Cover page

Star Talk- conversation amongst the stars of Indian Astrophysics

The four founding fathers and architectural pillars of IUCAA (Inter-University Centre for Astronomy and

Astrophysics) and, more broadly, the titans of modern Indian astrophysics left us in 2025.

P. P. Divakaran: A distinguished theoretical physicist (formerly of TIFR) who played a crucial role in the

conceptual and academic planning of IUCAA.

Govind Swarup: The ”Father of Indian Radio Astronomy.” He built the Ooty Radio Telescope and the Giant

Metrewave Radio Telescope (GMRT). His collaboration with the other three was essential in making Pune a

global hub for astronomy.

Jayant Narlikar: The Founder-Director of IUCAA. He is the face of Indian cosmology and was the primary

visionary behind creating a center that connected universities to world-class research.

Naresh Dadhich: A renowned relativist and the second Director of IUCAA. He was a key member of the original

team that walked the ”IUCAA trail” from its inception.

image credit : Atharva Pathak and ChatGPT

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.4, No.1, 2026

www.airis4d.com

This edition starts with: “Star Talk: Conversations

Amongst the Stars of Indian Astrophysics pays tribute

to the four founding fathers and architectural pillars

of the Inter-University Centre for Astronomy and

Astrophysics (IUCAA), whose collective vision shaped

modern Indian astrophysics and whose passing in

2025 marked the end of a seminal era. The cover

honours P. P. Divakaran, a distinguished theoretical

physicist and key architect of IUCAA’s academic

foundation; Govind Swarup, the Father of Indian Radio

Astronomy and creator of landmark facilities such as

the Ooty Radio Telescope and GMRT; Jayant Narlikar,

the founder-director of IUCAA and the iconic face

of Indian cosmology; and Naresh Dadhich, eminent

relativist and IUCAA’s second director. Together, they

transformed Pune into a global centre for astronomical

research, leaving behind a legacy of scientiﬁc excellence,

institutional vision, and inspiration for generations to

come.

Arun Aniyan’s article, The Energy Footprint of

AI: Scaling Systems and Environmental Concerns,

examines the rapidly growing energy footprint of

artiﬁcial intelligence as increasingly large and complex

models drive exponential rises in computational and

power demands. It highlights how both training and

everyday use of modern AI systems consume vast

amounts of electricity, strain power grids, increase

carbon emissions, generate e-waste, and intensify water

usage for cooling. In response, technology companies

are exploring high-stakes solutions such as space-based

data centres and dedicated nuclear power, reﬂecting the

severity of the challenge. The article concludes that the

future of AI depends on scaling sustainability alongside

performance through more eﬃcient algorithms, cleaner

energy sources, and improved hardware design, making

environmental responsibility central to AI’s long-term

viability.

Blesson George, in ”Applications of Group

Equivariant Convolutional Neural Networks,” discusses

group equivariant convolutional neural networks (G-

CNNs), which extend standard CNNs by embedding

symmetry principles, such as rotations, reﬂections,

and permutations, directly into neural architectures.

By formulating convolutions over mathematical

groups, G-CNNs ensure that feature representations

transform predictably under these symmetries, leading

to improved generalization and reduced reliance

on data augmentation. The article outlines the

mathematical foundations of equivariance, highlights

rotation-equivariant networks for image data, and

demonstrates applications across image classiﬁcation,

object detection, medical imaging, molecular modeling,

physics-informed learning, and graph-based data. It

concludes that while challenges remain in scalability

and computational cost, G-CNNs provide a powerful,

theory-driven framework for building robust and data-

eﬃcient models across diverse scientiﬁc and real-world

domains.

Why Simple Measures Still Matter in the Era

of Deep Learning by Jinsu Ann Mathew argues that

despite the transformative impact of deep learning,

simple analytical methods remain essential and often

highly eﬀective. It highlights that basic techniques

such as counting, averaging, and simple statistical

measures are eﬃcient, robust, and easier to interpret,

especially when data is limited, noisy, or patterns are

clear. While deep learning excels at capturing complex

details, simple methods provide better transparency,

stability, and practical value in real-world applications

like monitoring, healthcare, and decision-making. The

article concludes that technological progress depends on

balancing simplicity and complexity, choosing methods

that best ﬁt the problem rather than assuming deeper

models are always superior.

Abishek P S. -Plasma Physics- Fusion Energy-

explores the central role of plasma physics in achieving

fusion energy and examines the major challenges

and emerging solutions in this ﬁeld. It explains

why fusion is a crucial clean and abundant energy

source, highlighting its safety, minimal waste, and

potential to support global energy security and

climate goals. The article details key scientiﬁc

and engineering challenges, including sustaining

ultra-hot and unstable plasma, meeting conﬁnement

requirements, developing resilient materials, managing

tritium fuel, and reducing high economic costs. It

then outlines promising solutions such as advanced

magnetic and inertial conﬁnement techniques, improved

plasma control using real-time feedback and AI, novel

materials, high-temperature superconducting magnets,

and international collaboration. Overall, it presents

fusion as a global imperative whose success depends on

coordinated scientiﬁc innovation, engineering advances,

economic strategies, and public trust.

Black Hole Stories-23 Binary Neutron Star

Merger by Ajit Kembhavi recounts the landmark

2017 observation of a binary neutron star merger

(GW170817), the ﬁrst event detected both through

gravitational waves and across the electromagnetic

spectrum. It explains how observations by LIGO,

Virgo, and multiple space- and ground-based telescopes

conﬁrmed long-standing predictions about neutron star

mergers, including gamma-ray bursts, kilonovae, and

the production of heavy elements like gold and platinum.

The event provided deep insights into neutron star

physics, the possible formation of a black hole remnant,

and the origin of elements heavier than iron. It also

enabled an independent measurement of the Hubble

constant and conﬁrmed that gravitational waves travel

at the speed of light, marking a milestone in multi-

messenger astronomy and modern astrophysics.

The article “The Geometry of Truth: Shapley

Values for Model Interpretability” by Linn Abraham

explains how Shapley values can be used to interpret

deep learning models for solar ﬂare prediction by

fairly attributing importance to diﬀerent solar imaging

passbands. By treating each passband as a “player”

in a cooperative game, the method quantiﬁes how

individual wavelengths and their interactions contribute

to a model’s predictions, capturing non-linear synergies

that simpler importance measures miss. Shapley values

enable counterfactual reasoning, allowing researchers

to ask “what if” questions and distinguish meaningful

physical signals from spurious correlations. Overall,

the article presents Shapley-based interpretability as

a rigorous bridge between black-box AI models and

scientiﬁc understanding, helping validate whether

models are learning genuine solar physics rather than

accidental patterns.

The article ”DNA Damage Assessment at the

Single-Cell Level Using the Alkaline Comet Assay

by Aengela Grace Jacob presents the Alkaline Comet

Assay as a sensitive, rapid, and cost-eﬀective technique

for assessing DNA damage at the single-cell level, with

wide applications in biomonitoring, cancer research,

toxicology, and public health. By visualising DNA

strand breaks as comet-like tails formed during alkaline

electrophoresis, the assay quantitatively measures

damage using parameters such as tail length,

The article “Diabetic Retinopathy- Medical Image

Processing” by Geetha Paul provides a comprehensive

overview of diabetic retinopathy (DR), a progressive

microvascular complication of diabetes and a leading

cause of vision loss, detailing its pathogenesis, clinical

stages, symptoms, diagnosis, and management. It

explains how chronic hyperglycaemia damages retinal

blood vessels, leading from non-proliferative changes

such as microaneurysms, haemorrhages, and exudates to

proliferative disease marked by neovascularisation and

severe vision-threatening complications. The article

also describes key retinal features including drusen,

choroidal neovascularisation, and cystoid macular

oedema, and emphasises the importance of early

detection and strict metabolic control. A major focus

is placed on the role of medical image processing and

artiﬁcial intelligence, highlighting how retinal imaging,

automated lesion detection, and deep learning–based

grading pipelines enhance screening eﬃciency, enable

early diagnosis, and support timely intervention to

prevent irreversible vision loss.

The article by Neelima Dubey on “Neurological

Disorders: A Brief Overview” presents a

broad yet integrated overview of neurological

disorders, encompassing both neurodegenerative and

neuropsychiatric conditions, and highlights their

growing global health burden. It outlines how disorders

of the central and peripheral nervous systems lead to

diverse cognitive, motor, and emotional impairments,

with neurodegenerative diseases such as Alzheimer’s,

Parkinson’s, Huntington’s disease, ALS, and multiple

sclerosis characterised by progressive and often

irreversible neuronal loss. The article also emphasises

neuropsychiatric and mood disorders—including major

depressive disorder, bipolar disorders, postpartum mood

disorders, and seasonal aﬀective disorder—linking

behavioural symptoms to underlying neurological

dysfunction. By discussing disease mechanisms,

clinical features, and current therapeutic strategies, the

review underscores the need for integrated, multimodal

approaches that combine pharmacological treatment,

neuromodulation, psychotherapy, and early biomarker-

based diagnosis to improve patient outcomes and quality

of life.

The article “AI in India: From Policy Vision to

Everyday Governance” by Atharva Pathak examines

how artiﬁcial intelligence in India has evolved from

a policy ambition into a practical tool of everyday

governance, reshaping decision-making across sectors

such as welfare, healthcare, education, climate action,

public safety, and research. Anchored by the IndiaAI

Mission, India’s approach emphasises AI as a public

good that augments human judgment rather than

replacing it, supported by shared datasets, Centres

of Excellence, and capacity building for civil servants.

The article highlights AI’s growing role in policy design,

predictive analytics, environmental monitoring, citizen-

centric service delivery, and disaster response, while

underscoring the importance of ethics, transparency,

and accountability. Overall, it presents India as a

model for responsibly integrating AI into governance

to improve administrative eﬃciency, evidence-based

policymaking, and public service outcomes.

News Desk - Meomories Never Fade

Contents

Editorial iii

I Artiﬁcial Intelligence and Machine Learning 1

1 The Energy Footprint of AI: Scaling Systems and Environmental Concerns 2

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

1.2 The Exponential Scaling of AI and Energy Demand . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Innovative, High-Stakes Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.4 Future Energy Demands and Environmental Impact . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Applications of Group Equivariant Convolutional Neural Networks 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Group Equivariance: Mathematical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.3 Rotation-Equivariant CNNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.4 Applications in Computer Vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.5 Medical Imaging Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.6 Scientiﬁc and Physical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.7 Beyond Images: Permutation Equivariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.8 Challenges and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Why Simple Measures Still Matter in the Era of Deep Learning 9

3.1 The Limits of Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.2 Seeing the Big Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.3 Clarity as a Scientiﬁc Advantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.4 Practical Foundations for Sustainable Progress . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

II Astronomy and Astrophysics 12

1 Plasma Physics- Fusion Energy- Challenges and Solutions 13

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.2 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 Black Hole Stories-23

Binary Neutron Star Merger 17

2.1 Binary Neutron Star Merger Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.2 Neutron Star Binaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.3 Electromagnetic Counterpart of the Gravitational Wave Detection . . . . . . . . . . . . . . . . . 18

CONTENTS

2.4 Nature of the Binary Components: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.5 Nature of the Remnant: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.6 Production of Heavy Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.7 Determination of Hubble’s Constant and the Speed of Gravitational Waves . . . . . . . . . . . . 20

3 The Geometry of Truth: Shapley Values for Model Interpretability 22

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3.2 The Marginal Contribution Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3.3 Problem of Interaction (Synergy) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3.4 Causality as a ”Fair Share” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.5 The Counterfactual Gold Standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.6 The Geometry of Truth: The Vector Sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.7 Resolving the ”Clever Hans” Confusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.8 Local vs. Global Insights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

III Biosciences 25

1 DNA Damage Assessment at the Single-Cell Level Using the Alkaline Comet Assay 26

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

1.2 Sample preparation and analysis of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2 Diabetic Retinopathy- Medical Image Processing 30

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2 Drusen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.3 CNV (Choroidal Neovascularization): . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.4 CME (Cystoid Macular Edema): . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.5 Exudates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.6 Microaneurysms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.7 Haemorrhages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.8 Causes and Progression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.9 Symptoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.10 Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.11 Treatment and Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.12 Role of Image Processing in Diabetic Retinopathy . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.13 Automated Detection and Classiﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.14 Steps in Diabetic Retinopathy Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.15 DR grading pipeline end-to-end . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3 Neurological Disorders: A Brief Overview 36

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.2 Neurodegenerative Diseases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.3 Alzheimer’s disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.4 Parkinson’s disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

viii

CONTENTS

3.5 Huntington’s disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.6 Amyotrophic Lateral Sclerosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.7 Multiple Sclerosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.8 Bipolar Disorders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.9 Premenstrual Dysphoric Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.10 Persistent Depressive Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.11 Seasonal Aﬀective Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.12 Postpartum Depression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.13 Postpartum Psychosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.14 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

IV General 43

1 AI in India: From Policy Vision to Everyday Governance 44

1.1 India’s National AI Vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

1.2 Capacity Building for AI-Enabled Governance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

1.3 AI in Policy Design and Decision-Making . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

1.4 Environmental Governance and Climate Intelligence . . . . . . . . . . . . . . . . . . . . . . . . . 46

1.5 Public Safety, Law Enforcement, and Justice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

1.6 Citizen-Centric Service Delivery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

1.7 AI in Healthcare, Research, and Disaster Response . . . . . . . . . . . . . . . . . . . . . . . . . . 46

1.8 Ethics, Trust, and Responsible AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

1.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

Part I

Artiﬁcial Intelligence and Machine Learning

The Energy Footprint of AI: Scaling Systems

and Environmental Concerns

by Arun Aniyan

airis4D, Vol.4, No.1, 2026

www.airis4d.com

1.1 Introduction

The rapid expansion of Artiﬁcial Intelligence

(AI) has brought about revolutionary advancements

across industries, from autonomous vehicles and

personalized medicine to complex ﬁnancial modeling

and large language models. However, this progress is

underpinned by massive computational power, raising

signiﬁcant and escalating concerns about the energy

consumption and the resulting environmental impact of

AI systems.

The sheer scale and complexity of modern AI

models, particularly those based on deep learning and

transformer architectures, are pushing the limits of

current energy infrastructure. Training a single state-

of-the-art model can consume the energy equivalent of

several homes over a year, releasing substantial amounts

of carbon dioxide. This intense energy demand stems

from the computational requirements for both the initial

training phase—which can involve weeks or months of

continuous processing on thousands of specialized chips

(GPUs and TPUs)—and the ongoing inference phase

when the model is actively used to generate predictions

or content.

The escalating energy footprint is forcing

technology companies to explore drastic and innovative

solutions. As the number of data centers required

to house and power these systems grows, so does

the strain on local power grids and water supplies

(needed for cooling). Consequently, companies are

contemplating truly radical measures to secure the

necessary power supply. These include speculative,

cutting-edge proposals such as sending data centers

to space to leverage natural vacuum cooling and solar

energy more eﬃciently, or even building dedicated,

small modular nuclear power plants to guarantee a

stable, high-density, carbon-free energy source solely

for their computational operations.

Ultimately, the future growth of AI is inextricably

linked to addressing its environmental cost, demanding

a concerted eﬀort in hardware eﬃciency, algorithmic

optimization, and the adoption of genuinely sustainable,

low-carbon energy sources.

