India is letting the AI revolution bypass the country. It may have to pay a heavy price

It takes four years to complete a PhD at the Imperial College of Science, Technology and Medicine. But when a chatbot was given a PhD-level problem at a gathering of top mathematicians in Berkeley, it solved it in just ten minutes.

Illustration: Prajna Ghosh | ThePrint

In 1986, when the world was still largely and blissfully unaware of the meaning of artificial intelligence, I had an interesting one-on-one tutorial on a new software called Graffiti, developed by my friend and colleague at the University of Houston, Siemion Fajtlowicz. I was lecturing in the US, and at Simeon’s urging, I sat with him as he demonstrated the mathematical capabilities of Graffiti. 

To my mind, that was one of the early forays of computer science and mathematics into a kind of artificial intelligence. What Siemion had created was software capable of generating new and non-trivial mathematical problems in the field of graph theory, a branch of mathematics. When I next interacted with him in 1992, he had developed Graffiti into a major conjecture–generating tool for both chemistry and graph theory. In fact, The New York Times carried a detailed feature on Siemion’s programme. 

I could see that Graffiti was making waves — some of the world’s leading graph theorists were publishing results that proved conjectures made by the software. Of course, Graffiti is rather different from today’s large language models (LLMs), but it remains a pioneer of sorts. 

There are three reasons I mention this software. First, Graffiti qualifies as AI in the context of its time (1980s–1990s) and within its domain-specific capabilities. It exhibited several characteristics of AI, such as automated reasoning, problem-solving, and the generation of novel conjectures that mimicked aspects of human mathematical creativity.

Its autonomy in conjecture-making and its impact on mathematical research further support this classification. If memory serves me right, one of the greats of 20th century mathematics, Paul Erdős, once told me that he had worked on some of Graffiti’s conjectures.

Another reason I mention Graffiti is that, in all my years of interacting with graph theorists in India — and their numbers abound — not one has shown any knowledge of or interest in using computers for purposes similar to what Graffiti does. I have also not come across a single graph theory course that incorporates ideas similar to Graffiti. This worrisome state of affairs continues to this day. And now, as advances in AI accelerate, India is being left behind in taking advantage of the technology for deep and meaningful knowledge creation.

My final reason for mentioning Graffiti is to underscore that the recent spurt in AI LLMs owes much to early programmes like Graffiti and several other platforms whose seeds were sown in the 1980s and 1990s. Once significant progress was made in data science, neural networks, and related fields, this leap was inevitable.

Even though India had, at one point, achieved a degree of proficiency in coding, we allowed the AI revolution to bypass us. Primarily, this was due to our unimaginative curricula and a lack of intellectual boldness. My fear is that we may pay a heavy price for our neglect over the past four decades.

Exciting and alarming

While the current advances in AI are exciting, they are also somewhat worrying. In mid-May, around 30 top mathematicians, some traveling from the UK, held a secret gathering in Berkeley, California.

Their mission: to challenge a unique “reasoning” chatbot with a series of difficult mathematical problems they had created. For two days, these experts, including University of Virginia mathematician Ken Ono, who served as the leader and judge, posed professor-level questions to the bot. To their astonishment, the chatbot solved all of them. 

“I have colleagues who said these models are approaching mathematical genius,” Ono remarked. 

For perspective, my colleagues and I recently experimented with some highly trained chatbots that specialise in mathematical and logical reasoning. On one platform, we posed  advanced undergraduate–level questions. The chatbot gave us ‘solutions’ almost instantly, and asserted their veracity with supreme confidence.

Unfortunately, the solutions were found to be incorrect on close human scrutiny. Contrast this with the chatbot that faced off against expert mathematicians in Berkeley, and it becomes clear this is an entirely different cup of tea. Its ability to operate successfully in the realm of advanced deductive reasoning stems from o4-mini, a reasoning-focused large language model trained by OpenAI.

Let me give a clearer, starker glimpse of just how impressive these capabilities are. At the Imperial College of Science, Technology and Medicine — where I obtained my doctoral degree — the average time to complete a PhD has been about four years. Any student admitted to a doctoral programme at Imperial is already considered quite smart. But if someone earns their degree in under three years, they are considered exceptionally intelligent.

At the Berkeley gathering, the chatbot was presented with a difficult PhD-level problem. According to Ono, it solved the problem in just ten minutes and asserted its dominance with supreme confidence. This makes me wonder about the future of doctoral research, and even mathematical research more broadly.

In fact, Ono is worried that we may be too easily intimidated by such a chatbot into believing it without question.

“There’s proof by induction, proof by contradiction, and then proof by intimidation,” he said. “If you say something with enough authority, people just get scared. I think o4-mini has mastered proof by intimidation; it says everything with so much confidence.”

We mathematicians in particular, and the public at large, should be both excited and alarmed.

Dinesh Singh is the former Vice Chancellor of the University of Delhi and adjunct professor of mathematics at the University of Houston, Texas, USA. He tweets @DineshSinghEDU. Views are personal.

(Edited by Aamaan Alam Khan)