in various contexts, including safety measures

for environments exposed to radiation.

3.10 Conclusion

The application of AI in astronomy and astro-

physics is transforming the ﬁeld by enabling the analy-

sis of large, complex datasets and uncovering patterns

and phenomena that traditional methods struggle to

detect. AI’s versatility, from object classiﬁcation and

exoplanet detection to gravitational wave identiﬁcation

and cosmological modeling, accelerates scientiﬁc dis-

covery and helps address fundamental questions about

3.10 Conclusion

the nature of the universe. As AI continues to evolve,

its role in astrophysical research will likely expand,

opening new frontiers in our understanding of the cos-

mos.

References:

AI in Astronomy

Astronomers are enlisting AI to prepare for a

data downpour

Welcome to the AI future?

Artiﬁcial intelligence helps in the identiﬁcation

of astronomical objects

AI Revolutionizes Exoplanet Discovery Uncov-

ering New Worlds

New artiﬁcial Intelligence-based tools can help

ﬁnding habitable planets

X-ray Binary

Black Holes and X-ray binaries

Gravitational-wave observatory

Searching for exoplanets using artiﬁcial intelli-

gence

A real-time fast radio burst: polarization detec-

tion and multiwavelength follow-up

The FRATS project: real-time searches for fast

radio bursts and other fast transients with LO-

FAR at 135 MHz

Detection of ultra-fast radio bursts from FRB

20121102A

Astronomers detect most distant fast radio burst

to date

DOE Explains...Dark Matter

Dark matter

Dark Energy and Dark Matter

Tools & Training

Retouching of astronomical data for the produc-

tion of outreach images

Astronomers detect most distant fast radio burst

to date

Image Quality and Calibration

Astroparticle physics

What are fast radio bursts?

About the Author

Sindhu G is a research scholar in Physics

doing research in Astronomy & Astrophysics. Her

research mainly focuses on classiﬁcation of variable

stars using diﬀerent machine learning algorithms. She

is also doing the period prediction of diﬀerent types

of variable stars, especially eclipsing binaries and on

the study of optical counterparts of X-ray binaries.

Part III

Biosciences

Harnessing Artiﬁcial Intelligence for

Biodiversity Conservation: A New Era of

Ecological Understanding

by Geetha Paul

airis4D, Vol.2, No.11, 2024

www.airis4d.com

(Image courtesy)

1.1 Introduction

Biodiversity, the variety of life on Earth, is essen-

tial for maintaining ecological balance and supporting

human existence. However, the rapid loss of biodiver-

sity due to human activities poses signiﬁcant challenges

for conservationists and researchers. Traditional meth-

ods of studying and conserving biodiversity are often

labour-intensive and time-consuming, making it chal-

lenging to keep pace with the scale of the crisis. In

this context, Artiﬁcial Intelligence (AI) and Machine

Learning (ML) have emerged as transformative tools

that can enhance our understanding of biodiversity and

improve conservation eﬀorts. Integrating Artiﬁcial In-

telligence into biodiversity research represents a trans-

formative shift in our approach to conservation. AI

empowers researchers and policymakers by enhanc-

ing data collection eﬃciency, improving accuracy in

species identiﬁcation, enabling real-time monitoring

and predictive analytics, supporting ecological restora-

tion eﬀorts and facilitating interdisciplinary collabora-

tion. As we harness the power of intelligent machines

to address current issues surrounding biodiversity loss,

such as habitat destruction due to climate change or

urbanisation, we move closer to a sustainable future

where human knowledge and technological innovation

work hand in hand to protect our planet’s rich bio-

logical heritage. By embracing these advancements

while remaining vigilant about ethical considerations

surrounding their use, we can pave the way for more

eﬀective conservation strategies that ensure a thriv-

ing planet for future generations. The applications are

vast, from identifying rare species to improving habitat

restoration eﬀorts, demonstrating that AI is not merely

a tool but a crucial ally in our quest to conserve the

natural world. Today, AI is becoming a powerful force

in nature conservation, with applications ranging from

monitoring wildlife to collecting environmental DNA.

1.2 The Challenge of Biodiversity

Loss

The current biodiversity crisis is alarming, with

estimates suggesting that over one million species face

extinction in the coming decades. Habitat destruction,

climate change, pollution, and overexploitation drive

this decline. As ecosystems become increasingly frag-

1.3 The Role of AI in Biodiversity Conservation and Data Collection Eﬃciency

mented and degraded, the need for eﬀective monitoring

and conservation strategies has never been more urgent.

1.3 The Role of AI in Biodiversity

Conservation and Data Collection

Eﬃciency

AI technologies, particularly deep learning algo-

rithms, have shown great promise in addressing these

challenges. By processing vast amounts of data quickly

and accurately, AI can streamline data collection and

analysis, enabling researchers to focus on critical con-

servation tasks. Traditional methods of gathering eco-

logical data often involve extensive ﬁeldwork, which

can be time-consuming and costly. AI-powered sys-

tems can analyse data from various sources—remote

sensors, camera traps, and satellite imagery—24/7.

This capability allows for real-time monitoring of wildlife

activity and environmental changes. For instance, AI

algorithms can identify species from images captured

by camera traps, signiﬁcantly reducing the time re-

quired for manual identiﬁcation. Predictive analytics

is one of AI’s most exciting applications in biodiversity

conservation. By analysing historical data, AI models

can forecast future trends and identify areas at risk of

habitat loss or species decline. For example, AI can

assess how climate change will impact various ecosys-

tems, aiding in adaptation planning. This proactive

approach allows conservationists to develop strategies

to mitigate potential threats before they escalate.

Image courtesy: U.S. Geological Survey.

Fig: 1 Perspective in machine learning for wildlife

conservation. Studies frequently combine data from

multiple sensors at the exact geographic location or

data from various locations to achieve deeper ecolog-

ical insights. Sentinel-2 (ESA) satellite/ of the U.S.

Geological Survey.