1.2

The Exponential Scaling of AI and

Energy Demand

The relentless pursuit of peak performance in

state-of-the-art Artiﬁcial Intelligence (AI) models,

particularly within the domains of large language

models (LLMs) and advanced deep learning systems,

has established a direct and undeniable correlation

between model capability and sheer scale. Superior

accuracy, nuanced understanding, and broader

capability are achieved primarily through increasing

the model’s architectural size and the astronomical

volume of data ingested during training. This

fundamental drive for enhancement has initiated

an era of exponential growth in the computational

resources—and consequently, the energy—required to

develop and operate modern AI.

1.3 Innovative, High-Stakes Solutions

The environmental implications of this scaling

are stark. The training phase alone for a single,

massive AI model can necessitate an energy expenditure

that is comparable to, or even exceeds, the lifetime

carbon emissions of multiple passenger vehicles.

This signiﬁcant energy footprint is no longer an

isolated event; as AI technologies are woven into

the very fabric of daily life—powering everything

from bespoke personalized digital recommendations

and sophisticated autonomous vehicle systems to

groundbreaking scientiﬁc research and complex global

predictive modeling—the aggregate energy demand

experiences a steep and continuous ascent. This demand

encompasses both the initial, computationally intensive

training phase and the ongoing, widely distributed

inference phase, where the trained model is actively

used to generate results.

A principal catalyst behind this escalating energy

hunger is the unwavering industry commitment to

developing progressively larger and more powerful

models. This trend is most evident in the growth

trajectory of leading LLMs. In a remarkably short

period, the parameter count—a measure of a model’s

size and complexity—has ballooned from a few

million to over a trillion. Crucially, this aggressive

scaling shows no discernible signs of abating. This

incessant increase in computational load mandates the

construction, maintenance, and continuous operation

of colossal, highly optimized data centers. These data

centers, which require immense and constant streams

of electricity for both computation and cooling, serve

as the physical, infrastructural manifestation of AI’s

burgeoning and voracious appetite for power.

1.3 Innovative, High-Stakes Solutions

In response to this growing energy crisis, major

technology companies are exploring unprecedented

solutions.

1.3.1 Data Centers in Space

The concept of deploying data centers in space, as

recently brought to light by Google’s consideration,

underscores the increasing severity of the energy

and thermal management challenges facing modern

AI infrastructure. On Earth, the relentless growth

in AI model size and computational demands has

pushed existing terrestrial data centers to their limits,

particularly regarding the dissipation of immense waste

heat.

The primary motivation for this extraterrestrial

venture is the exploit the near-absolute zero temperature

of deep space, which oﬀers a virtually limitless and

energy-free heat sink. This would drastically reduce

the need for conventional, energy-intensive cooling

systems (like large-scale HVAC and liquid cooling),

which currently represent a signiﬁcant portion of a data

center’s total energy footprint.

However, moving data centers oﬀ-planet

introduces a new set of substantial logistical and

energetic hurdles:

Initial Deployment Energy: The energy

required for manufacturing, assembly, and most

critically, launching the data center hardware and

supporting infrastructure into orbit or beyond

would be astronomical. This initial ’carbon

investment’ must be oﬀset by long-term energy

savings to make the proposition environmentally

viable.

Maintenance and Longevity: Regular

maintenance, hardware upgrades, and repairs,

which are routine on Earth, become incredibly

complex, costly, and energy-intensive in space.

Missions to service or replace components

would require launching further materials and

personnel.

Communication and Latency: Communicating

with a data center in orbit or on the Moon

introduces signiﬁcant latency delays (light speed

limitations) that could be prohibitive for real-

time AI applications like autonomous vehicles,

high-frequency trading, or interactive virtual

assistants. While communication lasers could

be used, the continuous, high-bandwidth energy

required for uplinks and downlinks would still be

considerable.

Operational Energy: Even in space, power is

1.4 Future Energy Demands and Environmental Impact

required for the computational units themselves,

internal systems, attitude control, station-keeping,

and the sophisticated communication arrays.

While solar power in space is highly eﬃcient, the

power generation and storage infrastructure must

be rugged, redundant, and massive to support an

AI cluster.

Ultimately, while the vision of space-based data

centers oﬀers a potential long-term solution for thermal

management, it also represents an extreme and capital-

intensive attempt to manage the escalating energy

footprint of scaling AI. It is a stark indicator that

terrestrial solutions are rapidly becoming insuﬃcient

for the demands of the next generation of artiﬁcial

intelligence.

1.3.2 Dedicated Nuclear power plans

The escalating energy demands of artiﬁcial

intelligence, particularly for training and running large

language models and complex neural networks, are

pushing tech companies to consider radical energy

solutions. A prime example is the discussion around

industry leaders, such as OpenAI, actively exploring

the feasibility of developing and operating their

own dedicated nuclear power infrastructure. This

initiative, speciﬁcally aimed at powering massive, next-

generation AI data centers, highlights the sheer scale and

consistency of power required that traditional renewable

sources or the existing grid may struggle to meet.

Nuclear power presents a compelling, albeit

controversial, answer to the AI energy crisis. As a

carbon-free energy source, it aligns with sustainability

goals by not directly emitting greenhouse gases during

operation. Crucially, it oﬀers high-density power that

is consistently available—a critical requirement for

24/7 AI operations, unlike intermittent sources such as

solar or wind. This reliability makes it an attractive

proposition for the intense, constant computational load

of advanced AI.

However, the path to nuclear-powered AI is

fraught with signiﬁcant hurdles. The regulatory

landscape surrounding nuclear energy is notoriously

stringent and complex, demanding exhaustive safety

protocols, licensing, and oversight from national

and international bodies. Safety concerns remain

paramount; while modern reactor designs boast

enhanced safety features, the public and environmental

risk associated with any nuclear incident remains

a powerful factor. Furthermore, the long-term

management and disposal of nuclear waste present

an enduring environmental and logistical challenge

that requires robust, permanent solutions, adding

considerable cost and complexity to the entire endeavor.

The exploration of proprietary nuclear power for AI

underscores a pivotal moment where technological

advancement is directly confronting fundamental energy

and environmental policy challenges.

1.4 Future Energy Demands and

Environmental Impact

The accelerating trajectory of Artiﬁcial

Intelligence (AI) development is placing an

unprecedented and rapidly escalating demand on

global energy resources. If the current pace of AI

adoption and scaling continues unabated, the electricity

consumption of the world’s data centers—the essential

infrastructure for AI—is projected to consume a

signiﬁcantly substantial and unsustainable portion of

the total world’s energy usage within the next decade.

This growth is driven by the increasing computational

complexity of state-of-the-art models (like Large

Language Models and foundation models), which

require massive, continuous training and inference

cycles.

The environmental burden of AI extends far beyond

the simple consumption of electricity. A holistic view

reveals a complex set of environmental challenges tied to

every stage of the AI hardware lifecycle and operation.

Carbon Emissions: The Climate Cost of

Computation

The most direct and immediate environmental

concern is the contribution to greenhouse gas

(GHG) emissions. The vast majority of the

world’s electricity grid still relies heavily on

fossil fuels. Consequently, unless the energy

1.5 Conclusion

powering massive AI data centers is sourced

entirely from veriﬁable renewable (e.g., solar,

wind) or other low-carbon sources (e.g., nuclear),

their continuous operation directly contributes to

and exacerbates climate change. The process of

model training, in particular, can be equivalent to

the lifetime emissions of multiple automobiles,

making AI a signiﬁcant new driver of climate

vulnerability.

2. E-Waste and Hardware Lifecycle Challenges

AI is an extremely hardware-intensive

ﬁeld. The specialized computational

accelerators—primarily Graphics Processing

Units (GPUs), Tensor Processing Units

(TPUs), and other custom silicon—required

for high-performance AI processing have a

comparatively short eﬀective lifecycle. The

constant and rapid evolution of AI models and

architectures necessitates frequent upgrades

to more powerful, eﬃcient hardware. This

relentless cycle of replacement creates a

rapidly accelerating and challenging problem

of electronic waste (e-waste). This waste

stream often contains complex assemblies and

toxic materials (including heavy metals and

ﬂame retardants), which pose signiﬁcant risks

to human health and the environment when

improperly disposed of in landﬁlls. Furthermore,

the mining and manufacturing of this specialized

hardware are themselves resource-intensive and

environmentally degrading processes.

Water Usage: A Hidden Drain on Local

Resources

The immense concentration of power and

heat generated by AI data centers necessitates

sophisticated and equally immense cooling

systems. These systems, particularly those that

employ evaporative cooling, require colossal

amounts of water. This consumption of water,

often measured in the millions of liters per day for

a single large facility, puts substantial pressure

on local water resources, especially when data

centers are located in already arid or hot climates.

This excessive water withdrawal can contribute

to or exacerbate drought conditions, reduce

water availability for agriculture and human

consumption, and severely impact local aquatic

ecosystems and biodiversity. The competition for

water between technological infrastructure and

human/environmental needs is rapidly becoming

a critical socio-environmental conﬂict.

1.5 Conclusion

Eﬀectively addressing the profound environmental

challenge posed by the scaling of AI will require a

comprehensive and multi-faceted approach involving

policy, engineering, and research. The primary

objectives must include:

Algorithmic Eﬃciency: Prioritizing research

into developing and deploying more energy-

eﬃcient AI algorithms, including techniques like

model compression, sparse training, and eﬃcient

inference methods, to reduce the computational

resources needed for a given performance level.

Renewable Energy Transition: Mandating and

accelerating the shift of data center operations

entirely to renewable energy sources through

power purchase agreements (PPAs), on-site

generation, and strategic location planning that

optimizes access to clean power grids.

Hardware Optimization and Longevity:

Improving the energy and thermal eﬃciency

of specialized hardware, while also designing

for longer lifecycles, enhanced recyclability, and

establishing robust, safe end-of-life protocols for

e-waste management.

The conversation surrounding AI must fundamentally

shift its focus. It is no longer suﬃcient to merely discuss

scaling capabilities; the imperative now is to scaling

sustainability alongside performance to ensure the long-

term viability and ethical deployment of this powerful

technology.

1.5 Conclusion

About the Author

Dr.Arun Aniyan is leading the R&D for

Artiﬁcial intelligence at DeepAlert Ltd,UK. He comes

from an academic background and has experience

in designing machine learning products for diﬀerent

domains. His major interest is knowledge representation

and computer vision.

Applications of Group Equivariant

Convolutional Neural Networks

by Blesson George

airis4D, Vol.4, No.1, 2026

www.airis4d.com

2.1 Introduction

Let

f ∶ R

→ R

represent an image. A standard

convolutional layer applies a ﬁlter ψ to f as

(f ∗ψ)(x) =

∫

f(y)ψ(x −y)dy. (2.1)

This operation is translation equivariant, meaning

(f ∗ψ) = (T

f)∗ψ, (2.2)

where T

denotes a translation by a.

However, translations alone are insuﬃcient to

model many practical symmetries such as rotations

and reﬂections. Group equivariant CNNs extend this

idea by incorporating general transformation groups

into the convolution operation.

2.2 Group Equivariance:

Mathematical Framework

Let

be a group acting on a space

. A function

is said to be equivariant with respect to group actions

and ρ

out

Φ(ρ

(g)f ) = ρ

out

(g)Φ(f ), ∀g ∈ G. (2.3)

In G-CNNs, the convolution is deﬁned over the

group G:

(f ∗ψ)(g) =

∑

h∈G

f(h)ψ(g

−1

h), (2.4)

for discrete groups, or using integrals for continuous

groups.

This formulation ensures that feature maps

transform predictably under the action of G.

2.3 Rotation-Equivariant CNNs

For image data, a commonly used group is the

planar rotation group

SO(2)

. A rotation operator

acts on an image f as

f)(x) = f (R

−θ

x). (2.5)

A rotation-equivariant layer satisﬁes

Φ(R

f) = R

Φ(f). (2.6)

Such networks are particularly eﬀective in

applications where object orientation is arbitrary,

including aerial imagery and medical imaging.

2.4 Applications in Computer Vision

2.4.1 Image Classiﬁcation

In image classiﬁcation tasks, the goal is to learn a

mapping

F ∶ f → y, (2.7)

where

is a class label invariant under group

transformations.

Equivariance at intermediate layers allows the

network to build invariant representations at the ﬁnal

layer via pooling over the group:

y = max

g∈G

Φ(f)(g). (2.8)

This approach improves generalization while

reducing reliance on data augmentation.

2.7 Beyond Images: Permutation Equivariance

2.4.2 Object Detection and Segmentation

In segmentation tasks, equivariance ensures that

spatial predictions transform consistently:

S(R

f) = R

S(f), (2.9)

where S(f) denotes the segmentation map.

This property is crucial in remote sensing and

biomedical imaging, where object orientation varies

signiﬁcantly.

2.5 Medical Imaging Applications

Medical images can be modeled as functions

deﬁned on continuous domains with rotational

symmetry. G-CNNs incorporate these symmetries

directly, improving performance in tasks such as tumor

detection.

For a classiﬁer

, rotational invariance can be

enforced as

C(R

f) = C(f), (2.10)

which is achieved by combining equivariant layers with

invariant pooling.

2.6 Scientiﬁc and Physical

Applications

2.6.1 Physics-Informed Learning

Physical systems often obey symmetry laws. For

a physical ﬁeld u(x) governed by a PDE,

Lu = 0, (2.11)

equivariance ensures that learned solutions respect

transformation symmetries of L.

G-CNNs have been used to learn mappings

between physical states while preserving rotational

and translational symmetry.

2.6.2 Molecular Modeling

Molecules can be represented as point clouds in

. For a molecular property predictor P ,

P (Rx) = P (x), ∀R ∈ SO(3), (2.12)

ensuring physical consistency. Group equivariant

networks naturally satisfy this constraint.

2.7 Beyond Images: Permutation

Equivariance

For graph-structured data, node permutations form

a symmetry group. A graph neural network layer

permutation equivariant if

Φ(πX)= πΦ(X), (2.13)

where π is a permutation matrix.

This principle underlies applications in chemistry,

social networks, and recommendation systems.

2.8 Challenges and Future Directions

While G-CNNs oﬀer theoretical and practical

beneﬁts, challenges remain in scaling to large

continuous groups and reducing computational

overhead. Hybrid approaches combining group

equivariance with attention mechanisms and

transformers represent an emerging research direction.

2.9 Conclusion

By embedding symmetry principles directly into

neural architectures, group equivariant CNNs provide

robust, data-eﬃcient models across diverse application

domains. The inclusion of mathematically grounded

equivariance constraints bridges the gap between theory

and practice in deep learning.

About the Author

Dr. Blesson George presently serves as

an Assistant Professor of Physics at CMS College

Kottayam, Kerala. His research pursuits encompass

the development of machine learning algorithms, along

with the utilization of machine learning techniques

across diverse domains.

Why Simple Measures Still Matter in the Era

of Deep Learning

by Jinsu Ann Mathew

airis4D, Vol.4, No.1, 2026

www.airis4d.com

Deep learning methods such as neural networks,

transformers, and convolutional models have

transformed modern technology. They power voice

assistants, image recognition systems, and automatic

translation tools. These methods learn complex patterns

by processing large amounts of data through many

layers.

