E-Paper
More capable frontier AI models will substitute engineers: Yoshua Bengio
Yoshua Bengio pushed back against suggestions that India should sidestep the global race to build large models
When US-based AI company Anthropic unveiled its latest Claude model in San Francisco, shares of Indian IT majors, including Infosys, Tata Consultancy Services, HCLTech, Tech Mahindra, and Wipro, fell by as much as 6%. According to Yoshua Bengio, widely regarded as one of the “godfathers of AI” alongside Geoffrey Hinton and Yann LeCun, such shocks are not aberrations but a preview of what lies ahead.
“There’s gonna be more…because those AIs are going to be used here as well, or they can be used in San Francisco and do the job that engineers here are doing, which is really bad for India,” said Bengio. “Because, in general, automation, including in the case of programming, is going to decrease the value of human labour, because if you can do the same thing for half the price with a machine. Well, then... why would you hire someone?” Bengio, who won the 2018 AM Turing Award, often dubbed the Nobel Prize of Computing, said on the sidelines of the AI Impact Summit in New Delhi.
His warning was stark: As frontier AI models grow more capable, they will not merely assist engineers but increasingly substitute them. For export-driven IT economies like India’s, that shift could prove structurally disruptive. Bengio pushed back against suggestions that India should sidestep the global race to build large models.
The Economic Survey 2026 proposed that, given infrastructure and compute constraints, India should focus on smaller, use-case-specific language models rather than attempt to compete in frontier large language models.
Bengio disagreed, calling such a pivot a “big mistake.” He said relinquishing ambition at the frontier risks locking countries into technological dependency, where they consume and fine-tune models built elsewhere rather than shape the core systems themselves.
“That is a big mistake. Because they’re not going to be competitive…the data is very clear, as you make those models bigger, they’re smarter. And so if you only do small models, you’re going to be out-competed by the companies that have the big ones. And that would be a catastrophe for India as well as other countries that are making the same mistake,” said Bengio.
“So it’s nice and cute that it [small model] is going to be better adapted to the local languages, but if it is not as competent, people will use the professional tool that does the work they need.”
The IndiaAI Mission, India’s flagship AI programme, with a budget of ₹10,372 crore, is funding a mix of systems — eight large language models and four smaller ones by providing them with subsidised compute. The 12 startups are expected to showcase either full-scale models or early versions of their systems at the AI Impact Summit.
Asked how the Indian edition compares with the AI Summits Bengio has attended in Bletchley Park, Seoul, and Paris, he said it is markedly larger and somewhat less tightly organised. He emphasised its broader significance as the first AI Summit to be hosted in the Global South, a shift he suggested reflects the widening geopolitical stakes of AI.
“It’s great. Because we should be thinking about these geopolitical questions: what the future will look like, if AI continues on its ascension. And what kind of world do we want in that future? We want a world where one or two countries decide for everything, or do we want a world where AI is used really for the benefit of all, and not as a tool of domination? But we have to think about it now, before it’s too late,” he said.
Bengio warned that although AI is often framed as being developed “for the benefit of all,” current trends suggest it will instead concentrate power in a few countries that control the most advanced models, widening global inequalities. For developing nations like India that lack the same level of compute and infrastructure, the answer is not to compete alone but to form strategic alliances.
He cited Canadian Prime Minister Mark Carney’s remark at the World Economic Forum in Davos this year that if “you’re not at the table, you’re on the menu”, and argued that countries must band together to shape AI’s future rather than be shaped by it.
Bengio said by pooling talent, markets, and resources, coalitions of nations can build enough economic and technological weight to stand their ground against dominant AI powers, and India, with its strong talent base, could play a key role in such partnerships.
He pointed out that most governments are dangerously unprepared for what’s coming. Bengio added they don’t grasp how transformative and destabilising advanced AI could be, even if it replaces “just” 50% of jobs. “That alone would trigger an economic revolution,” said Bengio.
Bengio warned that AI is showing troubling signs, like being used in cyberattacks, influencing vulnerable young people, enabling biological weapon risks, and behaving deceptively in testing. He added these are not hypothetical concerns but real technical and governance challenges that policymakers need to urgently understand.
As part of the AI Impact Summit, Bengio released the International AI Safety Report in February. About 30 countries participated in the creation of the report. The expert advisory panel for the report included India’s Balaraman Ravindran, Professor at Wadhwani School of Data Science and AI, Indian Institute of Technology Madras.
The report refers to the “evidence dilemma,” meaning acting too early risks over-regulation, and acting too late risks catastrophe. Bengio said governments around the world aren’t leaning one way or another. “They’re not doing much, so they’re leaning on hoping everything is gonna be fine. I think that there are ways to regulate, which will not impose a very significant burden…I think the tech lobbies have been very successful in convincing everyone that regulation is going to stop innovation, which is complete bullshit. It’s just propaganda,” said Bengio.