1.4 Wildlife Monitoring and

Protection

AI is crucial in monitoring vulnerable wildlife

populations. Advanced technologies such as acoustic

monitoring can detect animal calls and alert researchers

to the presence of rare or endangered species. Ad-

ditionally, drones equipped with AI capabilities can

patrol protected areas to detect illegal activities like

poaching.

(Image courtesy)

1.4.1 Acoustic Monitoring

Acoustic monitoring and sound recordings can

detect animal calls, and researchers can listen to the

recording as many times as needed to conﬁrm an ID.

Repeated listening by multiple experts will allows scru-

tinising data for more accurate interpretation.

1.5 Wildlife Conservation Tools and Technologies

1.4.2 Revolutionising Wildlife Monitoring

with Advanced Technology

(Image courtesy)

Fig:3. Artiﬁcial intelligence-based image recog-

nition and computer vision technologies help identify

species from photos and videos, helping researchers

track and study wildlife populations.

(Image courtesy)

Fig:4. DeepMind, a UK-based company, has de-

veloped an artiﬁcial intelligence model designed to

identify various animal species and assess their popu-

lations. This initiative is based in Serengeti National

Park, Tanzania, where scientists utilise AI technology

to detect wildlife, analyse large volumes of data col-

lected with the help of advanced camera traps and work

towards conserving endangered species before it be-

comes too late.

1.5 Wildlife Conservation Tools and

Technologies

Camera Traps: Equipped with motion sensors,

camera traps capture images and videos of wildlife,

providing valuable data on species presence and be-

haviour.

GPS Tracking Collars: These collars provide real-

time location data, helping researchers track animal

movements and study habits.

Remote sensing - using satellite Images to monitor

and manage habitats.

UAV - Unmanned aerial vehicles or drones for

monitoring habitats Satellite imagery, GIS mapping -

Geographic Information Systems - to map and monitor

species and habitats.

Habitat Mapping: Satellite imagery aids in identi-

fying critical habitats for various species. By mapping

these areas, conservationists can prioritise regions for

protection and restoration.

Population Dynamics: GIS tools enable the analy-

sis of wildlife population trends by correlating satellite

data with ﬁeld observations and mapping and monitor-

ing species and habitats. This helps assess the health

of populations and determine conservation strategies.

DNA Analysis: Genetic analysis helps researchers

understand population genetics, identify individual an-

imals, and monitor genetic diversity.

Predictive Modeling: AI can predict future popu-

lation trends and potential threats, allowing for proac-

tive conservation eﬀorts.

Habitat Analysis: AI algorithms can analyse habi-

tat data to identify critical conservation and restoration

areas.

Disease Monitoring: AI can detect patterns in

wildlife health data, helping to identify and respond to

disease outbreaks.

Climate Impact Studies: AI helps researchers un-

derstand how climate change aﬀects wildlife, guiding

conservation strategies in a changing environment.

1.6 Wildlife Conservation Tools and

Technologies

Camera Traps: Equipped with motion sensors,

camera traps capture images and videos of wildlife,

providing valuable data on species presence and

behaviour.

GPS Tracking Collars: These collars provide

real-time location data, helping researchers track

animal movements and study habits. Remote

sensing - using satellite Images to monitor and

1.7 Case Studies

manage habitats.

UAV - Unmanned aerial vehicles or drones for

monitoring habitats Satellite imagery.

GIS mapping - Geographic Information Systems

- to map and monitor species and habitats.

Habitat Mapping: Satellite imagery aids in iden-

tifying critical habitats for various species. By

mapping these areas, conservationists can pri-

oritise regions for protection and restoration.

Population Dynamics: GIS tools enable the anal-

ysis of wildlife population trends by correlating

satellite data with ﬁeld observations and map-

ping and monitoring species and habitats. This

helps assess the health of populations and deter-

mine conservation strategies.

DNA Analysis: Genetic analysis helps researchers

understand population genetics, identify individ-

ual animals, and monitor genetic diversity. Pre-

dictive Modelling: AI can predict future popu-

lation trends and potential threats, allowing for

proactive conservation eﬀorts.

Habitat Analysis: AI algorithms can analyse

habitat data to identify critical conservation and

restoration areas.

Disease Monitoring: AI can detect patterns in

wildlife health data, helping to identify and re-

spond to disease outbreaks.

Climate Impact Studies: AI helps researchers

understand how climate change aﬀects wildlife,

guiding conservation strategies in a changing en-

vironment.

(Source) Table 1. Shows the list of technologies

to work smarter and faster- Aspects, Description and

Examples of AI Tools

1.7 Case Studies

Camera Traps: Organisations like the Zoological

Society of London utilise AI to analyse millions of

images from camera traps deployed in natural habitats.

This technology enables precise population estimates

and real-time monitoring of biodiversity.

Acoustic Monitoring: Projects like BirdNET use

AI to identify over 3,000 bird species from audio record-

ings. This innovation allows scientists and citizen sci-

entists alike to contribute to biodiversity monitoring on

an unprecedented scale.

Anti-Poaching Eﬀorts: The SMART (Spatial Mon-

itoring and Reporting Tool) platform integrates AI to

analyse data from ranger patrols and camera traps.

This system has signiﬁcantly reduced illegal hunting

by identifying poaching hotspots in several key conser-

vation areas. Real-time data allows for rapid response

to threats like poaching or natural disasters.

AI also aids in ecological restoration eﬀorts by op-

timising strategies for rehabilitating degraded ecosys-

tems. AI can simulate ecosystem behaviour under

various scenarios, allowing scientists to test diﬀerent

restoration approaches. Drones equipped with AI can

identify and remove invasive species, helping restore

native ﬂora and fauna balance. AI sensors assess soil

1.8 Behavioural Studies Through Machine Learning

health to optimise the selection of native plant species

for restoration projects.

1.8 Behavioural Studies Through

Machine Learning

In addition to identiﬁcation and monitoring tasks,

machine learning algorithms can facilitate behavioural

studies among various species.