However, alongside these powerful tools, simpler

analytical methods continue to be widely used.

Techniques based on counting, averaging, similarity,

or basic statistical summaries still play an important

role. Their continued use shows that not every problem

requires deep learning—and that simpler approaches

often provide strong, reliable answers.

3.1 The Limits of Complexity

Deep learning models are designed to capture very

detailed relationships in data. For example, a deep

neural network for text analysis may learn sentence

structure, word meaning, and context across many

layers. While this is impressive, it also introduces

challenges.

Such models require large datasets, long training

times, and careful tuning. They can also fail in

unexpected ways if the data changes slightly. For

instance, a deep learning spam ﬁlter trained on one type

of email may perform poorly when spam styles evolve.

In contrast, simple methods such as keyword

counting, frequency thresholds, or basic scoring rules

often work surprisingly well. A spam ﬁlter that checks

how often certain words or symbols appear can already

block many unwanted emails. In problems where

patterns are clear, complex models may add eﬀort

without providing much beneﬁt.

3.2 Seeing the Big Picture

Deep learning excels at learning ﬁne details, but

simple measures are better at capturing overall trends.

For example, image recognition models analyze pixel-

level features across many layers to identify objects.

Yet, in some cases, overall brightness, color distribution,

or shape size is enough to separate one category from

another.

In data monitoring, deep learning models may

analyze every signal variation, while a simple average

or trend line can reveal sudden changes immediately.

For instance, detecting unusual network traﬃc may only

require tracking the total number of requests over time,

rather than modeling each connection individually.

Because simple methods summarize data instead

of examining every detail, they often remain

stable even when data is noisy, incomplete, or

imperfect—conditions where deep learning models

may struggle.

3.3 Clarity as a Scientiﬁc Advantage

One major drawback of deep learning methods is

that they are often diﬃcult to explain. A neural network

might make an accurate prediction, but explaining why

3.5 Conclusion

it reached that conclusion can be challenging even for

experts.

Simple methods oﬀer clear reasoning. If a

document is classiﬁed as suspicious because it contains

repeated unusual patterns or extreme values, the

explanation is easy to understand. This transparency is

especially important in areas like healthcare, education,

and policy-making, where decisions must be justiﬁed.

For example, a medical system that ﬂags patients

based on simple risk scores is easier to trust than a

complex model that provides no clear explanation, even

if the complex model is slightly more accurate.

3.4 Practical Foundations for

Sustainable Progress

Simple methods play an important role in making

technology practical and sustainable. One of their

biggest strengths is eﬃciency. They work quickly, need

less data, and can run on ordinary computers. This

makes them useful in situations where time, money, or

technical resources are limited. For example, a basic

rule-based system can often detect unusual activity in

data immediately, without waiting for a complex model

to be trained.

Simple methods also act as a starting point for

more advanced systems. Before using deep learning,

researchers often begin with simple approaches to

understand the data. These early steps help answer

basic questions such as: Is there a clear pattern? Is the

problem really complex? In many cases, these simple

checks already provide useful results and may even

remove the need for a deep model.

Even when deep learning is used, simple methods

remain important for comparison. They act as

reference points. If a complex model does not perform

signiﬁcantly better than a simple one, then the extra

eﬀort may not be worthwhile. This comparison helps

ensure that technology is used wisely and responsibly.

Most importantly, simple methods encourage a

balanced view of progress. Advancement does not

mean replacing all old ideas with new ones. Instead,

it means choosing the right level of complexity for

each problem. Simple approaches remind us that

understanding, eﬃciency, and reliability are just as

valuable as accuracy. In this way, they support long-

term progress that is both practical and meaningful.

3.5 Conclusion

The rapid growth of deep learning has expanded

what machines can do, but it has not reduced the

value of simple methods. Instead, it has highlighted

their importance. Simple measures continue to matter

because they are easy to understand, eﬃcient to use,

and reliable across many real-world situations.

While deep learning is powerful for complex

problems, simple approaches often provide strong

results with far fewer resources. They help reveal

overall patterns, oﬀer clear explanations, and serve

as dependable foundations for advanced systems. In

many cases, they guide researchers in deciding whether

deeper complexity is truly necessary.

Ultimately, progress in technology is not about

choosing between simplicity and complexity, but about

balancing the two. Simple methods remind us that

eﬀective solutions are not always the most complicated

ones—they are the ones that best match the problem at

hand.

References

When to Use Machine Learning Instead of Deep

Learning — Why Simpler Still Wins

Are Statistical Methods Obsolete in the Era of

Deep Learning?

What is Deep Learning & Why Does It Matter?

Deep Learning vs. Machine Learning: What’s

the Diﬀerence?

The Case Against Deep Learning: When Simple

Models Beat Complex Ones

3.5 Conclusion

About the Author

Jinsu Ann Mathew is a research scholar

in Natural Language Processing and Chemical

Informatics. Her interests include applying basic

scientiﬁc research on computational linguistics,

practical applications of human language technology,

and interdisciplinary work in computational physics.

Part II

Astronomy and Astrophysics

Plasma Physics- Fusion Energy- Challenges

and Solutions

by Abishek P S

airis4D, Vol.4, No.1, 2026

www.airis4d.com

1.1 Introduction

Plasma is the essential medium in which nuclear

fusion takes place, and understanding its behaviour is

central to fusion energy research. Often referred to

as the “fourth state of matter,” plasma is an ionized

gas where electrons are stripped from atoms, leaving

behind positively charged ions[1]. Because both ions

and electrons carry charge, they interact strongly with

electromagnetic ﬁelds, making plasma both diﬃcult

to control and uniquely suited for fusion reactions. In

a fusion environment, plasma provides the conditions

where light nuclei such as deuterium and tritium collide

and fuse into heavier nuclei, releasing immense amounts

of energy. Fusion energy is increasingly seen as a

necessity because of the global challenges we face

in energy, climate, and sustainability. The world’s

population and industrial activity continue to grow,

driving up electricity demand at a time when fossil fuels

are depleting and causing severe environmental damage.

Fusion oﬀers a solution that is both abundant and clean.

Its fuel hydrogen isotopes like deuterium and tritium

can be sourced from water and lithium, which are widely

available and evenly distributed across the planet. This

means fusion energy could provide a virtually limitless

supply of power without the geopolitical tensions

associated with oil and gas reserves. Unlike fossil

fuels, fusion produces no greenhouse gases, making it

a cornerstone technology for combating climate change

and achieving net-zero emissions targets.

Another reason fusion is needed is its safety

advantages compared to nuclear ﬁssion. Fusion

reactions do not rely on chain reactions, so there is

no risk of runaway meltdowns like those seen in ﬁssion

reactors. The waste produced is minimal and short-

lived, unlike the long-lived radioactive byproducts of

ﬁssion. This makes fusion far more acceptable from

both an environmental and public safety perspective.

Fusion also has the potential to deliver continuous,

reliable baseload electricity, unlike solar and wind,

which are intermittent and weather-dependent. By

complementing renewables, fusion could stabilize

energy grids and ensure a steady supply of power for

industries, cities, and households.

Economically, fusion energy is needed to secure

long-term energy independence. Fossil fuels are ﬁnite

and subject to volatile markets, while fusion fuel

is abundant and inexpensive once the technology is

mastered. Fusion reactors could also be designed to

provide not just electricity but high-grade industrial

heat, desalination for freshwater, and even medical

isotopes, diversifying their beneﬁts. In the long run,

fusion could reduce reliance on imported fuels, lower

energy costs, and drive new industries built around

advanced materials and superconducting technologies.

Finally, from a societal perspective, fusion

represents hope for a sustainable future. It addresses

the dual challenge of meeting rising energy demand

while protecting the planet from climate catastrophe.

It also inspires international collaboration, as seen in

projects like International Thermonuclear Experimental

Reactor (ITER), where countries pool resources and

1.3 Solutions

expertise to achieve a shared goal. Public acceptance of

fusion is generally high because it is perceived as safer

than ﬁssion and cleaner than fossil fuels, but continued

investment and transparent communication are needed

to maintain trust.

In essence, the need for fusion energy stems

from its ability to provide abundant, safe, and climate-

friendly power that can sustain human progress without

compromising the environment. It is not just a

technological ambition, it is a global imperative for

energy security, climate stability, and a sustainable

future.

1.2 Challenges

Harnessing plasma for fusion energy presents

challenges across scientiﬁc, engineering, economic,

and societal domains. Scientiﬁcally, the greatest hurdle

is sustaining plasma at extreme temperatures of 100–150

million degrees Celsius. At these conditions, plasma

becomes unstable and prone to turbulence, disruptions,

and instabilities such as kink and ballooning modes,

which can cause sudden energy losses or damage to

the reactor. Another critical challenge is meeting the

Lawson criterion, which requires the right balance of

plasma density, temperature, and conﬁnement time. If

this balance is not achieved, plasma cools before fusion

reactions can occur, preventing net energy gain.

From an engineering perspective, plasma interacts

aggressively with reactor materials. Plasma-facing

components such as divertors and ﬁrst walls are

constantly bombarded by high-energy particles and

neutrons, leading to erosion, embrittlement, and

structural degradation. Developing materials that can

withstand these conditions over long durations is a major

challenge. Magnetic conﬁnement devices like tokamaks

and stellarators also demand extremely precise magnetic

ﬁelds to keep plasma stable, and even small deviations

can destabilize conﬁnement[2,3]. Fuel handling adds

further complexity, particularly with tritium, which

is radioactive and scarce. Safe and eﬃcient tritium

breeding and storage systems are essential for future

reactors.

Economically, fusion reactors are expensive

to build and operate. They require advanced

superconducting magnets, powerful laser systems,

and specialized materials, all of which drive up

costs. Current experiments often consume more

energy than they produce, so achieving net-positive

energy output remains a critical milestone. Scaling

up experimental devices to commercial power plants

introduces additional complexity, as larger systems

amplify engineering and ﬁnancial challenges.

On the societal and regulatory side, fusion is

generally considered safer than ﬁssion because it does

not involve chain reactions, but risks remain. Tritium

leakage, neutron radiation exposure, and accidents

during plasma disruptions are potential hazards. Strict

safety standards are needed to regulate radioactive

materials and manage neutron-activated waste. Public

acceptance is also a challenge, as fusion projects require

long timelines and massive investments, often leading

to skepticism about feasibility.

Plasma conﬁnement is one of the greatest obstacles

in fusion research. Because plasma cannot touch

physical walls without cooling instantly, scientists use

magnetic conﬁnement (tokamaks and stellarators) or

inertial conﬁnement (lasers or particle beams) to keep

plasma stable long enough for fusion reactions to

occur. Magnetic conﬁnement relies on strong magnetic

ﬁelds to trap plasma in a toroidal shape, while inertial

conﬁnement compresses fuel pellets so rapidly that the

plasma’s own inertia prevents it from dispersing.

1.3 Solutions

Addressing the challenges of plasma conﬁnement

in fusion energy requires solutions that span science,

engineering, economics, and society. On the scientiﬁc

side, researchers are working to better understand

and control plasma instabilities, which are one of

the biggest obstacles to sustained fusion. Advanced

magnetic conﬁgurations in devices like tokamaks and

stellarators are being designed to minimize turbulence

and disruptions. Real-time feedback systems that use

microwaves, neutral beams, or magnetic perturbations

are being developed to actively suppress instabilities

such as edge-localized modes. In addition, scientists

1.3 Solutions

are experimenting with shaping plasma proﬁles to

create transport barriers that reduce energy leakage,

while artiﬁcial intelligence is increasingly being used

to predict and prevent disruptions before they occur[4].

In inertial conﬁnement approaches, precision in laser

targeting and capsule design is being improved to

achieve perfectly symmetrical implosions, which are

critical for ignition.

From an engineering perspective, the focus is

on building reactors that can withstand the extreme

conditions created by plasma. Plasma-facing materials

such as tungsten composites, liquid metals like lithium,

and radiation-resistant alloys are being tested to handle

intense heat ﬂuxes and neutron bombardment. Divertor

systems with innovative geometries, such as the

snowﬂake or super-X designs, are being developed

to spread heat loads and protect reactor walls. High-

temperature superconducting magnets are another

breakthrough, allowing stronger magnetic ﬁelds in

smaller devices, which improves conﬁnement and

reduces reactor size. Tritium breeding blankets are

being designed to ensure a self-suﬃcient fuel cycle,

while robotic systems for remote handling are being

created to safely replace and repair components without

human exposure to radiation[3].

Magnetic conﬁnement devices include tokamaks,

which use a combination of toroidal and poloidal

magnetic ﬁelds along with plasma currents to conﬁne

plasma. While eﬀective, tokamaks are prone to

instabilities due to their reliance on plasma current.

Stellarators, by contrast, use twisted external magnetic

coils to generate complex three-dimensional ﬁelds that

conﬁne plasma without requiring a plasma current,

making them more stable for steady-state operation,

though their design is far more complex. Simpler

devices like magnetic mirrors trap plasma between

regions of stronger magnetic ﬁelds, but suﬀer from

particle losses and are less eﬃcient for large-scale

fusion.

Inertial conﬁnement takes a diﬀerent approach

by compressing tiny fuel pellets, typically deuterium-

tritium mixtures, with powerful lasers or particle beams.

The implosion is so rapid that the fuel’s inertia prevents

it from dispersing before fusion occurs. This method

achieves extremely high densities, far greater than

magnetic conﬁnement, but only for nanoseconds. The

main challenge lies in achieving perfect symmetry

in the implosion, as even small imperfections can

prevent ignition. Despite these diﬃculties, inertial

conﬁnement has shown promise, with facilities like the

National Ignition Facility (NIF) achieving signiﬁcant

breakthroughs in recent years.

Ultimately, the central challenge of conﬁnement

is plasma stability. Plasmas are highly dynamic and

prone to instabilities and turbulence, which degrade

conﬁnement and lead to energy losses. Plasma

also loses energy through radiation and particle

transport, requiring continuous input to sustain fusion

conditions. Designing magnetic geometries that

minimize instabilities while maximizing conﬁnement

time is one of the most diﬃcult aspects of fusion

research. In inertial conﬁnement, precision is

paramount: the implosion must be perfectly uniform,

and lasers must deliver energy with extreme accuracy.

Overcoming these challenges is key to unlocking the

promise of fusion energy as a safe, clean, and virtually

limitless power source.

Economically, fusion projects are exploring

ways to reduce costs and accelerate deployment.

Compact reactor designs, such as those using high-

ﬁeld superconducting magnets, aim to deliver the

same performance as larger devices but at lower cost.

Modular construction strategies allow reactors to be

built in standardized sections, simplifying maintenance

and reducing expenses. Staged deployment, where

components are tested in smaller facilities before scaling

up, helps spread investment and reduce technical risk.

Advances in manufacturing, including 3D printing and

precision coil winding, are lowering production costs

for complex reactor parts. Fusion plants may also

diversify their revenue streams by providing industrial

heat, desalination, or medical isotopes in addition to

electricity.

