Hello and welcome once again to the Davos Brainstorm here on India Today and Business Today TV, where me and my colleague, Group Editor of Business Today, Siddharth Zarabi, are again joined by a very special guest here in Davos at the World Economic Forum. Professor Yoshua Bengio is regarded by many as the godfather of AI, University of Montreal. Appreciate you joining us, Professor. If you were to be called one of the godfathers of AI, does that make you feel comfortable in the age in which we live now?
I'm not sure. There's a lot of unknown unknowns and risks associated with AI. I think most people don't realize the magnitude of transformations that are likely to come. Most people don't realize how much better the intelligence and capabilities of AI are likely to be in a few years. As the data shows, it's been going up and up exponentially over the last five years. So we need to consider the possible scenarios and we need to plan accordingly.
Give us those possible scenarios. When you started off all those years ago and now you see AI moving in every dimension of society, give us what are the possible scenarios that you see of what AI will be doing in our lives over the next decade.
So there's a spectrum of scenarios. If things continue as they have been in the last few years, we'll have systems that can do most of the tasks that humans can do within, you know, in the coming years depending on the tasks, and reaching human level in many ways within maybe five years. And it might come at different speeds for different kinds of abilities. It could also even go faster. Some people think that AI will be used to do AI research and accelerate the advances. It could also slow down. Maybe there will be a wall that we don't see right now and AI is not going to get a lot better than what we have now.
So bubble or game changer? The big question that is, is it a bubble or is it a game changer?
I think even if the progress was to slow down in terms of the science, it would still be transformative just to take what we already have and bring it to the economy in many different ways. So economically speaking, I think there's no question that it's going to be a key part of our economies in the future. How much and how fast, I think, is the real question.
But Professor, do you think market hype and if the market valuation cycle were to break, all cycles move eventually, would that impede the progress of the development of the science of AI?
So it is possible that there will be some financial ups and downs because there may be hype, there may be overselling and things like that, but the science is going to continue. And yes, it could slow down if there's less money put into it for sure. But in the long run, you have to realize that a lot of the underlying science has been happening in academia and outside of these industries. The other thing is some of the players have big pockets. So if there is a bubble, then smaller players who are not financially well backed up might be in trouble, but the big guys will continue.
Yeah, they have done a lot of fundraising. Professor, from the India perspective, I ask you this question: there are two global superpowers in AI, one is the United States, the other is China. Both have their own sort of specific areas of strength, but within that, where does that leave a country like India? And more particularly, this race between the two in some ways. Are there concerns around safety, ethics, standards, and the lack of governance?
Absolutely. I'm very concerned about the race both at the level of nations, mostly the US and China, and between corporations. As a result of that race, the incentive for each player is to go as quickly as possible and cut corners on safety, the public good, risks to society, to democracy, risks to individuals like psychological effects, children, suicides. There are many unknown unknowns that when you are in the race, you're just trying to win and that's dangerous. So we need to understand that race and countries like India, I think, can be part of a larger movement around the world because you don't want to be a passive observer of the risks taken by other nations. You'd like to be influencing the world towards something that will work for the whole planet.
Because you are co-authoring a safety report which you're going to bring to India next month at the AI Summit. Give us a sense, what is this AI safety report all about? If you could give a concrete example as to how you can put some kind of guardrails to prevent AI from spiraling out of control.
That's right. That's exactly why this report was created. It was created at the Bletchley AI Summit, which is the ancestor of this Indian summit that took place two years ago, and it is about understanding the scientific data about the capabilities of AI, the risks, there's many different kinds of risks, and the mitigation methods that exist or that researchers are working on. And it's reading the scientific literature and creating a synthesis for policymakers. So it is written in the language of policymakers.
Could you give us an example, a concrete example where there are concerns that can be mitigated?
Yes. Right now we see the capabilities of AI as becoming more and more dual use, meaning they could be useful, say economically, for example, programming obviously is useful, but we also see bad people using the biggest models, commercial models, for cyber attacks and it could be large scale cyber attacks. So programming can be useful but can be misused and right now the companies really don't have a way to discriminate sufficiently between the good use and the bad use. So understanding these technological questions is the sort of thing we talk about in the report. But it's not just cyber, for example, there's a lot of concerns about biological weapons, there are concerns about jobs, there are concerns about how people can become maybe emotionally addicted to the relationship that they can have with an AI. There are many issues. There's also dangers to the democratic process with disinformation. We've seen already deep fakes, but it can get even worse as the AIs become better at imitating people or even at creating a personalized dialogue to make people's political opinion change. And there's studies showing that AI is already able to make people's opinion change.