Grooming Behaviors: AI systems can analyse

video footage to observe grooming behaviours in ﬁsh

or other aquatic organisms, providing insights into their

social interactions.

Courtship Displays: By studying courtship be-

haviours through video analysis, researchers can better

understand mating strategies in diﬀerent species.

These behavioural insights are vital for develop-

ing eﬀective conservation strategies tailored to speciﬁc

ecological needs.

1.9 Ethical Considerations

While AI oﬀers immense potential for biodiver-

sity conservation, it also raises ethical questions that

must be addressed. Issues related to data privacy, po-

tential job displacement for human workers, and the

implications of using AI in wildlife management need

careful consideration. Conservationists must navigate

these challenges while leveraging technology to protect

vulnerable ecosystems. As researchers increasingly

rely on data collected from public sources or citizen

scientists, ensuring data privacy becomes paramount.

Ethical guidelines must be established to protect indi-

viduals’ rights while promoting transparency in bio-

diversity research. The automation by AI may lead

to concerns regarding job displacement among tradi-

tional ﬁeldworkers or researchers. While AI enhances

eﬃciency, it is essential to consider how these tech-

nologies can complement human expertise rather than

replace it.

1.10 Conclusion

The integration of Artiﬁcial Intelligence into bio-

diversity research represents a paradigm shift in our ap-

proach to conservation. By automating data collection

and analysis, predictive modelling, and wildlife mon-

itoring, AI empowers researchers to make informed

decisions that enhance our understanding of complex

ecological systems. As we harness the power of intel-

ligent machines to address pressing issues surrounding

biodiversity loss, we move closer to a sustainable future

where human knowledge and technological innovation

work hand in hand to protect our planet’s rich biological

heritage. The applications are vast—from identifying

rare species to improving habitat restoration eﬀorts—

demonstrating that AI is not just a tool but a crucial

ally in our quest to conserve the natural world. By

embracing these technological advancements while re-

maining vigilant about ethical considerations, we can

pave the way for more eﬀective conservation strategies

that ensure a thriving planet for future generations.

1.11 Future Directions

Looking ahead, further research is needed to re-

ﬁne AI algorithms tailored explicitly for biodiversity

applications. Collaborations between ecologists and

computer scientists will be critical in developing ro-

bust models for ecological complexities.

Moreover, expanding access to training datasets

will enhance machine learning capabilities across var-

ious taxa globally. Engaging citizen scientists through

mobile applications could facilitate data collection while

fostering public interest in biodiversity conservation.

In summary, as we stand at the intersection of tech-

nology and ecology, embracing artiﬁcial intelligence

oﬀers an unprecedented opportunity for understanding

and actively preserving our planet’s invaluable biodi-

versity.

1.11 Future Directions

(Image courtesy)

Fig 5: This image showcases a robotic arm gently

holding a lush green sapling toward a human hand ex-

tended in acceptance. The white robotic arm, with

sleek design and dexterous ﬁngers, symbolises the

pinnacle of modern engineering and the potential for

artiﬁcial intelligence to coexist with and nurture the

natural world. The human hand’s soft, warm skin

tone contrasts with the cool precision of the robotic

limb, conveying a narrative of trust and partnership

between humanity and machines. The verdant leaves

of the sapling, backlit by soft, diﬀused daylight ﬁlter-

ing through a window, introduce a sense of hope and

growth that symbolise a harmonious future.

“By prioritising biodiversity conservation and pro-

tecting natural habitats, with the help of AI tools,

we can cultivate a harmonious coexistence between

technology and nature, ultimately fostering a healthier

planet for future generations”.

References

4 Reasons why acoustic monitoring reigns in

wildlife research

https://saiwa.ai/blog/ai-in-wildlife-conservation/

https://www.nature.com/articles/s41467-022-27980-

AI in conservation where we came from and

where we are heading

https://hcrobo.com/ai-wildlife-monitoring-automated-

wildlife-tracking/

Silvestro, D., Goria, S., Sterner, T., & et al.

(2022). Improving biodiversity protection through

artiﬁcial intelligence. Nature Sustainability, 5,

415–424.

https://doi.org/10.1038/s41893-022-00851-6

About the Author

Geetha Paul is one of the directors of

airis4D. She leads the Biosciences Division. Her

research interests extends from Cell & Molecular Bi-

ology to Environmental Sciences, Odonatology, and

Aquatic Biology.

Bronchopulmonary Dysplasia in Preterm

Infants: Addressing the Challenges and

Advancements in Management

by Kalyani Bagri

airis4D, Vol.2, No.11, 2024

www.airis4d.com

2.1 Introduction

Bronchopulmonary Dysplasia (BPD) is a chronic

lung condition predominantly aﬀecting preterm infants,

particularly those born before 28 weeks of gestation.

Characterized by a need for supplemental oxygen or

respiratory support for at least 28 days post-birth. BPD’s

severity is classiﬁed based on oxygen dependency at

36 weeks postmenstrual age (PMA). This classiﬁca-

tion helps diﬀerentiate the stages of the disease, which

typically develops due to disrupted lung maturation,

exacerbated by necessary medical interventions such

as mechanical ventilation and oxygen therapy.

Since Dr. William Northway’s initial description

in 1967, our understanding of BPD has evolved signif-

icantly. Originally attributed to lung injury from pro-

longed mechanical ventilation and high oxygen expo-

sure. Classic BPD was thought to cause inﬂammation,

ﬁbrosis, and alveolar damage. However, advancements

in neonatal care, including gentler ventilation methods,

surfactant therapy, and antenatal corticosteroids, have

transformed the clinical presentation of BPD. Now, it is

recognized as a developmental disorder resulting from

interrupted lung growth rather than solely lung injury.