On the societal and regulatory side, safety and

public trust are critical. Fusion is inherently safer than

ﬁssion because it does not involve chain reactions, but

risks such as tritium leakage and neutron radiation must

still be managed. Strict international safety standards

1.3 Solutions

are being established to regulate fuel handling and

waste management. Transparent communication of

milestones and performance data helps build credibility

and public support, while global collaboration through

projects like ITER pools resources and expertise to

accelerate progress. Public engagement programs

are also important to explain fusion’s beneﬁts and

challenges, helping overcome skepticism about its long

timelines and high costs.

Taken together, these solutions form a

comprehensive roadmap for overcoming the obstacles of

plasma conﬁnement. By combining scientiﬁc advances

in plasma control, engineering innovations in materials

and magnets, economic strategies to reduce costs, and

societal frameworks to ensure safety and acceptance,

fusion energy can move closer to becoming a practical,

clean, and virtually limitless power source for the future.

References

[1] Francis, F, Chen., (2015). “Introduction to

Plasma Physics and Controlled Fusion (3rd ed.)”.

[2] Miyamoto, K. (1979). “Plasma physics for

nuclear fusion”.

[3] Miyamoto, K. (2005). “Plasma physics

and controlled nuclear fusion.” Berlin, Heidelberg:

Springer Berlin Heidelberg.

[4] Tajima, T. (2018). “Computational plasma

physics: with applications to fusion and astrophysics”.

CRC press.

About the Author

Abishek P S is a Research Scholar in

the Department of Physics, Bharata Mata College

(Autonomous) Thrikkakara, kochi. He pursues

research in the ﬁeld of Theoretical Plasma physics.

His works mainly focus on the Nonlinear Wave

Phenomenons in Space and Astrophysical Plasmas.

Black Hole Stories-23

Binary Neutron Star Merger

by Ajit Kembhavi

airis4D, Vol.4, No.1, 2026

www.airis4d.com

In this story we will describe the merger of

the two components of a binary neutron star, which

was observed by the LIGO and Virgo detectors in

August 2017. The electromagnetic emission from the

merger was also observed from the ground and space,

which led to unprecedented interest in the event by

the astronomical community. A very large number

of research papers were written, and some eﬀects

predicted decades ago were observed. At the time

of writing, this remains the only merger observed

by gravitational wave detectors which has also been

observed at electromagnetic wavelengths.

2.1 Binary Neutron Star Merger

Detection

The story began on August 17, 2017, when

the advanced LIGO (aLIGO) detectors at Hanford

and Livingston, and the advanced VIRGO (aVIRGO)

detector in Italy, detected a signal which lasted about

100 seconds. Following convention, the signal was

named GW170817. From simple visual inspection of

the two aLIGO signals it was apparent that the detection

could have been generated by the spiral-in of a neutron

star binary.

A cosmic Gamma-ray burst was independently

observed by the Gamma-ray burst Monitor (GBM)

on the Fermi satellite just 1.7 seconds after the end

of the gravitational wave detection. A Gamma-ray

burst (see below for a description) is expected in a

binary neutron star merger, because of the enormous

amount of electromagnetic energy emitted. A world-

wide alert was therefore generated so that astronomers

could prepare to observe the event with ground and

space based telescopes. Quick analysis of data from

the three gravitational wave detectors made it possible

to locate the source within an area of about 31 square

degrees of the sky.

2.2 Neutron Star Binaries

In the case of the black hole binaries detected by

aLIGO, the component black holes before the merger,

the system during the merger, the ringdown phase after

the merger and the ﬁnal black hole are all invisible

electromagnetically. Only the gravitational waves

emitted can be detected. Neutron star binaries are

quite diﬀerent from black hole binaries. In BHS-17

and BHS-18, we have discussed in detail the binary

pulsar PSR1913+16, in which one component of the

binary is a neutron star which is radio pulsar, while

the other component is a non-pulsating neutron star.

It is the radio pulsations which enabled the discovery

of the object, and accurate measurements of the small

changes in the observed period led to the identiﬁcation

of the binary system and led to the determination of the

properties of the neutron stars and the binary. It was

observed that the size of the binary is shrinking, which

could be attributed to energy loss due to the emission

of gravitational waves by the binary. As the binary

becomes more compact, the rate at which energy is

2.3 Electromagnetic Counterpart of the Gravitational Wave Detection

lost increases, so that the binary shrinks in size even

faster. It is estimated that the spiral-in process will be

completed and the neutron stars will merge in about

300 million years. GW170817 is just such a merger.

The process of merger of the two neutron stars

is very violent. As the stars approach each other ever

more closely, the gravitational ﬁeld of each neutron star

distorts the shape of the other star, like the tides produced

in the oceans of the Earth due to the gravitational ﬁelds

of the Sun and the Moon. As the distance between the

two neutron stars becomes comparable to their radius,

which is about 10 km, the tidal force becomes so large

that each star is torn apart and the matter from the two

merges together. The gravitational energy released in

the process leads to a tremendous explosion in which

some of the matter is ejected from the system. The

remaining matter undergoes a collapse due to the large

gravitational force dragging the matter inwards. If the

mass of the collapsing core is less than the maximum

permissible mass of a neutron star, then as discussed

in earlier stories, the collapsing object forms a neutron

star. But if the mass of the remnant is more than the

maximum mass limit for a neutron star, a black hole

must be formed from the merger.

2.3 Electromagnetic Counterpart of

the Gravitational Wave Detection

Gamma-Ray Bursts:

The explosion which results during the neutron star

merger is expected to produce a Gamma-ray burst. In

such bursts a tremendous amount of energy is released,

in which ﬁrst there is explosive release of Gamma-

rays, which is followed by electromagnetic radiation of

various kinds, including X-rays, optical radiation and

radio waves. Hundreds of Gamma-ray bursts lasting for

various duration, ranging from a fraction of a second

to hundreds of seconds and longer have so far been

observed. The amount of energy emitted in such a short

time is equivalent to the total energy emitted by a star

like the Sun in about a trillion years. Some of the bursts

with short duration of less than 2 seconds are believed

to be produced by merging neutron stars.

Properties of the System:

The detection of the Gravitational wave source

GW170817 and the independent detection of a

Gamma-ray burst within 1.7 seconds of it generated

tremendous excitement, as the event could lead to better

understanding of gravitational wave sources, Gamma-

ray bursts and very high density matter in its most

extreme form. On the gravitational side, analysis of the

data from the aLIGO and aVIRGO detectors led to very

interesting results. The total mass of the system before

the merger was found to be in the range of 2.73 and

2.82 Solar masses, the mass of one of the components

was in the range of 0.86 to 1.36 Solar masses, while

the mass of the other component was in the range of

1.36 to 2.26 Solar masses. The masses cannot be more

accurately determined because these are linked to how

much spin the two objects have, and that is not known

at the present. Since the two components have merged

and do not any longer exist as independent objects,

their further observation is not possible. But over time

it will be possible to analyse the data in increasingly

sophisticated ways, so that more information will be

obtained and used to determine the properties more

accurately.

A very important quantity associated with the

source is its distance from us. For a gravitational wave

detection, the luminosity i.e. the total gravitational

wave energy emitted by system per second is known, as

well as the ﬂux measured on the Earth. From these two

quantities the distance to the source is determined. The

distance to GW170817 determined from gravitational

wave observations is about 130 million light years,

which is much less than the distance to the detections

made earlier. As a consequence, GW170817 is the

most intense gravitational wave source detected at that

time. Because it was so intense, the source could be

observed by the aVIRGO as well, in addition to the

two aLIGO detectors. That allowed the position in the

sky to be determined to within an area of 31 square

degrees. The source is near the southern end of the

constellation Hydra. Astronomers searched this part of

the sky with a multitude of telescopes and soon found

a transient source which had appeared in the galaxy

NGC3993, about 75,000 light years from its centre.

2.6 Production of Heavy Elements

Detailed observations of the source were carried out at

optical, near-infrared, X-ray and radio wavelengths by

various ground and space based telescopes and a great

deal of data was obtained.

2.4

Nature of the Binary Components:

The mass of each component was well within

the range of known neutron star masses which were

determined from the many known binary neutron stars

and X-ray binaries. The two masses were also well

below the mass of known black holes associated with

binary stellar systems, as determined from gravitational

wave and X-ray data. It is therefore reasonable to

assume that the two components were neutron stars,

and much of the work on the source has proceeded

on that assumption. Nevertheless, it is important to

keep in mind that either of the objects could have

been compact exotic objects like quark stars, or they

could be low mass black holes. Further information

will be obtained as the analysis of data on GW170817

progresses and more such systems are observed in the

future. But it is clear that both objects could not have

been black holes. The observation of Gamma-rays and

other electromagnetic radiation from GW170817 mean

that at least one of the objects must have had ﬁnite size,

with matter ﬂow having taken place from this object to

the other one. If both components were black holes then

no electromagnetic radiation would have been detected.

2.5 Nature of the Remnant:

What is the nature of the remnant formed from

the merger? The mass of the remnant, which is 2.82

Solar masses, is again in the range of known neutron

star masses, but close to the upper limit. So while a

large mass neutron star could be formed, depending on

its mass an spin, it would be short lived, collapsing in

less than a second, or it could be long lived, existing for

10,000s, or even much longer duration, before collapsing

to a black hole. It could also be long lived stable neutron

star. A neutron star remnant would emit gravitational

waves upon formation because of irregularities present

in its structure, but these are at high frequencies which

cannot be eﬃciently detected by aLIGO. The merger

could also lead to the formation of a black hole, in

which case there would be the ringdown gravitational

wave emission as in the case of GW150914. But this

again is at high frequencies around 6000 Hz, at which

the detectors do not have suﬃcient sensitivity. The

nature of the object remains unresolved as of 2025.

2.6 Production of Heavy Elements

Observations of the electromagnetic counterpart

of GW170817, known as EM170817, at various

wavelengths have led to a great deal of information

about the nature of the spectrum in the optical and

near-infrared regions. From the shape of these spectra

it can be concluded that several elements heavier than

iron must be present in the matter being expelled during

the merger. The production of such elements has so

far remained a puzzle. It is known that the lightest

elements hydrogen and helium and some light elements

like lithium, beryllium and boron in trace quantities are

produced in the Big Bang. More helium and all the

heavier elements continuing up to iron are produced

in the interiors of stars due to nucleosynthesis. It has

long been believed that elements heavier than iron,

particularly those which require a mechanism involving

the rapid capture of neutrons by atomic nuclei, in what

is known as r-process, were produced in supernovae.

More recently it has been argued that such elements

could be produced in the neutron rich matter present

during neutron star mergers. The observed spectra of

EM170817 are consistent with this expectation. A total

of 16,000 times the mass of the Earth in heavy elements

is believed to have formed, including approximately

10 Earth masses just of the two elements gold and

platinum (Wikipidia). Some of the nuclei produced in

this manner are unstable and undergo radioactive decay,

leading to explosive electromagnetic emission which

is known as a kilonova. The observed emission from

EM170817 is consistent with known theoretical models

of kilo novae.

2.7 Determination of Hubble’s Constant and the Speed of Gravitational Waves

2.6.1 Kilonova:

A number of short lived Gamma-ray bursts have

so far been observed. It is believed that the emission

from such sources is produced by matter moving in a

narrow jet at speeds very close to the speed of light

towards the observer. It has also been believed that

such short period Gamma-Ray bursts are produced

during the merger of neutron stars. Observation of the

emission from EM170817 show that while it is indeed a

short duration Gamma-ray burst, the rate of Gamma-ray

emission from it is about 10,000 times smaller than

other such known bursts. Detailed considerations show

that EM170817 must be quite diﬀerent from the earlier

sources, and new models are required to explain its

behaviour. While several models have been considered,

there is no agreement yet as to the correct one which is

consistent will all observations.

2.7 Determination of Hubble’s

Constant and the Speed of

Gravitational Waves

Our Universe can be considered to be

homogeneous and isotropic, which means that the

distribution of its contents is the same at all positions

Universe and the same in all directions, when viewed

from any position The Universe is known to be

expanding, and its scale at a given cosmological epoch

is determine by an expansion factor S(t). The form of

S(t) is determined by solving Einstein’s equations of

gravitation whose form is greatly simpliﬁed due to the

homogeneity and isotropy. The rate of expansion of the

Universe is determined by the Hubble parameter

H =

S(t)

dS(t)

The value of the of H at the present epoch t

S(t

)

dS(t

)

known as Hubble’s constant is very important for

cosmology. The current age of the Universe and the

cosmological distances of objects depend on it.

The expansion of the Universe leads to redshift of

light received from objects at cosmological distances.

The redshift z for an object is

z =

∆λ

where

∆λ

is the change in the wavelength

of some

speciﬁc spectral line emitted by a source. It is found that

for objects at cosmological distances,

∆λ

and therefore

z is positive, so distant objects are moving away from

us, which establishes that the Universe is expanding.

In the 1930s, it was discovered by Edwin Hubble and

Georges Lemaitre that the velocity with which distant

objects are receding from us is proportional to their

distance, which can be written as

v =

where c is the speed of light. For v smaller than c, it

can be shown from cosmological considerations that

z =

These relations are approximate and valid only for v

c, i.e. z

1. Observationally distant galaxies

and quasars are found to have z much larger than 1. In

such cases exact counterparts of the above equations,

derived taking into account the curvature of space-time

over large distances, are used.

The redshift of an object can be determined from

its observed spectrum. If the distances to a number of

objects with known z can be measured, then

can

be determined. This is a rather diﬃcult exercise, and

currently the value of the Hubble constant determined

using Supernovae Type Ia as distance indicators is

73km/sec/M pc

, where Mpc is one million parsec.

A smaller value of 67 km/sec/Mpc is obtained using

cosmic microwave background data from the Planck

mission.

The above methods for determining

use distances obtained through electromagnetic

measurements. Observations of the neutron star binary

provide an independent way of measuring distances.

The distance to the binary merger is determined

from the gravitational wave observations as discussed

above, while the cosmological redshift is determined

from the optical spectrum of the electromagnetic

2.7 Determination of Hubble’s Constant and the Speed of Gravitational Waves

counterpart. Combining the two leads to determination

of Hubble’s constant. For small redshifts, Hubble’s

constant obtained in this manner is consistent with the

value obtained by conventional means which use only

electromagnetic observations. With further neutron

star binary observations, the precision with which

Hubble’s constant is determined from gravitational

wave observations will improve.

Distances are also determined for black hole binary

mergers. But here the redshift is not determined, since

there is no electromagnetic radiation. But if the host

galaxy for in which the merger is located is known, then

the measured redshift of the galaxy provides the redshift

of the merger. The diﬃculty here is that because the

position in the sky of a gravitational wave source is

diﬃcult to determine, the host galaxy is not known

precisely. There are also other techniques which can be

used to estimate the redshift. It is expected that as more

data becomes available, a precise value of the Hubble’s

constant based on gravitational wave measurements

will emerge.

The Gamma-ray burst associated with the neutron

star merger was observed 1.74 s after the merger. From

the delay it is possible to determine that the speed

gravitational waves is the same as the speed of light, to

within one millionth of a billionth of the light speed,

conﬁrming with great precision Einstein’s prediction

that the two speeds must be the same.

About the Author

Professor Ajit Kembhavi is an emeritus

Professor at Inter University Centre for Astronomy

and Astrophysics and is also the Principal Investigator

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

activities from the late 80s to promote astronomy

research in Indian universities.