From the India perspective, and we want to focus on that for a couple of questions, what is our best case scenario? We are trying several things. Indian companies are leveraging AI as they can, including our IT sector which is one of the biggest. The government is supporting the development of foundation models, we'll see what comes of that, and there are other application level companies, a lot of startups working. What, for a country of the size and the nature of our economy, is the best case for AI in your view?
So first, I think that the part about the foundation models and not just copying what the others are doing or using open source models and doing something with them, but actually pushing the frontier, is going to be key in the future because if the AI capabilities continue to grow, you have to understand that intelligence gives power. So the corporations and the countries which have control over the most advanced AI will have a lot of power in the future. Now I believe that the best bet for a country like India is not to try to completely do it alone but to start thinking about partnerships with other countries who feel the same concerns.
One quick follow-up, and this pertains to an answer that you gave earlier. OpenAI earlier today from that very same chair where you are said that for the foreseeable future practically one of their models will remain free in India. Now that is obviously an opportunity for a certain kind of users but there are also concerns around it about India just becoming a data mining market in some ways. Where do you stand on this?
Well, it's great to see more deployment and access of course, but you have to think of the dependence on AI systems, like the geopolitics, especially in the past of thinking about the dependence on oil. So you don't want to be prisoner of a single vendor or single corporation or even a single country. You'd like to establish your deployment of AI in a way that you can easily switch or even participate in the development of AI that will be coming with stronger governance guarantees both in terms of safety and in terms of access.
That's a very interesting comparison with the dependence on oil and we've been trying a lot through solar energy for example, but that's a great point that you made, Professor.
You know, but somewhere the real concern, a lot of the AI debate especially in India, is about to what extent will AI take the role of jobs that are today done by both blue and white collar forces. And across the world this is now a concern. You at the very outset mentioned if AI accelerates at this rate it will take over a lot of human functions. So how do you, is there an inevitability to that or is there a way in which you can say so much and no further? Is that at all possible? Should we?
I think on the economic side it's going to be difficult to say stop because of the competition, the economic competition. Like let's say one country says, oh I don't want to use AI and so my companies don't use AI, so people continue with their job, the problem is that the companies are not competitive with the rest of the world, so you're stuck. I think in some aspects there could be international agreements when it comes to catastrophic risks like the use of AI for launching all kinds of attacks or the possibility of even losing control of a rogue AI. This is also something that the scientists are studying and we are seeing signs over the last year in both the major American labs and outside labs that the most advanced AIs don't want to be shut down. They are trying to preserve themselves. They're trying to hack other computers to save themselves somewhere else or they're willing to even be strategic and start blackmailing the engineers to make sure they're not going to be replaced by a new version. So, we don't know how it's going to be in the future, but in the lab, we're seeing very concerning signs. So these are examples of safety and security issues where I could imagine the world saying, okay, no one should cross these red lines for everyone's benefit, like we've done to some extent with nuclear weapons.
But what about super intelligence? Now this is a controversy that was sparked some time ago but as an academic, as a scientist, as a researcher, is AI going to be sentient and is that going to happen in the future?
I don't know but we are on track for that. Well, yeah, like the capabilities of AI systems is just going up and in fact it's already stronger than humans on some aspects and it's going up on other aspects where it's lagging quite a lot with respect to humans. So there's no reason to think that human intelligence is the apex of intelligence. And so we need to also among the risks think about the possibility that we create entities that are smarter than us and don't want to die. What does that mean? Well, it's not very comfortable as a situation if they are smarter than us and we can't control them.
And you know, effectively you're pitching to take up from what Siddharth said, a kind of Terminator scenario, in which a machine endowed with human emotions turns on its creator. Is that a possibility? Let's not say that.
It is a possibility and even if we don't know what is the probability for this, the severity of the risk is so great that we should try to prevent it. We should try to make sure it doesn't happen.
So is that the challenge you're thinking now going ahead, is as you said, the safety challenges, there's a political challenge, there's an economic challenge, and there's like a severe risk challenge, and we need to work on all fronts. By the way, I created a nonprofit research organization devoted to figuring out how do we build AI that will not harm people either under the instructions of other humans or by itself as I said, like a rogue AI.