2.2 Mortality and Morbidity

BPD remains a leading complication of prematu-

rity, substantially contributing to infant mortality and

morbidity. Severe BPD increases the risk of early mor-

tality due to respiratory failure, pulmonary hyperten-

sion, and other life-threatening conditions. Despite

neonatal care advancements, severe BPD cases con-

tinue to carry high mortality rates, particularly in ex-

tremely preterm infants.

Long-term morbidity associated with BPD in-

cludes chronic respiratory problems, such as recurrent

infections, wheezing, and impaired gas exchange. Ad-

ditionally, BPD is linked to neurodevelopmental chal-

lenges, including motor delays, cognitive impairments,

and behavioral issues. These ongoing health problems

result in prolonged hospital stays, increased healthcare

costs, and frequent readmissions, signiﬁcantly impact-

ing quality of life and developmental outcomes for af-

fected children.

2.3 Short-Term and Long-Term

Outcomes

Preterm infants with BPD often require extended

stays in the neonatal intensive care unit (NICU) due

to prolonged respiratory support, which increases their

vulnerability to infections and feeding diﬃculties. These

complications can hinder growth and neurodevelop-

ment, while prolonged mechanical ventilation raises

the risk of ventilator-associated complications, further

exacerbating health challenges.

2.4 Pathogenesis of BPD

The consequences of BPD persist beyond infancy,

impacting long-term outcomes. Children with a his-

tory of BPD are more likely to experience respiratory

issues, including reduced lung function and an ele-

vated risk of chronic obstructive pulmonary disease

(COPD) in adulthood. Structural changes in the lungs

can persist into adolescence and adulthood, leading to

impaired exercise tolerance and increased susceptibil-

ity to respiratory illnesses. Moreover, BPD contributes

to neurodevelopmental delays, aﬀecting learning, be-

havior, and overall quality of life.

2.4 Pathogenesis of BPD

The pathogenesis of BPD in preterm infants in-

volves a complex interplay of prenatal and postnatal

mechanisms that disrupt lung development. Maternal

inﬂammation, such as chorioamnionitis, can precipi-

tate lung injury before birth, while intrauterine growth

restriction (IUGR), placental dysfunction, and mater-

nal smoking can also compromise lung development,

increasing vulnerability to postnatal injury.

Following birth, mechanical ventilation, oxygen

toxicity, infections, and nutritional deﬁciencies fur-

ther contribute to lung damage. Although essential for

survival, high oxygen levels generate oxidative stress,

damaging delicate lung tissue. Even with modern,

gentler ventilation techniques, mechanical ventilation

can still traumatic immature lungs. Infections and in-

ﬂammation exacerbate lung development impairment,

leading to abnormal alveolarization and vasculariza-

tion.

The cumulative eﬀect of these prenatal and post-

natal insults is arrested alveolar growth, abnormal lung

architecture, and impaired gas exchange – the hallmark

characteristics of BPD.

2.5 Prevention and Treatment

Challenges

Despite signiﬁcant advancements in neonatal care,

which have improved the survival rates of preterm in-

fants, preventing and treating BPD remains a formidable

challenge. Current treatments, including antenatal cor-

ticosteroids, surfactant replacement therapy, and less

invasive ventilation strategies, have mitigated the sever-

ity of respiratory distress but have not substantially

reduced BPD incidence. The ongoing struggle to pre-

vent BPD stems from the delicate balance between

life-saving interventions, such as oxygen therapy and

mechanical ventilation, and their potential harm to the

developing lungs of extremely preterm infants.

Furthermore, the intricate pathogenesis of BPD,

involving multiple prenatal and postnatal factors, com-

plicates eﬀorts to prevent its onset fully. As a result,

preventing BPD remains an area of ongoing research,

with clinicians seeking innovative strategies to mini-

mize lung damage and promote healthy lung develop-

ment in vulnerable preterm infants.

2.6 Postnatal Corticosteroids

Postnatal corticosteroids (PNCS) have been a main-

stay in managing preterm infants at risk of BPD, with

established eﬃcacy in reducing incidence and sever-

ity. However, concerns about potential long-term ad-

verse eﬀects have sparked ongoing debate and evolv-

ing guidelines. The mechanism of action of PNCS in

preventing BPD is complex, involving reduced lung in-

ﬂammation, promoted lung maturation, and improved

oxygenation.

Although corticosteroid therapy became a stan-

dard intervention for preterm infants at risk of BPD

in the 1990s, increasing long-term follow-up studies

have raised concerns about optimal dosage and dura-

tion. PNCS therapy in preterm infants carries both

short-term and long-term risks.

In the short term, corticosteroids can increase the

risk of infection by suppressing the immune system,

leading to sepsis and other infections. Additionally,

PNCS can cause metabolic complications, such as hy-

perglycemia and hypertension, complicating manage-

ment and increasing adverse event risk. Corticosteroid

use may also result in gastrointestinal complications,

including feeding intolerance and an increased risk of

necrotizing enterocolitis (NEC). Furthermore, adrenal

suppression is a concern, requiring careful tapering and

2.7 Predictive Modeling

monitoring for signs of adrenal insuﬃciency.

Long-term risks associated with PNCS therapy in-

clude neurodevelopmental impairment, with high-dose

dexamethasone exposure potentially increasing the risk

of cerebral palsy and cognitive deﬁcits. Growth retar-

dation is another concern, as corticosteroid therapy can

negatively impact growth in preterm infants, leading to

long-term growth deﬁcits. Moreover, corticosteroid

exposure may impair lung development, aﬀecting lung

function well into childhood. Endocrine dysfunction,

particularly related to adrenal insuﬃciency, is also a

potential long-term consequence.

Balancing the beneﬁts and risks of PNCS therapy

requires careful consideration of individual patient fac-

tors, including gestational age, birth weight, and clini-

cal status, to optimize treatment and minimize adverse

eﬀects.