The Geometry of Truth: Shapley Values for

Model Interpretability

by Linn Abraham

airis4D, Vol.4, No.1, 2026

www.airis4d.com

3.1 Introduction

In the study of solar ﬂares, we have moved past the

era of simple observation into the era of deep learning.

While models can now predict ﬂares from SDO/AIA

images with high accuracy, they remain ”black boxes.”

To turn these predictions into scientiﬁc discovery, we

can try to answer a question like which speciﬁc AIA

passband (e.g., 94

A, 171

A, 131

A) is the primary driver

of the model’s logic. We achieve this using Shapley

values, a method from cooperative game theory that

provides a mathematically ”fair” way to distribute credit

among diﬀerent physical inputs.

3.2 The Marginal Contribution

Experiment

To understand the importance of a passband, we

treat the AIA channels as players in a game. The

”payout” of the game is the ﬂare probability predicted

by the model. The method calculates the Marginal

Contribution of each passband. Imagine we are testing

the importance of the 94

A (ﬂaring iron) channel. The

SHAP (SHapley Additive exPlanations) method follows

this protocol:

It takes a subset of other channels (e.g., just

171

A and 304

A) and records the model’s ﬂare

prediction.

It then adds the 94

A channel to that group and

records the new prediction.

The diﬀerence between those two scores is the

”marginal contribution” of 94

A for that speciﬁc

combination.

The Calculation: The ﬁnal Shapley value is the

weighted average of these marginal contributions

across every possible combination of channels.

By testing the channel in every possible context,

we ensure that we aren’t just seeing a ﬂuke; we are seeing

the indispensable value of that speciﬁc wavelength to

the model’s ”reasoning.”

3.3 Problem of Interaction (Synergy)

Solar ﬂares are inherently non-linear, emergent

phenomena. Often, a single passband carries very little

predictive weight on its own. For example, a slight

brightening in the 171

A channel (the quiet corona)

might be routine. However, when that brightening

occurs in tandem with a speciﬁc magnetic conﬁguration

in the 1600

A channel (the photosphere), it becomes a

critical precursor.

Traditional importance metrics—such as

Permutation Importance or Saliency Maps—struggle

with this because they often treat features as independent

variables. Because the Shapley method calculates

the contribution of 171

A within the context of every

possible subset (including the subset containing

1600

A), it captures these Interaction Eﬀects. It

essentially maps the ”team chemistry” of the AIA

channels, revealing how diﬀerent layers of the solar

atmosphere work in concert to trigger an eruptive

event.

3.6 The Geometry of Truth: The Vector Sum

3.4 Causality as a ”Fair Share”

Philosophically, we often argue about ”The Cause”

of a ﬂare. Is it magnetic shear or ﬂux emergence? In

reality, it is a ”cooperative game” played by multiple

physical layers.

Shapley values allow us to move away from looking

for a ”silver bullet” and instead accept a Distributed

Causality. The method is governed by the Axiom of

Eﬃciency, which states that the sum of all Shapley

values must exactly equal the ﬁnal prediction. This

satisﬁes the scientiﬁc need for a quantitative, additive

explanation: ”0.6 of this ﬂare probability was ’caused’

by the 94

A channel, and 0.4 was ’caused’ by the 131

channel.”

3.5 The Counterfactual Gold

Standard

The most powerful aspect of this method

is Counterfactual Reasoning. In philosophy, a

counterfactual asks: ”What would have happened in

a world where everything stayed the same, but this

one passband was diﬀerent?” In a standard neural

network, you cannot simply ”turn oﬀ” a passband

because the model expects a complete tensor input.

To bypass this, the Shapley method uses a ”Reference

Baseline”—usually the average value of a channel

across the entire dataset—to simulate the absence of

information.

By mathematically simulating these ”alternative

solar worlds” during the marginal contribution tests, the

method provides a rigorous answer to the ”What if?”

question. It allows the researcher to move from a vague

intuition like ”I think this channel is important” to a

concrete, evidence-based claim: ”I have mathematically

proven that this speciﬁc 94

A emission is the pivot point

upon which the model’s entire prediction rests.”

3.6 The Geometry of Truth: The

Vector Sum

This approach provides what can be described as a

Geometry of Truth because of the ”Axiom of Eﬃciency.”

In many statistical models, the ”importance” of various

features doesn’t necessarily add up to the ﬁnal result;

there is often unexplained variance or overlapping credit.

Shapley values, however, create a perfect closed system.

If the model starts at a ”base rate” (the average

ﬂare probability across your entire dataset, say 5%)

and ends at a 90% ﬂare prediction for a speciﬁc active

region, the Shapley values of the individual passbands

act as vectors that must bridge that 85% gap exactly.

Each AIA channel pushes the prediction toward or

away from the ﬂare. When you sum these vectors, they

land precisely on the model’s ﬁnal output. There is

no ”missing logic” and no overlap. Every percentage

point of the model’s certainty is accounted for and

assigned to a speciﬁc physical layer of the Sun. This

geometric decomposition ensures that the explanation

is as rigorous as the mathematical model itself.

3.7 Resolving the ”Clever Hans”

Confusion

A common point of confusion in interpretability

is whether SHAP shows us how the Sun works. It is

vital to clarify: SHAP shows the truth of the model, not

necessarily the truth of nature. In machine learning,

we often encounter the ”Clever Hans” eﬀect—where a

model produces the right answer for the wrong reason.

For instance, a model might achieve high accuracy by

looking at a speciﬁc calibration artifact in the 171

channel rather than the actual magnetic reconnection

physics. By using Shapley values, we can perform

a ”sanity check.” If the model consistently assigns

high importance to the 94

A and 131

A channels (the

hottest plasma), we gain conﬁdence that the model

has successfully captured the underlying physics. If

it ignores these and relies on irrelevant data, SHAP

exposes the model as a statistical ﬂuke rather than a

scientiﬁc instrument.

REFERENCES

3.8 Local vs. Global Insights

Finally, the Shapley method allows us to oscillate

between two scales of discovery:

Local Attribution: For one speciﬁc ﬂare, we can

generate a spatial heatmap. This shows us not just

which passband mattered, but where in the image the

model was looking—perhaps a speciﬁc brightening at

the magnetic neutral line.

Global Interpretation: By averaging these values

across thousands of images, we can rank the AIA

passbands. This tells us which physical ”thermometer”

the model trusts most across the entire solar cycle.

3.9 Conclusion

By treating the model as a ”cooperative game,” we

bypass messy debates about subjective importance. We

treat the model’s decision-making process as a territory

to be mapped. Shapley values provide the coordinates,

showing us exactly which physical ”expert” the model

trusted most to call the ﬂare.

References

[Munn and Pitman(2022)]

Michael Munn and David

Pitman. Explainable AI for Practitioners:

Designing and Implementing Explainable ML

Solutions. O’Reilly, Beijing, Sebastopol, CA, 2022.

ISBN 978-1-0981-1913-3.

[Sturmfels et al.(2020)Sturmfels, Lundberg, and Lee]

Pascal Sturmfels, Scott Lundberg, and Su-In Lee.

Visualizing the Impact of Feature Attribution

Baselines. Distill, 5(1):e22, January 2020. ISSN

2476-0757. doi: 10.23915/distill.00022.

[Sundararajan et al.(2017)Sundararajan, Taly, and Yan]

Mukund Sundararajan, Ankur Taly, and Qiqi Yan.

Axiomatic Attribution for Deep Networks, June

2017.

[Molnar(2019)]

Christoph Molnar. Interpretable

Machine Learning: A Guide for Making Black

Box Models Interpretable. Lulu, Morisville, North

Carolina, 2019. ISBN 978-0-244-76852-2.

[Lundberg and Lee()]

Scott M Lundberg and Su-In

Lee. A Uniﬁed Approach to Interpreting Model

Predictions.

[Mumford(2012)]

Stephen Mumford. Metaphysics: A

Very Short Introduction. Number 326 in Very Short

Introductions. Oxford University Press, Oxford,

United Kingdom, ﬁrst edition edition, 2012. ISBN

978-0-19-965712-4.

[Menzies and Beebee(2025)]

Peter Menzies and Helen

Beebee. Counterfactual Theories of Causation. In

Edward N. Zalta and Uri Nodelman, editors, The

Stanford Encyclopedia of Philosophy. Metaphysics

Research Lab, Stanford University, winter 2025

edition, 2025.

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 classiﬁcations

of galaxies from optical images surveys and radio

galaxy source extraction from radio observations.

Part III

Biosciences

DNA Damage Assessment at the Single-Cell

Level Using the Alkaline Comet Assay

by Aengela Grace Jacob

airis4D, Vol.4, No.1, 2026

www.airis4d.com

1.1 Introduction

The cell stands as the fundamental unit of

life, a microscopic powerhouse that orchestrates

the intricate mechanisms sustaining living systems.

This foundational ideology highlights how metabolic

changes within cells propagate outward, aﬀecting

organismal health, disease progression, and

environmental resilience. Disruptions at the

cellular level, particularly DNA damage from genotoxic

agents like radiation, chemicals, or oxidative stress, can

cascade into systemic failures, manifesting as cancer,

infertility, or accelerated ageing. Understanding these

processes demands precise tools to quantify damage,

repair kinetics, and genotoxic potential, bridging

molecular biology with public health implications.

Central to this evaluation is the Alkaline Comet

Assay, also known as single-cell gel electrophoresis

(SCGE), a versatile and highly sensitive technique

for assessing DNA integrity at the individual cell

level. Developed in the 1980s and reﬁned over

decades, it detects single- and double-strand breaks,

alkali-labile sites, and incomplete repair intermediates

by exploiting the principle of DNA migration under

alkaline conditions. Cells are embedded in agarose on

microscope slides, lysed to expose nucleoids, subjected

to high-pH electrophoresis (pH

13), and stained for

ﬂuorescence microscopy. Damaged DNA fragments

migrate from the nuclear head to form a comet-like

tail, whose morphology directly correlates with lesion

severity.

This low-cost, rapid method (yielding results

in hours) excels in biomonitoring occupational

exposures, predicting cancer radiosensitivity, evaluating

chemotherapy eﬃcacy, and screening for environmental

pollutants. Quantitative metrics, such as % tail DNA,

tail length, and Olive tail moment, provide robust,

statistically analyzable data, often surpassing traditional

assays in sensitivity. The Olive tail moment (OTM)

provides robust, statistically analyzable data in the

Alkaline Comet Assay by integrating both the amount

of DNA in the tail (% tail DNA) and its migration

distance (tail centre of mass relative to the head). By

illuminating DNA’s vulnerability, the Alkaline Comet

Assay enables researchers to safeguard the cellular

foundation of life against invisible threats.

1.2 Sample preparation and analysis

of data

Preparation of cells

The analysis is conducted on harvested whole

blood cells, divided into two sets: one is UV-treated,

and the other is UV-un-treated. These are suspended

in a physiological buﬀer, such as PBS, and the culture

is embedded in low-melting agarose. The agarose gel

is used in proper proportions to avoid gel thickening,

and a thin layer of it is added to a microscopic slide.

This is done to immobilise the cells and protect their

morphology during the subsequent steps.

Cell Lysis

1.2 Sample preparation and analysis of data

Cell lysis is performed to access intracellular

materials for various scientiﬁc and industrial

applications. A primary objective is to obtain speciﬁc

components, such as DNA, RNA, proteins, or organelles,

for detailed study. Researchers might extract DNA

for genomic sequencing, RNA for gene expression

analysis, or proteins to understand their functions

and interactions. In molecular biology research, it

enables the isolation of genetic material for cloning,

sequencing, and genetic manipulation. For instance, in

drug discovery, lysing cells can help identify potential

drug targets by analysing speciﬁc proteins or cellular

pathways involved in diseases.

Therefore, to induce cell lysis, the slides coated

with agarose are submerged in chilled detergent lysis

buﬀer (containing Triton X-100 and SDS), which

removes cellular membranes and proteins, leaving the

DNA in a nucleoid structure.

Lysis buﬀer contains Proteinase K, which digests

proteins, allowing the separation of DNA from proteins,

and RNase A, which degrades RNA. It also includes salt

solutions, such as sodium chloride and sodium acetate,

for DNA precipitation, as well as buﬀer solutions like

Tris-EDTA for storing and resuspending DNA.

Electrophoresis

The slide is placed in an electrophoresis buﬀer

under an electrical ﬁeld, causing the fragmented DNA

to migrate away from the nucleus toward the anode.

The rate and extent of DNA migration indicate the level

of DNA damage: more damaged DNA migrates further,

resembling a comet tail. An electrophoresis tank is

used to allow DNA migration under an electric ﬁeld of

25-30V and 300-400mA for 20 minutes.

Neutralisation and staining

After electrophoresis, the slides are neutralised

with a buﬀer to stop the DNA unwinding. Now we have

to stain the DNA using dyes like Ethidium Bromide

(carcinogenic, so it is less commonly used); instead,

silver staining is preferred since it is more sensitive to

a large number of DNA lesions, and it also ensures a

low background noise. The primary disadvantage of

silver staining is its cost. Here, in our experimental

study, we use SYBR GREEN, a ﬂuorescent dye which is

visualised under a ﬂuorescent microscope with the help

Figure 1: GEL RUN Stained with SYBR GREEN

]tiny Image courtesy: https://www.thermoﬁscher.com

/order/catalog/product/S7563

of MagVision software. The fragments were examined

to quantify comet parameters, including tail length,

intensity, and tail moment.

Interpretation of data

Alkaline Single Cell Gel Electrophoresis (SCGE)

is a standard comet assay for detecting single- and

double-stranded breaks, alkali-labile sites, and other

forms of DNA damage induced by genotoxic agents like

chemicals or radiations or other physiological conditions

like oxidative stress; the body’s defense against free

radicals is provided by antioxidants, but when free

radicals outnumber antioxidants, oxidative stress is

triggered leading to damage in DNA. The comet image

was captured using Magvision Software. The head

region represents DNA that migrates outside of the

tail. The tail region represents the DNA migrating

out of the nucleus due to fragmentation and loss of

structure. The more the migration, the more the

damage. The primary purpose of the comet assay

is to detect these DNA damages and evaluate the DNA

repair mechanisms that can be employed. It can also

be used to assess DNA damage in sperm cells for male

1.2 Sample preparation and analysis of data

(a) Total tail analysis

(b) Comet tail analysis

Figure 2: (a) Total comet primary cellular analysis

object masks. Object masks are automatically placed

around comets from all control cell populations. (b)

Comet tail analysis.

infertility studies and for biomonitoring DNA damage

in patient samples to assess cancer risk. It is a strenuous

process that requires valuable time and eﬀort to assess

small quantities of DNA; hence, it’s not widely carried

out in all laboratories. The new advanced versions

utilise comet chips, speciﬁcally HTP Comet assay

systems (COMPAC-50), which can combine multiple

slide processing and analysis steps for higher throughput

and also use of automated microscopy systems like

Metafer for high content screening of large gel formats

Here, sample images from the study, conducted on

UV-treated and UV-untreated blood, are shown. UV-

treated blood was observed to have denser, longer tails

compared to the untreated blood.

Figure 3: UV-treated blood sample stained by SYBR

GREEN.

