A 35-Year Odyssey with AI

By Laxmidhar Behera, Director, IIT Mandi and Professor, IIT Kanpur

Published on: February 18, 2026

Introduction

Today, when students at IIT Mandi ask ChatGPT to write a poem or debug code, they often take for granted the invisible magic that makes it possible. For them, Artificial Intelligence (AI) is a utility, as ubiquitous as electricity. However, for me, AI is not merely a product; it is a journey—a scientific romance that began in 1988, long before Deep Learning became a buzzword, and has culminated in a national imperative for India.

Early Beginnings

My tryst with AI began during my Master’s program (1988–1990) and subsequent PhD at IIT Delhi (1991–1995). Under the mentorship of distinguished professors like Prof. Madan Gopal and Prof. Santanu Chaudhury, I dove into the then-esoteric world of Neural Networks and Robotic Manipulators.

To explain this to a layperson, imagine teaching a child to recognize animals. You show them thousands of pictures and say, “This is a cat,” or “This is a dog.” Eventually, the child’s brain forms a pattern—a model—that can identify a new animal instantly. This is known as the Forward Problem – System Identification: Given an input (x), tell me what it is (y).

Advancing Research

My research, however, pushed this boundary further into Network Inversion. I asked the opposite question: “If I tell you ‘African Cat’ (y), can you draw it for me (x)?” In the early 90s, this was a radical proposition. We were essentially trying to reverse-engineer the brain of the machine to generate data from concepts—a precursor to today’s Generative AI.

We published these ideas in prestigious journals like IEEE Transactions on Neural Networks, yet the significance was largely academic. The world wasn’t ready. The computing power required to simulate even a single hidden layer of neurons on our VAX machines was excruciatingly slow. We were trying to simulate a brain with the power of a calculator.

The AI Winter

Then came the AI Winter (1995–2005). Funding dried up, and skepticism grew. But my innate interest—unraveling how the inner engineering of human thinking could be replicated in the outer engineering of machines—kept me going. We waited for the hardware to catch up to our math.

The Thaw and Revolution

The thaw finally arrived with Geoffrey Hinton’s work on Deep Belief Networks and Restricted Boltzmann Machines. Suddenly, Deep Learning wasn’t just a theory; it was a revolution. The emergence of Graphics Processing Units (GPUs) changed the game entirely. Where we once struggled with one hidden layer, we could now simulate networks with 100 layers with ease.

A Watershed Moment

The vindication of our persistence came in 2017 at the Amazon Robotics Challenge in Nagoya, Japan. Our team arrived with an 8-GPU server—a beast of a machine compared to the VAX of my PhD days. The challenge was brutal: The AI had to learn to recognize and pick up 50% of unknown objects in just 20 minutes. While teams from MIT, CMU, and Princeton struggled, our algorithm adapted in real-time. We secured the 3rd position in the picking task. It was a watershed moment, proving that Indian algorithms could compete with—and defeat—the best in the world.

Entrepreneurship and New Beginnings

Fueled by this success, I ventured into entrepreneurship, founding a company at IIT Kanpur in 2019 to revolutionize AI education. We were making great strides until the Covid-19 pandemic disrupted the world. I closed that chapter, but another opened when I accepted the responsibility of director at IIT Mandi in January 2022.

Current Landscape

By then, the world had changed again. ChatGPT, Gemini, and Sarvam AI had made Generative AI a household name. However, as I looked at the landscape, I realized a critical gap. AI is hungry for data, but the diet it is fed is overwhelmingly western. If India is to be a leader, and not just a consumer, we need Indian data. We need AI that understands our diverse languages, diagnoses diseases based on Indian genetic profiles, improves our unique agricultural patterns, and solves our chaotic road safety issues.

The Government of India’s IndiaAI Mission is a commendable step in this direction, ensuring we do not remain backward in technology.

The Philosophical Approach

However, my journey has taught me that technology alone is not enough. While the West chases big data and massive computational models, I am reminded of the wisdom of our ancient seers. They taught that true cognition often arises not from noise, but from silence—from little data. The Indian Knowledge System (IKS) offers a profound counter-narrative to the brute-force approach of modern AI. It suggests that intelligence is not just about processing terabytes of information, but about inner engineering—understanding the consciousness that perceives that information.

The Next Frontier

This is the next frontier for us. We must build AI that is not only smart but ethical and empathetic. We need algorithms that care for mental health as much as they care for efficiency. By marrying the computational power of the GPU with the cognitive depth of the rishi, India can offer the world a new kind of AI: one that is powerful, purposeful, and profoundly human.

Mind-Matter Interaction

One of the most exciting directions in this regard is the field of mind-matter interaction, in which we have taken a significant lead at IIT Mandi. We are exploring how human consciousness—the “mind”—can interface directly with physical systems—the “matter.” This research bridges the gap between ancient Indian wisdom and modern neuroscience, potentially opening doors to AI systems that are responsive to human intent and mental states.

Conclusion

From the slow hum of a VAX machine at IIT Delhi to the lightning-fast servers of the Amazon Challenge, and now to the strategic corridors of IIT Mandi, my journey has been one of persistent curiosity. The winter is over. The Indian summer of AI is just beginning.

Note