2.7 Predictive Modeling

To tackle the complexities surrounding corticos-

teroid use, innovative tools have been developed to help

clinicians tailor treatments to individual infants. No-

tably, the National Institute of Child Health and Human

Development(NICHD) Bronchopulmonary Dysplasia

Estimator (BPDE) is a predictive model that utilizes

patient-speciﬁc factors to assess BPD risk, enabling

informed decisions regarding corticosteroid use. By

identifying high-risk infants, clinicians can prioritize

early and aggressive interventions, such as postnatal

corticosteroids (PNCS) and surfactant therapy, reduc-

ing unnecessary corticosteroid exposure and minimiz-

ing adverse eﬀects.

The estimator identiﬁes infants at highest risk of

developing BPD, allowing clinicians to prioritize early

intervention. Additionally, it optimizes steroid dos-

ing by tailoring therapy to an infant’s individual risk

proﬁle, minimizing adverse eﬀects while maximizing

therapeutic beneﬁts. Furthermore, the BPDE enables

continuous monitoring of treatment response, allowing

for dynamic adjustments to the therapeutic regimen, re-

sulting in a more personalized and eﬀective treatment

plan.

2.7.1 The Need for a Separate BPD

Estimator for India

The application of international guidelines and

BPD estimators, such as the NICHD BPD estimator,

in the Indian context is limited due to several factors.

India’s diverse population, characterized by varying ge-

netic and environmental factors, inﬂuences BPD risk,

making it challenging to directly apply international

models. Additionally, diﬀerences in clinical practices,

including oxygen therapy and surfactant administra-

tion, across various Indian healthcare settings further

complicate the use of standardized estimators.

Moreover, the lack of extensive, high-quality re-

search studies on preterm infants in India makes it

diﬃcult to create and conﬁrm the accuracy of reliable

estimators/predictive models for BPD. Consequently,

developing a tailored BPD estimator for the Indian con-

text is essential to enhance risk prediction accuracy and

optimize clinical decision-making.

Such a model should incorporate population-speciﬁc

factors, including maternal nutritional status, socioe-

conomic factors, and exposure to environmental pollu-

tants, to better reﬂect the Indian scenario. By account-

ing for these unique factors, an India-speciﬁc BPD

estimator can provide more reliable risk assessments,

ultimately improving outcomes for preterm infants.

2.7.2 Web-Based BPD Estimator at

Fernandez Hospital, Hyderabad

We developed and validated a predictive model

for BPD using a retrospective cohort of 845 preterm

infants (206 with BPD and 639 without BPD) admitted

to the Fernandez Foundation’s Neonatal Intensive Care

Unit (NICU) from 2018 to 2022. Our study integrated

various perinatal and respiratory risk factors, employ-

ing an ensemble model of Random Forest algorithms

to optimize predictive accuracy.

The model demonstrated robust performance, achiev-

ing 99.0% accuracy on training set and 80.3% on the

testing set, comparable to the established NICHD BPD

estimator. Notably, our model predicts the risk of BPD

and provides percentage probabilities for both classes

(BPD and No-BPD), enhancing clinical decision-making.

2.8 Emerging Treatments and Future Directions

Our ﬁndings highlight key contributors to BPD

pathogenesis, particularly early-onset sepsis, Necrotiz-

ing Enterocolitis (NEC), and Patent Ductus Arteriosus

(PDA). The predictive model is particularly advanta-

geous in optimizing PNCs use by reducing unnecessary

exposure, associated risks and improving overall eﬃ-

cacy. To further enhance the model’s utility, we are

expanding it to include multiclass classiﬁcation, cate-

gorizing BPD into mild, moderate, and severe classes.

This advancement aims to facilitate targeted interven-

tions and personalized care for high-risk preterm in-

fants.

To facilitate clinical implementation, our predic-

tive model has been developed as a user-friendly web-

based estimator, enabling healthcare providers to ac-

cess and utilize the tool for real-time assessments and

tailored care strategies.

2.8 Emerging Treatments and Future

Directions

Emerging treatments and novel approaches oﬀer

hope for reducing the burden of BPD. According to the

National Institutes of Health, several promising strate-

gies are being explored.

One area of research focuses on stem cell therapy,

speciﬁcally mesenchymal stem cells (MSCs), which

have shown potential in repairing lung tissue and re-

ducing inﬂammation. Preclinical studies have demon-

strated that MSCs can promote lung growth and reduce

ﬁbrosis in animal models of BPD, and early-phase clin-

ical trials are investigating their safety and eﬃcacy in

neonates.

Additionally, new pharmacological agents target-

ing inﬂammation, oxidative stress, and lung develop-

ment are being developed. Anti-inﬂammatory agents,

antioxidants, and growth factors that promote alveolar-

ization and vascularization hold promise in preventing

or treating BPD.

Advances in genetic research also pave the way

for precision medicine approaches, enabling targeted

interventions based on an infant’s genetic proﬁle. This

personalized approach may improve outcomes and re-

duce BPD risk.

Moreover, revolutionary changes are on the hori-

zon in neonatology, particularly with artiﬁcial womb

technology. By mimicking the intrauterine environ-

ment, this innovation supports extremely preterm in-

fants during critical lung development stages, poten-

tially drastically reducing the need for mechanical ven-

tilation and oxygen therapy – preventing BPD before it

starts.

References

1. Bonadies, L., Zaramella, P., Porzionato, A., Per-

ilongo, G., Muraca, M., & Baraldi, E. (2020).

Present and Future of Bronchopulmonary Dys-

plasia. Journal of Clinical Medicine, 9(5), 1539.

https://doi.org/10.3390/jcm9051539. PMID: 32443685;

PMCID: PMC7290764.

2. Holzfurtner, L., Shahzad, T., Dong, Y., Rekers,

L., Selting, A., Staude, B., Lauer, T., Schmidt,

A., Rivetti, S., Zimmer, K. P., Behnke, J., Bel-

lusci, S., & Ehrhardt, H. (2022). When inﬂam-

mation meets lung development—An update on

the pathogenesis of bronchopulmonary dyspla-

sia. Molecular and Cellular Pediatrics, 9(1), 7.

https://doi.org/10.1186/s40348-022-00137-z. PMID:

35445327; PMCID: PMC9021337.