Figure 4: UV-untreated blood sample stained by SYBR

GREEN

1.2 Sample preparation and analysis of data

References

What Is Cell Lysing and Why Is It Important? -

Biology Insights

https://www.thermoﬁscher.com/order/catalog/p

roduct/S7563

www.bio-equip.cn/enshow1equip.asp?equipid=44603

https://pmc.ncbi.nlm.nih.gov/articles/PMC84136

25/

https://www.agilent.com/cs/library/application

s/automated-comet-assay-genotoxicity-5994-2597E

N-agilent.pdf

About the Author

Aengela Grace Jacob is Final Year Student

of Bsc Biotechnology , Chemistry ( BSc BtC ) Dual

Major at Christ University Central campus, Bangalore

Diabetic Retinopathy- Medical Image

Processing

by Geetha Paul

airis4D, Vol.4, No.1, 2026

www.airis4d.com

2.1 Introduction

Diabetic retinopathy (DR) is a progressive eye

disease and a leading cause of vision impairment and

blindness among individuals with diabetes mellitus.

This microvascular complication primarily aﬀects the

retina, the light-sensitive tissue lining the back of

the eye, by damaging its small blood vessels due to

prolonged exposure to high blood glucose levels. DR

can develop in both type 1 and type 2 diabetes and

typically advances over the years, often without initial

symptoms. As diabetes prevalence grows globally,

diabetic retinopathy has become a major public health

concern, particularly impacting adults of working age.

The primary risk factors include the duration of diabetes,

poor glycemic control, hypertension, and dyslipidemia.

Early identiﬁcation and management, through regular

eye examinations and optimised diabetic care, are

essential to reduce the burden of visual disability caused

by this disease.

The pathogenesis of diabetic retinopathy involves

continuous injury to the retinal microvasculature.

Chronic high blood sugar levels induce metabolic

and biochemical changes, resulting in thickening of

the capillary basement membrane, loss of pericytes,

and the formation of microaneurysms. As the disease

progresses, vascular permeability increases, resulting

in ﬂuid leakage, retinal swelling (macular oedema),

and eventually ischemia due to capillary non-perfusion.

In advanced stages, marked by proliferative diabetic

retinopathy, abnormal new blood vessels proliferate

on the retinal surface in response to ischemia. These

vessels are fragile, prone to bleeding, and can precipitate

complications such as vitreous haemorrhage or retinal

detachment, ultimately risking permanent vision loss if

untreated.

Clinically, DR is classiﬁed into non-proliferative

(NPDR) and proliferative diabetic retinopathy (PDR)

stages. NPDR is characterised by microaneurysms,

haemorrhages, hard exudates, and sometimes

macular oedema, while PDR is distinguished by

neovascularisation. The onset and progression of DR

may be insidious; therefore, annual comprehensive

dilated eye examinations are recommended for all

diabetic patients. Treatment options include optimising

systemic risk factors, intravitreal pharmacotherapy (e.g.,

anti-VEGF agents), laser photocoagulation, and surgical

intervention for severe cases. Despite advances in

therapy, prevention remains the cornerstone of reducing

the impact of diabetic retinopathy by achieving strict

glycemic, blood pressure, and lipid control.

2. Non-proliferative diabetic retinopathy

Non-proliferative diabetic retinopathy (NPDR) is

the early stage of the disease in which symptoms will be

mild or nonexistent. In NPDR, the blood vessels in the

retina are weakened. Tiny bulges in the blood vessels,

called microaneurysms, may leak ﬂuid into the retina.

This leakage may lead to swelling of the macula.

3. Proliferative diabetic retinopathy

Proliferative diabetic retinopathy (PDR) is the

more advanced form of the disease. At this stage,

2.2 Drusen

Figure 1: High quality to low quality of DR Retinal

images,

Image courtsey:https://www.nature.com/articles/s41598-021-93632- 8?fromPaywallRec=false

Figure 2: (A) Illustration of the parts of the eye, (B)

Retina of the eye, (C) Drusen are yellow deposits under

the retina, (D) CNV (Choroidal Neovascularisation)

is abnormal blood vessel growth beneath the retina,

causing leakage; and CME (Cystoid Macular Oedema):

is ﬂuid accumulation causing retinal swelling.

Image courtesy: https://www.nature.com/articles/s41467-021-23458- 5

circulation problems deprive the retina of its oxygen

supply. As a result, new, fragile blood vessels can

begin to grow in the retina and into the vitreous, the

gel-like ﬂuid that ﬁlls the back of the eye. The new

blood vessels may leak blood into the vitreous, causing

vision to become cloudy.

Diabetic retinopathy is a leading cause of blindness

resulting from damage to retinal blood vessels caused

by diabetes. Image processing plays a central role in

detecting, grading, and monitoring diabetic retinopathy

using retinal fundus photographs.

Figure 3: Illustration of Drusen and its formation

beneath retina.

Image courtsey: https://www.brightfocus.org/resource/why-is-my-doctor-always-talking-about-dru

sen/

2.6 Microaneurysms

2.2 Drusen

These are small yellowish deposits of cellular

debris, proteins, and lipids that accumulate between

the retina and the underlying layer called Bruch’s

membrane. Drusen are a hallmark feature, especially in

age-related macular degeneration, but can also be seen

in other retinal conditions. They appear as small spots

beneath the retina and indicate waste accumulation

resulting from dysfunction of the retinal pigment

epithelium.

2.3 CNV (Choroidal

Neovascularization):

This is the abnormal growth of new blood

vessels from the choroid layer beneath the retina into

the subretinal space. CNV is often associated with

advanced forms of macular degeneration and can cause

leakage and bleeding, leading to vision loss. It is

a serious complication that can be seen in diabetic

retinopathy as well.

2.4 CME (Cystoid Macular Edema):

This refers to swelling in the macula (central

retina) caused by ﬂuid accumulation in cyst-like spaces.

CME is a common consequence of diabetic retinopathy

that results from vascular leakage and inﬂammation. It

leads to central vision impairment and is a major cause

of vision loss in DR.

2.5 Exudates

Exudates in the context of diabetic retinopathy

are deposits composed primarily of lipids (fats) and

proteinaceous material that leak out from damaged

blood vessels in the retina. These deposits accumulate

in the outer layers of the retina and appear as yellow or

white patches on retinal images.

Figure 4: Exudates in Dibetic Retinopathic eye.

Image courtsey: https://pmc.ncbi.nlm.nih.gov/articles/PMC4568614/

Figure 5: Microaneurysms in Diabetic Retinopathic

eye

Image courtsey: https://onlinelibrary.wiley.com/doi/10.1155/2023/1305583

They are caused by the breakdown of the blood-

retina barrier due to damaged retinal capillaries,

allowing serum proteins and lipids to escape and settle

in the retinal tissue. Exudates often indicate leakage

and swelling in the retina, and their presence, especially

near the macula (central retina), can lead to signiﬁcant

visual impairment.

2.6 Microaneurysms

Microaneurysms are tiny bulges or swellings in

the walls of the small blood vessels (capillaries) of the

retina, occurring as an early sign of diabetic retinopathy.

They form when high blood sugar levels caused by

diabetes weaken the retinal blood vessels, causing them

to balloon out locally. These microaneurysms can leak

ﬂuid or blood into the retina, potentially leading to

vision problems.

2.7 Haemorrhages

Haemorrhages in diabetic retinopathy refer to

bleeding that occurs when the fragile blood vessels

in the retina break and leak blood into the retinal tissue.

They can appear in various forms depending on their

location and size:

Dot and blot haemorrhages: Small, round

haemorrhages found in the deeper layers of the retina,

typically resulting from the rupture of microaneurysms.

Flame-shaped hemorrhages: Occur in the

2.11 Treatment and Management

Figure 6: Haemorrhages in Diabetic Retinopathic eye.

Image courtsey: https://www.goodeyes.com/diabetic-retinopathy/

superﬁcial layers of the retina, resembling ﬂame shapes

along the retinal nerve ﬁber layer

Haemorrhages are a sign of worsening diabetic

retinal damage and are associated with increased

vascular permeability and vessel wall weakening. In

severe diabetic retinopathy, especially proliferative

diabetic retinopathy, new fragile blood vessels may form

and bleed into the vitreous humour, causing vitreous

haemorrhage, which can lead to vision loss. The

presence and extent of haemorrhages are important

clinical indicators used in diagnosing and staging

diabetic retinopathy.

2.8 Causes and Progression

High blood sugar from diabetes damages retinal

blood vessels, causing increased permeability and blood

vessel loss, which leads to retinal ischemia and swelling.

The retina reacts by growing abnormal new vessels that

are fragile and prone to bleeding. Vision loss occurs

particularly when macular edema develops or in the

advanced proliferative stage with retinal detachment

risks.

2.9 Symptoms

Common symptoms include blurry or cloudy

vision, ﬂoaters, dark spots or areas in the vision, and

eventual vision loss. Early stages may be asymptomatic,

making regular screening important for people with

diabetes.

2.10 Diagnosis

Diagnosis involves dilated eye exams using

ophthalmoscopy or retinal imaging to detect

microaneurysms, haemorrhages, exudates, and new

vessel growth. Regular screening is crucial since early

diabetic retinopathy can be managed before signiﬁcant

vision loss.

2.11 Treatment and Management

Treatment aims to manage diabetes (maintaining

blood sugar, blood pressure, and cholesterol control)

and directly address eye damage.

Treatments include Anti-VEGF injections to

reduce abnormal blood vessel growth and macular

swelling, Steroid injections to reduce inﬂammation

and swelling, Laser photocoagulation to seal leaking

vessels and prevent bleeding, and Vitrectomy surgery

for severe cases with vitreous haemorrhage or retinal

detachment.

Strict glycemic control and routine eye exams are

essential to prevent progression. Early detection and

intervention can prevent 90% of severe vision loss cases.

Overall, diabetic retinopathy requires a combination

of systemic disease management and targeted ocular

treatments for best outcomes

2.12 Role of Image Processing in

Diabetic Retinopathy

Retinal imaging techniques, such as fundus

photography and optical coherence tomography (OCT),

provide high-resolution images of the retina.

Image processing enhances these images through noise

reduction, contrast adjustments, and colour channel

extraction to highlight key retinal structures.

Segmentation algorithms isolate anatomical features

like blood vessels, the optic disc and detect lesions

(microaneurysms, haemorrhages, exudates), which are

signs of DR.

2.13 Automated Detection and

Classiﬁcation

Machine learning and deep learning models,

especially convolutional neural networks, analyse

processed images to automatically detect and classify

2.15 DR grading pipeline end-to-end

Figure 7: Illustration of the steps in DR -Image

processing

Image courtsey: https://www.nature.com/articles/nrdp201612

DR stages.

These systems recognise subtle changes in retinal

features associated with early and advanced DR that

might be missed in manual examination.

Automated grading and early detection improve

screening eﬃciency, allowing timely intervention to

prevent severe vision loss.

2.14 Steps in Diabetic Retinopathy

Image Processing

Preprocessing: Involves green channel extraction

for higher contrast, histogram equalisation for

brightness normalisation, resizing images, and

denoising to enhance image clarity.

Feature Extraction: Quantiﬁes biological

features such as exudates, microaneurysms,

haemorrhages, blood vessel area, and bifurcation points.

These features are critical markers of retinopathy and

its severity.

Figure 8: The local fundus image dataset was randomly

divided into training and validation datasets. Seventy

per cent of the total images in the dataset were used

for training the image quality assessment sub-network.

The lesion detection sub-network was trained using

gradable images with annotations of retinal lesions.

Then, the total gradable images in the training set

were used to train the DR grading sub-network. All

images in the local validation dataset were used to test

the image quality sub-network. Finally, the gradable

images labelled with retinal lesions were used to test the

lesion detection sub-network. DR, diabetic retinopathy.

Image courtesy:https://www.nature.com/articles/s41467-021- 23458-5

Segmentation: Localises regions of interest (optic

disc, fovea, lesions) using thresholding and edge

detection algorithms.

Classiﬁcation: Machine learning and deep

learning models (like CNNs, SVMs, and RSG-Net)

automatically detect and classify stages of diabetic

retinopathy by analysing extracted features.

Grading: Automated systems grade images into

retinopathy stages (none, mild, moderate, severe,

proliferative) to guide clinical decision-making.

2.15 DR grading pipeline end-to-end

Data Collection and Preparation

Acquire a large annotated dataset of retinal fundus

images graded by DR severity levels (e.g., from public

datasets like Kaggle’s EyePACS).

Preprocess images by resizing (commonly to 224x224

pixels), normalisation, noise reduction, and enhancing

contrast.

Utilise data augmentation techniques (such as rotation,

ﬂipping, and zooming) to enhance training data

diversity and mitigate overﬁtting.

Model Architecture and Training

Choose a robust convolutional neural network

2.15 DR grading pipeline end-to-end

(CNN) architecture such as ResNet-50 or a specialized

model like RSG-Net for feature extraction.

Apply transfer learning by ﬁne-tuning a pretrained

backbone on DR data to leverage prior knowledge.

Train the model for multiclass classiﬁcation

corresponding to DR grading stages (no DR, mild,

moderate, severe, proliferative).

Use suitable loss functions (categorical cross-entropy

for multiclass) and optimisers like Adam or SGD with

tuned learning rates.

Implement regularisation techniques like dropout and

batch normalisation to prevent overﬁtting.

Monitor validation metrics to apply early stopping or

learning rate adjustments during training.

Model Evaluation and Validation

Evaluate performance using accuracy, AUC score,

sensitivity, speciﬁcity, and confusion matrices.

Validate the model on separate test data to ensure

generalisation.

Utilise robust metrics to evaluate misclassiﬁcations

between adjacent DR grades and mitigate class

imbalance.

Pipeline Automation and Deployment

Build a workﬂow to handle data input,

preprocessing, prediction, and grading output.

Optionally integrate a user interface for uploading

images and displaying results.

Containerise the model using Docker and deploy on

cloud platforms for scalability.

Implement continuous monitoring and retraining based

on new data.

This pipeline encapsulates automated DR detection with

high accuracy and practical usability, enabling timely

diagnosis and referral for diabetic retinopathy patients.

References

https://www.nature.com/articles/s41467-021-2

3458-5

https://eyewiki.org/Age-Related Macular Degen

eration

https://www.sciencedirect.com/org/science/articl

e/pii/S1546221824007951

https://pmc.ncbi.nlm.nih.gov/articles/PMC10411

652/

https://www.health.harvard.edu/newsletter artic

https://biomedpharmajournal.org/vol10no2/dia

betic-retinal-fundus-images-preprocessing-and-featu

re-extraction-for-early-detection-of-diabetic-retinop

athy/

About the Author

Geetha Paul is one of the directors of

airis4D. She leads the Biosciences Division. Her

research interests extends from Cell & Molecular

Biology to Environmental Sciences, Odonatology, and

Aquatic Biology.

Neurological Disorders: A Brief Overview

by Neelima Dubey

airis4D, Vol.4, No.1, 2026

www.airis4d.com

3.1 Introduction

According to the World Health Organization

(WHO), neurological disorders encompass all diseases

that aﬀect the nervous system and its components

entirely or partially. These disorders may involve

neurons or neural tracts of the central nervous

system (CNS) including the spinal cord and the

whole brain or its individual components such as the

cerebrum (cerebral cortex), basal ganglia, diencephalon,

brainstem (midbrain, pons, and medulla oblongata),

and cerebellum. In addition, conditions aﬀecting the

peripheral nervous system (PNS) include disorders of

the cranial nerves and their nuclei, spinal nerve roots

and plexuses, peripheral nerves, the autonomic nervous

system, the neuromuscular junction, and skeletal

muscles. These abnormalities can be electrochemical,

biochemical or structural which ranges from sensory

disturbance, diﬃculty in coordination, confusion,

seizures, muscular atrophy, and memory disturbance.

There are numerous etiologies attributed to the

advent of various neurologic ailments. According to the

WHO, almost one billion people globally suﬀer from

neurologic disorders; this is set to rise in the next few

years, making it one of the most important challenges

in public health. All age groups and geographical areas

are aﬀected by these illnesses. Neurological disorders

are broadly classiﬁed into neurodegenerative diseases

(such as Alzheimer’s disease, Parkinson’s disease,

Huntington’s disease, and amyotrophic lateral sclerosis)

and neuropsychiatric diseases (such as depression,

schizophrenia, bipolar aﬀective disorders, autism, mood

disorders, attention-deﬁcit or hyperactivity disorder, and

tardive dyskinesia).

3.2 Neurodegenerative Diseases

Neurodegenerative diseases are a group of

disorders that gradually destroys the structure and

function of neurons in central and peripheral nervous

system. These usually develop slowly, and their

eﬀect and symptoms tend to appear in the later life.

Neurodegenerative diseases are mostly associated with

aging, but there are certain types that can develop

in childhood or early adulthood. These diseases are

chronic, progressive and debilitating, posing signiﬁcant

challenges for both aﬀected individuals and healthcare

systems. Understanding the causes, symptoms and

mechanism behind neurodegenerative diseases is crucial

for developing eﬀective treatments and interventions.

Unlike some diseases that involve temporary

or reversible damage, neurodegenerative diseases

usually cause irreversible damage and progressively

worsen over time. Symptoms of these diseases vary

signiﬁcantly depending on the type of the area of brain

aﬀected, individual’s age, and the stage of the diseases.

Common categories of symptoms include memory loss,

confusion and disorientation, sleeping disturbances,

apathy, agitation, anxiety, ataxia, tremors or shaking,

depression, and personality shifts. On the other hand,

the exact causes of neurodegenerative diseases are

unknown, but researchers have identiﬁed a combination

of genetic, environmental, and age-related factors that

contribute to these diseases.

The most common neurodegenerative diseases

include Alzheimer’s disease, Parkinson’s disease,

3.5 Huntington’s disease

Huntington disease, Amyotrophic Lateral Sclerosis,

and Multiple Sclerosis.

3.3 Alzheimer’s disease

Alzheimer’s disease was named after the German

Psychiatrist and neuroanatomist Alois Alzheimer in

1906. It is the most common form of dementia and

primarily aﬀects memory and cognitive abilities. It

is typically seen in patients over the age of 65, but

about 5-10% of cases are considered “early-onset” with

symptoms appearing before 65. It is characterized by

the accumulation of abnormal proteins such as amyloid

plaques outside neurons and tau tangles inside neurons,

interfering with neuronal functions and causes cell

death.

Currently, there is no cure for Alzheimer’s

disease, although there are available treatments that

just improves the symptoms, such as cholinesterase

inhibitors, and memantine (glutamate regulator).

Donanemab is a recently FDA-approved treatment for

earlier onset of Alzheimer’s disease.

3.4 Parkinson’s disease

Parkinson’ s disease is a progressive movement

disorder. It is the second most prevalent

neurodegenerative disease. Although, its incidence

increases with age for above 70 years old, 5-10% of

cases are reported to be early onset. It was medically

described as “shaking palsy” by British physicist James

Parkinson in 1817. Later, William Rutherford Sanders

named the condition after him in 1865.

Several known risk factors contribute to the

development of Parkinson’s disease, such as age,

gender, ethnicity, rapid eye movement, sleep disorders,

traumatic brain injury, high consumption of dairy

products, genetics, and pesticides or herbicides.

However, the deﬁciency of dopamine in the substantia

nigra of brain stem is recognized to be the main

pathology. Parkinson’s disease largely aﬀects the motor

system. Common symptoms include tremors, stiﬀness,

and bradykinesia (slowness of movement). Non-

motor signs such as dementia (also called Parkinson’s

disease dementia) and cognitive decline are also noticed.

Parkinson’s disease dementia is believed to cause by

the accumulation of abnormal protein alpha-synuclein.

Cognitive decline can also result from the buildup of

this abnormal protein, causing neuronal malfunction

and cell death. It also shares common features with

Alzheimer’s disease.

Medications like levodopa and dopamine agonist

mostly helps in alleviating motor symptoms. Surgical

interventions such as deep brain stimulation may also

be helpful in some cases.

3.5 Huntington’s disease

Huntington’s disease is an autosomal dominant

disorder characterized by movement disorder and

cognitive decline. It was discovered by George

Huntington in 1872, and its genetic cause was later

discovered by Nancy Wexler in 1993. Movement defects

include chorea and loss of coordination. Psychiatric

disorders such as depression, psychosis, obsessive

compulsive disorder are common in the patient with

Huntington’s disease. This disease is caused by

abnormal expansion of CAG repeats in HTT gene

on chromosome 4, which results in the production

of an abnormal protein called huntingtin. This protein

gets aggregated and accumulates in the brain region,

especially in basal ganglia which is critical for motor

control, and cognition.

No treatments can cure the course of Huntington’s

disease, but medications such as tetrabenazine,

citalopram, and Haloperidol may be eﬀective in

lessening the symptoms of this disease. Speech therapy

and counselling also hold promising in slowing the

progression of this disease.

3.6 Amyotrophic Lateral Sclerosis

Amyotrophic Lateral Sclerosis was originally

deﬁned as motor neuron disease by Jean-Martin

Charcot in 1869. Now it is understood as multisystem

neurodegenerative disease. It generally causes

degeneration of upper and lower motor neuron, leading

to progressive muscle weakness and paralysis. It is

3.7 Multiple Sclerosis

categorized into two forms- sporadic (90-95%) which

has no genetically inherited component, and familial

(5-10%) which is a genetic dominant inheritance factor.

The drugs such as Riluzole and Edaravone focuses

on slowing progression, symptoms of the disease and

towards enhancing the quality of life.

3.7 Multiple Sclerosis

Multiple Sclerosis is a chronic autoimmune

disease primarily aﬀecting the brain, spinal cord,

and optic nerves. It shares common features with

other neurodegenerative diseases such as progressive

neuronal loss, axonal damage, and brain atrophy

contributing to long-term disability. Symptoms of

this disease generally include muscle weakness, visual

impairment, fatigue, and impaired coordination.

There is currently no cure for Multiple Sclerosis,

but Disease-Modifying Therapies (DMT), and cognitive

therapies may be helpful in slowing the disease

progression.

3.7.1 Neuropsychiatry

In contrast to neurodegenerative diseases like

Alzheimer’s, which involve progressive loss of neurons,

neuropsychiatry covers mental symptoms that result

from neurological malfunctioning. Studies highlight

its emphasis on brain-behaviour interfaces in diseases

like epilepsy, traumatic brain injury, and movement

disorders, combining psychiatry and neurology

for comprehensive care. With advancements in

neuroimaging (MRI), this discipline has demonstrated

similar pathophysiology in behavioural, emotional, and

cognitive symptoms across neural substrates.

Antiepileptics like valproate for mood stabilization

in epilepsy, antipsychotics like quetiapine for psychosis

in Parkinson’s disease, and benzodiazepines for acute

agitation in traumatic brain injury are just a few

examples of the multimodal approaches used in

neuropsychiatry today to treat underlying neurological

perturbation. Psychotherapy is tailored to manage

cognitive deﬁciencies, deep brain stimulation for drug

resistant patients, and cognitive rehabilitation are

examples of non-pharmacological therapies.

3.7.2 Mood disorders

Mood disorders are a group of mental illnesses

that have substantial eﬀects on day-to-day functioning

and neurocognitive functions. They are deﬁned by

disruptions in emotional regulation, which typically

demonstrate as sadness, mania, or hypomania. Their

variability is highlighted by core symptoms like self-

blame, worthlessness, anhedonia, and altered approach

or withdrawal behaviours, which are frequently

connected to anterior temporal and subgenual brain

networks.

Persistent sadness, anger, emotional lability,

exhaustion, psychomotor changes, and cognitive

diﬃculties like indecision are among the primary

symptoms of mood disorders. These symptoms

frequently coexist with physiological problems such

disturbed sleep, hunger, and libido. Treatment

methods focus on a multimodal approach, incorporating

psychotherapy like dialectical behaviour therapy

(DBT) and cognitive behavioural therapy (CBT),

pharmacotherapy like antidepressants and mood

stabilizers, lifestyle changes like exercise, sleep hygiene,

and nutritional support, and electroconvulsive therapy

(ECT) for refractory cases. Mood disorders can

be broadly classiﬁed as either bipolar disorder or

depression, according to the Diagnostic and Statistical

Manual of Mental Disorders, Fifth Edition (DSM-5).

3.7.3 Major Depressive Disorder

According to the DSM-5 criteria, major depressive

disorder is a severe form of unipolar depression

characterized by persistent low mood or anhedonia,

along with symptoms like guilt, low energy,

concentration problems, changes in appetite, sleep

disturbances, and suicidal thoughts. The diagnosis

requires at least ﬁve symptoms. Its signiﬁcant frequency

and treatment resistance in up to 60% of cases are

highlighted in recent pharmacologic reviews, prompting

research into new treatments that address underlying

biology in addition to conventional antidepressants.

The prognosis includes a higher risk of suicide in two-

3.9 Premenstrual Dysphoric Disorder

thirds of patients and chronic recurrence with untreated

episodes lasting six to twelve months.

Signiﬁcant weight gain or loss, recurring

thoughts of suicide, and feelings of emptiness

lasting more than two weeks are all signs of

major depressive disorder (MDD). Rapid-acting

medications such as ketamine for patients who

are resistant to treatment, esmethadone that targets

opioid pathways, and dextromethorphan-bupropion

combinations that produce 39–46% remission rates

better than monotherapy are examples of emerging

treatments for the condition.

3.8 Bipolar Disorders

Bipolar disorders are characterized by cyclic mood

swings between mania or hypomania and depression.

They show more functional impairment compared

to unipolar depression. Four basic mood states are

identiﬁed by analyses: depressed, anxious, irritated,

and euphoric. Blame or praise biases and variations in

self-worth are identiﬁed as an important mechanisms

that inform neurocognitive targets such as subgenual

networks.

Elevated energy during manic instances, rapid

thoughts, impulsivity, distractibility, and depressive

lethargy with diﬃculty concentrating throughout

episodes are common symptoms. Atypical

antipsychotics like quetiapine, mood stabilizers

like lithium or lamotrigine, and supplementary

psychotherapy like interpersonal and social rhythm

therapy are used in management to control daily

activities and stressors.

3.8.1 Bipolar Disorder – I

At least one manic episode, frequently

accompanied by depressive phases that cause signiﬁcant

disability and even psychosis, is a hallmark of

bipolar I disorder. The major symptoms and the

severity of the disorder is associated with a ﬂawed

or defective self-structure. However, these patients

show variable degrees of personality organization, with

less severe identity and aggressiveness deﬁcits than

those associated with other disorders such as borderline

personality disorder (BPD). BP-I has more overall mood

instability but less time spent in depressive states than

other form of Bipolar Disorder such as BP-II.

During mania, BP-I symptoms include

talkativeness, increased engagement in risky

activities, grandiosity, and a decreased desire for

sleep, often accompanied by psychotic characteristics.

The main therapeutics for BPD-I include Lithium for

manic prophylaxis, valproate for acute stabilization. In

addition, family-focused therapy is provided to improve

adherence and episode detection.

3.8.2 Bipolar Disorder – II

Recurrent depressive episodes and hypomania

without complete mania are characteristics of bipolar

II illness. This leads to greater chronicity, frequent

episodes, and extended depression phases. Up to 40%

BP-II patients spend more time depressed than observed

in BP-I. Due to the predominance of rapid episodes

and prominent depressive symptoms, it is frequently

misinterpreted as unipolar depression and carries an

increased risk of suicide during mixed states. Patients

report higher rates of relapse, decreased mood, and

increased symptom variability in BP-II.

Increased goal-directed activity, inﬂated self-

esteem, and ﬂight of ideas without signiﬁcant

impairment or inpatient requirements are examples

of hypomanic symptoms in BP-II. Eﬀective treatments

include lurasidone or lumateperone for acute bipolar

depression, lamotrigine for depression prevention. The

psychoeducation to the common people and to the family

members of the patients is given to reduce misdiagnosis

and promote long-term adherence.

3.9

Premenstrual Dysphoric Disorder

Premenstrual dysphoric disorder (PMDD),

according to the DSM-5 criteria in prospective

monitoring, is a severe version of premenstrual

syndrome that aﬀects women of reproductive age

with luteal-phase onset of severe emotional symptoms

such as anxiety, depression, irritability, and functional

3.11 Seasonal Aﬀective Disorder

impairment. It is underdiagnosed and thus remains

unidentiﬁed in the society particularly in Indian

society thus making day-to-day living of women with

PMDD more diﬃcult. PMDD mandates cycle-speciﬁc

treatments as the pathology is limited to the luteal phase

of the reproductive cycle but impacting the life of young

women throughout.

Severe mood swings, stress, enduring wrath,

feelings of overwhelm, and physical complaints like

breast tenderness or luteal phase joint pain are

common symptoms of PMDD. Continuous-dosing oral

contraceptives are used to suppress ovulation, luteal-

phase SSRIs are used for quick symptom relief, CBT is

used to reframe emotional patterns, and GnRH agonists

are used for resistant instances. However, these cannot

be the permanent cure for PMDD and could adversely

aﬀect the reproductive health of the women with PMDD.

3.10 Persistent Depressive Disorder

Formerly known as dysthymia or chronic major

depression, persistent depressive disorder (PDD) is

characterized by depressed mood on most days for

at least two years (one year in adolescents), with

symptoms like hopelessness and poor self-esteem that

never completely go away. The DSM-5 combines

chronic forms with severe comorbidity, disability, and

suicide risk, emphasizing duration over severity. Double

depression arises when large episodes overlap. It reacts

to antidepressants such as SSRIs or MAOIs, and CBT

is frequently used to help it go into remission.

Poor focus, indecision, excessive guilt, social

disengagement, and persistent low-grade hopelessness

without complete inter-episode recovery are some of

the symptoms of PDD.

Novel mood stabilizers like lamotrigine are used

in conjunction with SSRIs or MAOIs in therapeutic

approaches, which have been shown to provide better

remission when partnered with psychotherapy for longer

than two years.

3.11 Seasonal Aﬀective Disorder

Seasonal aﬀective disorder is characterized by

recurring signiﬁcant depression that coincides with

seasonal light variations. It is primarily winter-type

and is characterized by hypersomnia, hyperphagia, and

low energy as a result of less daylight. Although it

has received less attention in evaluations from 2024–

2025, it maintains the fundamental characteristics of

MDD but exhibits circadian patterns that worsen under

stressful conditions such environmental disruptions. In

order to prevent relapses, interventions combine regular

antidepressants with light therapy.