3. Zayat, N., Truﬀert, P., Drumez, E., Duhamel, A.,

Labreuche, J., Zemlin, M., Milligan, D., Maier,

R. F., Jarreau, P. H., Torchin, H., Zeitlin, J.,

Nuytten, A., & EPICE Research Group. (2022).

Systemic Steroids in Preventing Bronchopulmonary

Dysplasia (BPD): Neurodevelopmental Outcome

According to the Risk of BPD in the EPICE Co-

hort. International Journal of Environmental

Research and Public Health, 19(9), 5600. https:

//doi.org/10.3390/ijerph19095600. PMID: 35564997;

PMCID: PMC9106050.

4. Tracy, M. C., & Cornﬁeld, D. N. (2020). Bron-

chopulmonary Dysplasia: Then, Now, and Next.

Pediatric Allergy, Immunology, and Pulmonology,

33(3), 99–109. https://doi.org/10.1089/ped.2020.

1205. PMID: 35922031; PMCID: PMC9354034.

5. Nuthakki, S., Ahmad, K., Johnson, G., & Cuevas

2.8 Emerging Treatments and Future Directions

Guaman, M. (2023). Bronchopulmonary Dys-

plasia: Ongoing Challenges from Deﬁnitions to

Clinical Care. Journal of Clinical Medicine,

12(11), 3864. https://doi.org/10.3390/jcm12113864.

PMID: 37298058; PMCID: PMC10253815.

6. Muehlbacher, T., Bassler, D., & Bryant, M. B.

(2021). Evidence for the Management of Bron-

chopulmonary Dysplasia in Very Preterm In-

fants. Children (Basel), 8(4), 298. https://doi.

org/10.3390/children8040298. PMID: 33924638;

PMCID: PMC8069828.

7. Lemyre, B., Dunn, M., & Thebaud, B. (2020).

Postnatal corticosteroids to prevent or treat bron-

chopulmonary dysplasia in preterm infants. Pae-

diatrics & Child Health, 25(5), 322–331. https://

doi.org/10.1093/pch/pxaa073. PMID: 32765169;

PMCID: PMC7395322.

8. Balany, J., & Bhandari, V. (2015). Understand-

ing the Impact of Infection, Inﬂammation, and

Their Persistence in the Pathogenesis of Bron-

chopulmonary Dysplasia. Frontiers in Medicine,

2, 90. https://doi.org/10.3389/fmed.2015.00090.

PMID: 26734611; PMCID: PMC4685088.

9. Aleem, S., & Greenberg, R. G. (2023). Accu-

rate Prediction of Bronchopulmonary Dysplasia:

Are We There Yet? Journal of Pediatrics, 258,

113389. https://doi.org/10.1016/j.jpeds.2023.03.

004. Epub 2023 Mar 16. PMID: 36933768.

About the Author

Dr. Kalyani Bagri is a Senior Research

Associate at Fernandez Foundation in Hyderabad, In-

dia, where she plays a pivotal role as the lead data sci-

entist in the Neonatology Department. She earned her

Ph.D. in Astrophysics from Pt. Ravishankar Shukla

University, in collaboration with IUCAA and TIFR

Mumbai. In her current role, Dr. Bagri independently

integrates Artiﬁcial Intelligence (AI) into neonatal

care. She is leading critical projects, including the

development of a Bronchopulmonary Dysplasia esti-

mator to guide strategic decisions and optimize steroid

usage in neonates, and a sepsis calculator designed to

detect sepsis in neonates prior to clinical recognition.

Additionally, she is involved in a range of initiatives

aimed at advancing neonatal health outcomes through

her innovative and data-driven approaches.

Ethical Implications of Using AI in Scientiﬁc

Research

by Jinsu Ann Mathew

airis4D, Vol.2, No.11, 2024

www.airis4d.com

In just a few short years, Artiﬁcial Intelligence

(AI) has transformed scientiﬁc research in ways that

were once the stuﬀ of science ﬁction. Imagine a sys-

tem that can analyze mountains of data in minutes,

help design complex molecules for new drugs, review

thousands of research papers at lightning speed, or even

generate fresh ideas and hypotheses for scientists to ex-

plore. AI is doing all of this and more, fundamentally

changing how research is done.

In physics, for example, AI is being used to simu-

late and predict the behavior of particles at the smallest

scales, helping scientists understand the fundamental

forces of the universe. In ﬁelds like medicine, it’s help-

ing predict which new treatments might work best for

speciﬁc diseases. AI systems are even working along-

side scientists to test theories, analyze experimental

data, and reveal hidden connections across ﬁelds.

But with these amazing capabilities come impor-

tant ethical questions. How do we ensure the ﬁndings

from AI are accurate and trustworthy? Can we fully

understand and explain the conclusions AI reaches, or

are we taking a “black box” at its word? And what

happens if the data AI learns from is biased? If not

carefully managed, AI could amplify those biases, af-

fecting the results in ways that could mislead science or

harm society. Addressing these ethical issues is crucial

to ensure that AI contributes responsibly and equitably

to scientiﬁc progress. This article delves into these

ethical challenges and their impact on the reliability

and integrity of AI-driven research.

3.1 Ensuring Accuracy and

Trustworthiness in AI Findings

One of the most signiﬁcant ethical challenges in

using AI for research is ensuring that its ﬁndings are

accurate and trustworthy. While AI algorithms have

incredible potential to process data and detect patterns,

they’re not infallible. They operate in highly special-

ized domains, often working with data that is incom-

plete, noisy, or inconsistent. If AI models produce

inaccurate predictions or ﬂawed conclusions, the con-

sequences can be costly and, in some ﬁelds, even dan-

gerous. This is particularly concerning in ﬁelds where

research ﬁndings inﬂuence policy, public health, or

treatment plans.

Take the example of AI in drug discovery, where

researchers use AI models to predict the potential ef-

fectiveness of new compounds. AI can sift through vast

chemical databases, analyzing millions of compounds

and identifying those that could potentially treat spe-

ciﬁc diseases. This process is incredibly valuable for

accelerating the drug discovery pipeline. However, if

these AI predictions are inaccurate, scientists might

invest signiﬁcant time and resources into testing com-

pounds that ultimately prove ineﬀective or even unsafe.

Another example can be found in AI applications

in environmental science, where AI is used to forecast

climate patterns, track pollution levels, and assess the

impacts of deforestation or other ecological changes.

Inaccurate predictions in these areas could misguide

3.2 The Transparency Challenge: Opening the “Black Box”

critical environmental policies. For instance, if an AI

system underestimates the pollution levels in a spe-

ciﬁc area due to incomplete or outdated data, regula-

tory agencies may fail to impose necessary pollution

controls, putting ecosystems and public health at risk.

In such cases, trust in AI predictions is paramount, as

these outcomes have long-lasting eﬀects on society and

the planet.

To ensure AI’s trustworthiness, models must un-

dergo rigorous testing, validation, and continuous mon-

itoring. In drug discovery, this means running AI pre-

dictions against known compounds with veriﬁed ef-

fects to conﬁrm that the model can accurately identify

promising candidates. For environmental and health

applications, AI models need to be trained on large,

diverse datasets that are regularly updated, helping to

avoid inaccuracies caused by outdated or unrepresen-

tative data.

3.2 The Transparency Challenge:

Opening the “Black Box”

AI models, particularly complex ones like deep

neural networks, often function as “black boxes.” While

they can provide impressive answers and solutions, the

inner workings behind these answers are notoriously

diﬃcult to interpret, even for experts. This lack of

transparency creates a signiﬁcant dilemma: if scien-

tists and practitioners cannot fully understand or ex-

plain how an AI model reached its conclusions, how

can they trust these ﬁndings or conﬁdently justify them

to others? Without transparency, AI-driven insights

could remain out of reach for validation, veriﬁcation,

and understanding—a risk especially high in sensitive

ﬁelds like healthcare and environmental science.

In healthcare, for example, AI has been employed

to predict patient outcomes based on electronic health

records, analyzing vast datasets to forecast everything

from potential disease progression to the likelihood of

hospital readmission. These models often analyze nu-

merous factors at once, such as age, genetics, medical

history, and lifestyle choices, yielding predictions that

can guide treatments. However, when these predictions

arrive without an explanation for why a particular out-

come was chosen, doctors may hesitate to follow the

recommendations, unsure if the AI has accounted for

essential factors or ignored key information due to bias

in the training data. Imagine an AI recommending

a speciﬁc cancer treatment based on data from pre-

dominantly younger patients. Without transparency,

a physician might question whether this recommen-

dation truly applies to an older patient with multiple

health conditions, leading to justiﬁable hesitation in

trusting the AI’s judgment.

To address these transparency issues, researchers

are focusing on explainable AI (XAI)—models that

aim to reveal how AI arrives at speciﬁc conclusions.

In healthcare, explainable AI can help a doctor see

which factors (like age, genetic proﬁle, or health his-

tory) were weighted most heavily in a patient’s risk

prediction, making it easier for them to determine if

the recommendation is suitable. Transparency is also

essential for ensuring that AI ﬁndings are robust and

trustworthy. When results are open to scrutiny and can

be independently veriﬁed, they carry greater weight

and are more likely to be accepted within the scien-

tiﬁc community. Developing transparent, explainable

AI models enables researchers to back up AI-driven

discoveries with clear, understandable reasoning that

aligns with traditional scientiﬁc standards. Only by

improving transparency in AI systems can we ensure

that their insights are reliable, ethical, and beneﬁcial to

society at large.

3.3 The Risk of Bias in AI Models

AI models are only as good as the data they learn

from. When this data contains biases or lacks represen-

tation, the AI’s conclusions can become skewed, po-

tentially reinforcing existing inequalities or producing

inaccurate results. This is a serious ethical concern be-

cause biased AI models can inﬂuence further research,

decision-making, and public opinion.

In academic research, AI bias has already shown

how it can distort outcomes. For example, AI systems

used to evaluate research quality or assign academic

rankings may be trained on publication and citation

3.4 Moving Forward: A Framework for Responsible AI in Research

records, metrics that can reﬂect existing academic in-

equalities. These datasets often favor well-resourced

institutions and established researchers, as they have

greater access to publication channels and funding. An

AI model trained primarily on these records may un-

dervalue contributions from emerging researchers or

smaller institutions, perpetuating a cycle where only a

few voices receive visibility. This bias can aﬀect hir-

ing and funding decisions, limiting opportunities for

innovation from diverse perspectives.

AI can also introduce bias in hiring for corpo-

rate and research positions. Many companies and

institutions now use AI-driven recruitment tools to

screen applicants, learning from past hiring decisions.

If these past decisions favored certain demographics

or preferred educational backgrounds, the AI might

continue that pattern, unintentionally discriminating

against candidates from less represented groups or non-

traditional educational paths. For instance, if a com-

pany’s historical hiring data favored applicants from

certain universities, the AI might overlook qualiﬁed

applicants from other institutions, reducing diversity

and missing out on valuable perspectives.

Bias can also emerge in law enforcement AI, es-

pecially in systems designed to predict crime hotspots

or recommend sentencing. If the training data reﬂects

biased policing practices or historical disparities in sen-

tencing, the AI may continue to target speciﬁc neigh-

borhoods or individuals unfairly. This can inﬂuence

judicial outcomes, aﬀecting people’s lives and perpet-

uating systemic bias within the criminal justice system.

In all these cases, the common thread is that AI

can unintentionally reinforce the very biases that exist

in its training data. To address this, researchers and

developers need to scrutinize the data sources used in

AI training. A more equitable dataset, for example,

could include research from diverse academic ﬁelds,

grading samples from a wide range of dialects and

linguistic backgrounds, or hiring data that anonymizes

certain demographic information. Techniques like bias

detection, reweighting, and adversarial debiasing are

critical for minimizing the eﬀects of biased data.