Carbohydrate cravings that result in weight gain,

oversleeping, daytime tiredness, and an increased

appetite for carbohydrates during shorter daylight hours

are all signs of SAD. First-line treatments include

bright light therapy, which reduces mild-to-moderate

depression scores when compared to a placebo. CBT

and bupropion are also used to prevent seasonal relapses.

3.12 Postpartum Depression

The symptoms of postpartum depression (PPD)

are similar to Major Depressive Disorder (MDD), such

as persistent sorrow, anxiety, and diﬃculties connecting

with the newborn. However, maternal-infant dynamics

are severely impacted by the occurrence of PPD in

new mother. PPD can develop weeks to months

after parturition. The diagnosis is often missed due

to the common occurrence of mood swings during

the ﬁrst week after giving birth called postpartum

blues. Therefore, screening of the new mothers using

instruments like EPDS identiﬁes patients at the early

stages.

The speciﬁc symptoms of PPD include excessive

concern, panic attacks, hostility toward the infant, and

feelings of separation that arise due to the hormonal

changes in the body of the mother at the time parturition

and days following the birth. Sertraline is a ﬁrst-line

antidepressant given safely to the PPD mother. However,

this is not the permanent cure.

3.14 Conclusion

3.13 Postpartum Psychosis

Postpartum psychosis (PP) is not a very common

condition like other mood disorders. There are

ongoing advocacy to include PP under the “Rare

Disease” umbrella. However, it is an extremely

serious neurological disorder around the globe. It

is a psychiatric emergency and required immediate

intervention and usually inpatient hospitalization,

ideally in a mother-baby unit. Symptoms of PP are as

delusions, hallucinations, disordered behaviour, mood

swings, and cognitive abnormalities. These symptoms

typically manifest in the new mother within the ﬁrst

two to four weeks after parturition. It is characterized

by the abrupt onset, and may strike one in every 1000

new mothers in general. According to population-based

studies, the prevalence of ﬁrst-lifetime onset postpartum

psychosis ranges from 0.25 to 0.6 per 1,000 births. This

risk is much higher among mothers diagnosed with

prenatal bipolar diseases. Despite the low absolute

prevalence, the relative risk for the onset of aﬀective

psychosis is 23 times higher within 4 weeks of birth

than it is at any other point in the lifetime of a woman.

When delusional perception is present, the infant is

frequently the subject of the situation.

The mother may become more protective as a

result, or she may be at risk of abuse or neglect the

new born in certain situations. Infanticide is also

known however is uncommon, occurring in 1–4.5%

of all cases. The indications of the homicidal ideas

are more common in postpartum psychosis than in

non- psychotic episodes of postpartum mood disorder.

Mothers with postpartum psychosis are more likely to

report thoughts of self-harm than those with psychiatric

issues that started at other times. So, the necessity

of prompt diagnosis, hospitalization, and treatment

commencement highlights the severity of this ailment.

Lithium is the eﬀective drug prescribed for the

treatment. It is considered a gold standard for

prophylaxis (prevention of relapse) of postpartum

psychosis (PPP) and is often used in combination with

antipsychotics and benzodiazepines for acute treatment.

However, its long-term impact on the individual as

well as on the new-born is not yet comprehensively

validated.

3.14 Conclusion

Neurological diseases, which include

neuropsychiatric symptoms that disturbs emotional

and cognitive balance as well as neurodegenerative

conditions with unrelenting progression, have a

signiﬁcant impact on the individuals, their families, and

the overall society. Aging populations, environmental

variables, and changing lifestyle choices are the

causes of the increasing prevalence, which exacerbates

problems with everyday functioning, healthcare

systems, and economic productivity in a variety of

geographical areas.

As evidenced by recent paradigms for mood

and behavioural disorders, this growing impact

highlights the urgent need for integrated treatment

approaches that combine pharmacotherapy for symptom

control, neuromodulation techniques like deep brain

stimulation for refractory cases, and psychotherapy

adapted to neurological contexts. The identiﬁcation

of early biomarkers for accurate diagnosis and the

creation of tailored therapies to target underlying

pathophysiological mechanisms must be given top

priority in future research with the ultimate goal of

reducing suﬀering and improving quality of life globally.

References:

Berk, M., Corrales, A., Trisno, R., Dodd, S.,

Yatham, L. N., Vieta, E., McIntyre, R. S., Suppes, T.,

& Agustini, B. (2025). Bipolar II disorder: a state-of-

the-art review, World psychiatry : oﬃcial journal of the

World Psychiatric Association (WPA)(24)(2), 175–189.

Chen, Z. W., Zhang, X. F., & Tu, Z. M. (2024).

Treatment measures for seasonal aﬀective disorder: A

network meta-analysis., Journal of aﬀective disorders

(350), 531–536.

Cui, L., Li, S., Wang, S., Wu, X., Liu, Y., Yu,

W., Wang, Y., Tang, Y., Xia, M., & Li, B. (2024).

Major depressive disorder: hypothesis, mechanism,

3.14 Conclusion

prevention and treatment. Signal transduction and

targeted therapy, 9(1), 30.

Gadhave, D. G., Sugandhi, V. V., Jha, S. K.,

Nangare, S. N., Gupta, G., Singh, S. K., Dua,

K., Cho, H., Hansbro, P. M., & Paudel, K. R.

(2024). Neurodegenerative disorders: Mechanisms

of degeneration and therapeutic approaches with their

clinical relevance. Ageing research reviews, 99, 102357.

Lamptey, R. N. L., Chaulagain, B., Trivedi, R.,

Gothwal, A., Layek, B., & Singh, J. (2022). A Review

of the Common Neurodegenerative Disorders: Current

Therapeutic Approaches and the Potential Role of

Nanotherapeutics. International journal of molecular

sciences, 23(3), 1851.

Salpekar, J. A., Mula, M., Agrawal, N., &

Kaufman, K. R. (2025). Neuropsychiatry as a

paradigm propelling neurology and psychiatry into

the future.(BJPsych open)(11)(2), e38.

About the Author

Dr. Neelima Dubey is currently working

as an Associate Principal Research Scientist at

Dr. Reddy’s Institute of Lifesciences, Hyderabad.

She formerly worked as a Project Scientist and as

Assistant Professor at National Centre for Cell Science

(NCCS), Pune and Dr. D.Y. Patil Biotechnology &

Bioinformatics Institute (DYPBBI), Pune, respectively.

She is interested in investigating endocrine-related

mood disorders. Her lab utilizes patient-derived

in vitro disease models and other multidisciplinary

approach to investigate molecular, cellular, and circuit

level changes underlying endocrine-related mood

disorders.

Part IV

General

AI in India: From Policy Vision to Everyday

Governance

by Atharva Pathak

airis4D, Vol.4, No.1, 2026

www.airis4d.com

Artiﬁcial Intelligence (AI) has rapidly transitioned

from an experimental technology to a strategic

instrument of governance in India. What was

once conﬁned to academic research labs and private-

sector innovation hubs is now inﬂuencing how

policies are framed, implemented, monitored, and

reﬁned. Across ministries, state governments, urban

bodies, and research institutions, AI—combined with

machine learning, data analytics, and high-performance

computing—is reshaping administrative eﬃciency and

policy-driven decision-making.

India’s approach to AI is distinctive. Rather than

focusing solely on automation, the emphasis lies on

augmenting human decision-making, improving service

delivery, and enabling evidence-based governance

at scale. This balance between technology and

institutional responsibility deﬁnes the country’s

evolving AI ecosystem.

1.1 India’s National AI Vision

The cornerstone of India’s AI strategy is the

IndiaAI Mission, approved by the Union Cabinet in

2024. The mission positions AI as a national capability,

not merely a commercial product. Its scope spans

governance, healthcare, agriculture, education, climate

action, and scientiﬁc research, with a strong focus on

accessibility, ethics, and inclusivity.

A senior oﬃcial associated with the mission

remarked during its launch that *“AI must become a

public good—available, explainable, and accountable—

rather than a black-box privilege.” This philosophy

underpins initiatives such as the IndiaAI Dataset

Platform, which aggregates anonymised datasets from

1.3 AI in Policy Design and Decision-Making

multiple ministries and departments. These datasets—

covering demographics, health, environment, transport,

and socio-economic indicators—form the backbone for

training machine learning models used in governance

and research.

Complementing data access are **Centres of

Excellence (CoEs)** established across thematic

domains. These centres translate policy priorities

into deployable solutions, ensuring that innovations

move beyond pilot projects into real-world applications.

The Union Budget 2025’s allocation for an AI

Centre of Excellence in education reﬂects a long-

term commitment to institutionalising AI-driven

transformation.

1.2

Capacity Building for AI-Enabled

Governance

Technology alone cannot deliver better governance

without skilled institutions. Recognising this, India has

introduced an AI Competency Framework for civil

servants, designed to build foundational AI literacy

across administrative roles. The objective is not to turn

administrators into data scientists, but to enable them

to interpret AI outputs, question model assumptions,

and make informed decisions.

This institutional eﬀort aligns with the National

Education Policy (NEP) 2020, which embeds AI,

data science, and computational thinking into school

and university curricula. Initiatives under the

IndiaAI Mission further support postgraduate and

doctoral research, ensuring a steady pipeline of skilled

professionals for governance, industry, and academia.

As one district collector involved in an AI pilot

project observed, “AI doesn’t replace administrative

judgement—it sharpens it by showing patterns we would

otherwise miss.”

1.3 AI in Policy Design and

Decision-Making

One of the most transformative roles of AI

lies in policy design and simulation. By analysing

large administrative datasets, machine learning models

can identify ineﬃciencies, predict outcomes, and

simulate alternative policy scenarios before nationwide

implementation.

In welfare governance, AI systems trained on

census data, socio-economic surveys, and Direct Beneﬁt

Transfer (DBT) records help identify inclusion and

exclusion errors. Policymakers can test changes in

eligibility criteria digitally, reducing leakages and

improving targeting. This marks a shift from reactive

governance to anticipatory and adaptive policymaking.

Budget allocation is another area where AI-driven

analytics is gaining traction. Predictive models assess

programme performance across regions, enabling more

rational distribution of resources based on evidence

rather than precedent.

1.6 Citizen-Centric Service Delivery

1.4 Environmental Governance and

Climate Intelligence

AI has become indispensable in environmental

monitoring and climate governance. Urban air

quality forecasting systems now integrate satellite

data, traﬃc patterns, meteorological inputs, and

industrial emissions to generate short-term pollution

forecasts. The proposed collaboration between the

Delhi government and IIT Kanpur demonstrates how

such models can guide traﬃc restrictions, school

advisories, and public health alerts.

In disaster-prone regions, AI-powered ﬂood

forecasting models analyse rainfall, river gauge data,

and terrain maps to issue early warnings. States such as

Assam and Bihar have begun experimenting with these

systems to support evacuation planning and resource

deployment. Even a few hours of advanced warning

can signiﬁcantly reduce loss of life and infrastructure

damage.

1.5 Public Safety, Law Enforcement,

and Justice

AI applications in law enforcement focus on

decision support rather than autonomous enforcement.

The Maharashtra Police’s MARVEL (Multi-Agency

Research and Vigilance for Enforcement of Law)

initiative uses AI to analyse crime patterns across

jurisdictions, helping investigators identify repeat

oﬀenders and emerging trends.

In the judicial system, natural language processing

tools are being explored to summarise lengthy case

documents, retrieve relevant precedents, and assist legal

research. Legal experts consistently stress that such

systems must remain advisory. As one senior jurist

noted, “Technology can assist the mind, but justice must

always remain a human responsibility.”

1.6 Citizen-Centric Service Delivery

AI is increasingly visible at the citizen interface.

Platforms such as UMANG integrate hundreds of

government services into a single digital ecosystem. AI-

powered chatbots, grievance classiﬁcation systems, and

sentiment analysis tools help administrations respond

faster and more eﬀectively.

For citizens, this reduces bureaucratic friction

and waiting times. For administrators, it provides

dashboards highlighting recurring issues and service

gaps—enabling continuous service improvement and

better accountability.

1.7 AI in Healthcare, Research, and

Disaster Response

Public healthcare has emerged as a major

beneﬁciary of AI-driven analytics. Disease surveillance

systems analyse hospital admissions, pharmacy sales,

1.9 Conclusion

mobility data, and weather trends to detect early signs

of outbreaks. These insights allow health departments

to mobilise resources proactively.

In research, AI is accelerating discoveries across

domains such as climate science, astronomy, genomics,

and materials engineering. By automating data-

intensive tasks and enabling complex simulations, AI

is transforming how scientiﬁc knowledge is generated

and applied in India.

1.8 Ethics, Trust, and Responsible AI

As AI becomes embedded in governance, ethical

considerations are paramount. India’s evolving

AI governance framework emphasises transparency,

fairness, privacy protection, and human oversight.

Bias audits, explainable models, and accountability

mechanisms are essential to ensure that AI strengthens

democratic institutions rather than undermining them.

The guiding principle remains clear: AI should

enhance human judgement, not replace it.

1.9 Conclusion

Artiﬁcial intelligence in India has moved beyond

experimentation into practical governance. Through

national missions, institutional capacity building, and

real-world deployments, AI is enabling smarter policies,

eﬃcient administration, and accelerated research.

By aligning policy vision with technological

capability and ethical responsibility, India oﬀers a

compelling model of how AI can serve as a force

multiplier for governance—empowering administrators,

supporting researchers, and improving the everyday

lives of citizens.

References

1. Oﬃce of the Principal Scientiﬁc Adviser,

Government of India. IndiaAI Mission.

[https://www.psa.gov.in/ai-

mission](https://www.psa.gov.in/ai-mission)

2. Ministry of Electronics and Information

Technology (MeitY). India AI Governance Guidelines.

3. UNESCO. India Launches AI Competency

Framework to Transform Public Service.

4. Government of India. National Education

Policy 2020.

5. Times of India. Delhi Government Explores

AI-Enabled Air Pollution Management with IIT Kanpur.

6. Maharashtra Police. MARVEL – Maharashtra

Advanced Research and Vigilance for Enhanced Law

Enforcement.

7. Government of India. UMANG – Uniﬁed

Mobile Application for New-age Governance.

8. arXiv. AI-Based Legal Judgment Prediction

and Summarisation for India.

9. NITI Aayog. Responsible AI for All – National

Strategy for Artiﬁcial Intelligence.

About the Author

Atharva Pathak currently work as

a Software Engineer & Data Manager for the

Pune Knowledge Cluster, A project under the

Oﬃce of Principal Scientiﬁc Advisor, Govt. of

India & Supported by IUCAA, Pune, IN. Before

this, I was an Astronomer at the Inter-University

Centre for Astronomy & Astrophysics, IUCAA. I

have also worked on various freelance projects,

development required for websites and applications,

And localization of diﬀerent software. I am also a

life member of Jyotirvidya Parisanstha, India’s Oldest

association of Amateur Astronomers, and I look after

the IOTA-India Occultation section as a webmaster

and data curator.

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.