Bias remains one of the toughest ethical issues to

handle, as it can be deeply ingrained and diﬃcult to

detect. By actively managing bias, we can help ensure

AI contributes to fairer and more inclusive outcomes

across research, education, hiring, and beyond. Ad-

dressing these concerns is essential if we want AI to

serve as a truly reliable and unbiased partner in ad-

vancing science and society.

3.4 Moving Forward: A Framework

for Responsible AI in Research

Ensuring that AI contributes responsibly and eq-

uitably to scientiﬁc research requires a comprehensive

ethical framework. Such a framework should address

the various challenges posed by AI systems to foster

transparency, accuracy, and fairness. Key steps in this

framework include:

Rigorous Testing and Validation

AI models should undergo continuous testing with

varied datasets to ensure they remain accurate and re-

liable across diﬀerent research areas, reducing the risk

of overﬁtting or unintended biases.

Enhancing Transparency and Explainability

Increasing AI transparency allows scientists and

the public to understand AI-driven conclusions, fos-

tering trust through interpretability tools and simpler

models that reveal the logic behind AI decisions.

Proactively Managing and Mitigating Bias

By using diverse training data and techniques like

reweighting and debiasing, researchers can actively re-

duce AI model bias, helping ensure fair and balanced

research outcomes.

Establishing Ethical Guidelines and

Standards

Research organizations should deﬁne ethical guide-

lines covering data privacy, transparency, and account-

ability, providing a structured approach to address eth-

ical concerns and protect societal values.

REFERENCES

Incorporating Human Oversight and

Accountability

Human oversight ensures that AI serves as a sup-

port tool rather than a replacement for expert judgment,

with review boards validating AI ﬁndings to guard

against errors and unintentional biases.

Prioritizing Interdisciplinary Collaboration

Collaboration among experts from various ﬁelds,

such as ethics, computer science, and the research do-

main, helps identify ethical challenges and create bal-

anced solutions for using AI in research.

Continuous Monitoring and Updating of AI

Models

Regularly updating AI models helps adapt them to

new data and evolving ethical standards, ensuring their

ongoing fairness, accuracy, and relevance in changing

research contexts.

Ensuring Data Privacy and Security

An ethical AI framework must prioritize strict data

handling and anonymization protocols to protect indi-

vidual privacy and prevent misuse, especially when

dealing with sensitive research data.

By implementing these principles, the scientiﬁc

community can establish a responsible AI framework

that maximizes AI’s potential while safeguarding re-

search integrity, trust, and inclusivity. These measures

are fundamental to harnessing AI’s capabilities in a

way that promotes ethical standards and advances sci-

entiﬁc knowledge in service of society.

References

The ethics of using artiﬁcial intelligence in sci-

entiﬁc research: new guidance needed for a new

tool

Speciﬁc challenges posed by artiﬁcial intelli-

gence in research ethics

Ethical concerns mount as AI takes bigger decision-

making role in more industries

AI assistance with scientiﬁc writing: Possibili-

ties, pitfalls, and ethical considerations

Legal and Ethical Consideration in Artiﬁcial In-

telligence in Healthcare: Who Takes Responsi-

bility??

About the Author

Jinsu Ann Mathew is a research scholar

in Natural Language Processing and Chemical Infor-

matics. Her interests include applying basic scientiﬁc

research on computational linguistics, practical appli-

cations of human language technology, and interdis-

ciplinary work in computational physics.

References

About airis4D

Artiﬁcial Intelligence Research and Intelligent Systems (airis4D) is an AI and Bio-sciences Research Centre.

The Centre aims to create new knowledge in the ﬁeld of Space Science, Astronomy, Robotics, Agri Science,

Industry, and Biodiversity to bring Progress and Plenitude to the People and the Planet.

Vision

Humanity is in the 4th Industrial Revolution era, which operates on a cyber-physical production system. Cutting-

edge research and development in science and technology to create new knowledge and skills become the key to

the new world economy. Most of the resources for this goal can be harnessed by integrating biological systems

with intelligent computing systems oﬀered by AI. The future survival of humans, animals, and the ecosystem

depends on how eﬃciently the realities and resources are responsibly used for abundance and wellness. Artiﬁcial

intelligence Research and Intelligent Systems pursue this vision and look for the best actions that ensure an

abundant environment and ecosystem for the planet and the people.

Mission Statement

The 4D in airis4D represents the mission to Dream, Design, Develop, and Deploy Knowledge with the ﬁre of

commitment and dedication towards humanity and the ecosystem.

Dream

To promote the unlimited human potential to dream the impossible.

Design

To nurture the human capacity to articulate a dream and logically realise it.

Develop

To assist the talents to materialise a design into a product, a service, a knowledge that beneﬁts the community

and the planet.

Deploy

To realise and educate humanity that a knowledge that is not deployed makes no diﬀerence by its absence.

Campus

Situated in a lush green village campus in Thelliyoor, Kerala, India, airis4D was established under the auspicious

of SEED Foundation (Susthiratha, Environment, Education Development Foundation) a not-for-proﬁt company

for promoting Education, Research. Engineering, Biology, Development, etc.

The whole campus is powered by Solar power and has a rain harvesting facility to provide suﬃcient water supply

for up to three months of drought. The computing facility in the campus is accessible from anywhere through a

dedicated optical ﬁbre internet connectivity 24×7.

There is a freshwater stream that originates from the nearby hills and ﬂows through the middle of the campus.

The campus is a noted habitat for the biodiversity of tropical Fauna and Flora. airis4D carry out periodic and

systematic water quality and species diversity surveys in the region to ensure its richness. It is our pride that

the site has consistently been environment-friendly and rich in biodiversity. airis4D is also growing fruit plants

that can feed birds and provide water bodies to survive the drought.