- 11 hours ago
India Today AI Summit 2026: Vishal Sikka says AGI talks are nonsense
Category
🗞
NewsTranscript
00:00Dr. Vishal Sikka is a prominent Indian-American technology leader, entrepreneur, and AI expert.
00:07He holds a PhD in Artificial Intelligence from Stanford University, and that was way back in 1996.
00:16So, Dr. Sikka, for the students who are here, tell us, were you the only one in the class then?
00:23Oh, no, there were a lot of us. That time period was, believe it or not, there was another similar
00:31AI hype cycle in the 1980s.
00:34And the time that I did my PhD was, things were kind of mellowing out, and we used to call
00:40it the AI winter.
00:41But there were plenty of people studying AI.
00:44So, what is a PhD in AI like? I mean, what were you preparing yourself for?
00:49So, the idea of making a machine that is intelligent has fascinated people, you know, forever.
01:00And so, around that time, we were sort of in one of these generations where formal systems, reasoning systems,
01:08were being built to apply mathematical reasoning to complex problems.
01:14For example, medical diagnosis, or detecting fraud in credit card transactions, or these kinds of things.
01:21And that technique had run into some significant limitations.
01:25I worked on a very promising approach to connect efficient subsystems into a general purpose reasoner.
01:33It's a little bit like plugins today, except the outer system was not a large language model.
01:39It was a reasoning system.
01:41So, it has been fascinating to me over the last 37 or so years that I have been involved in
01:47the field to see its development.
01:50And what, you know, what we have today is…
01:53You know, you have famously said that AI's impact on services and in software is real but not sudden.
02:03When you say it's real but not sudden, who all will survive this onslaught at present?
02:11So, it is a…
02:13You know, there are two sides to this coin or tale of two cities or whatever metaphor you want to
02:17use.
02:18On the one hand, with AI today, it is possible for a person to be so incredibly more productive with
02:28whatever they need to do.
02:30I have seen examples where people can do things 20 times, 30 times faster, more efficiently than they did before.
02:40One of my friends who was a student with me in AI, when we did our PhDs together, I was
02:47just talking to him this morning.
02:48He rebuilt a system that was built by 15 engineers over nine months.
02:55He recently rebuilt that system by himself in 14 days using one of these…
03:00So, what took nine months then?
03:02What took nine months by 15 people with 15 very high-level engineers was rebuilt by himself alone within 14
03:11days.
03:12So, that's more than a hundred times productivity gain.
03:15So, this is possible today.
03:18On the other hand, there is also, if we don't go after these new opportunities, so you can do with
03:23that kind of a power, you can do unprecedented things.
03:27But on the other hand, you know, the kind of work that we do, the routine work, the knowledge work,
03:34that can also be done much more efficiently.
03:36And so, there is clearly a disruption there.
03:39So, both of these truths are there at the same time.
03:43Because you have highlighted the fact that what was being done by 15 individuals is now being done by one
03:50person and in just 14 days.
03:52Earlier, it was in nine months.
03:54So, is AI taking jobs, replacing humans?
03:59So, it is doing both.
04:00It is creating opportunities for completely new kinds of things that were simply not possible before.
04:07And also, it is making certain kinds of work completely unnecessary.
04:12And both of these are happening at the same time.
04:15So, it is like a creative destruction that is happening at hyperspeed.
04:20And the, so, the important thing is to focus on what is it that was possible to, that is possible
04:27to do now that is of value.
04:31And so, I will give you two examples.
04:33One is that at the same time that this friend of mine, you know, did his work 100 times more
04:41efficiently,
04:43there are also people who don't see the benefit from AI because they end up making more mistakes,
04:48they don't end up catching the mistakes and then fixing the mistakes that AI makes,
04:52takes them longer than it would have taken them to do the whole thing from scratch, etc.
04:56So, I think the statement there is that in the hands of people who understand how to use this,
05:01it is an incredibly powerful tool.
05:04The second thing I want to say is that you can do completely new kinds of things before.
05:10And one of my customers, it is a large distributor in the Middle East.
05:16He was telling us that, he was actually telling some of my other customers that one of their sub-major
05:23suppliers
05:25decided to shut down their manufacturing in one of the countries in the region.
05:30And they had to quickly understand the consequences of this.
05:33That now this main FMCG company is no longer going to produce locally.
05:37What do we do as a result of this?
05:39And they ran a bunch of these complex analysis using our product, using my company's product.
05:46And as a result of that analysis, they decided to shut down their operations in that country.
05:51Now, think about how this would have unfolded before AI.
05:56You would have, maybe, somebody shuts down their factory, you hire, you know, expensive consultants.
06:02And you say, guys, what are the consequences of this?
06:05You pay them millions of dollars, you know, three months later they come back and give you this analysis.
06:10Instead, you are doing this within five minutes, within ten minutes.
06:13And you take such a major decision.
06:15So, work that was done by, if you wanted to even do it, would be done over months by teams
06:22of people,
06:23can now be done in minutes.
06:24So, then the question is, what can you do with this?
06:27You know, what powerful things can this enable for you?
06:30Yes.
06:30So, I think, you know, I am an optimist and an entrepreneur,
06:34and I like to think of what is it that we can do with this,
06:38rather than what things of the past will it disrupt?
06:42You know, because there was this very interesting conversation
06:45that you were having with our executive editor-in-chief,
06:49and she used a phrase, boss mode.
06:51So, does the AI respond best in the boss mode?
06:56The, I think bosses who are bosses, not just, I mean, she is a real boss.
07:05The, I think, if you interpret, if you say, take boss mode as,
07:12understanding the limitations of AI, taking the work that AI produces for you,
07:17in a way that amplifies what you are doing, and yet you are careful,
07:22you apply your rigour, your scrutiny to what it has done.
07:25My God, it is an unbelievably powerful tool.
07:28But if you just blindly take what it gives you, then it is not so powerful.
07:33Then it is just, it regresses you to the mean of what is known.
07:35So, AI is not a friend?
07:39It is a tool, you know.
07:42So, how you engage with AI becomes important to you?
07:45Yes.
07:45How you engage with it, how you use this tool?
07:48It is, you know, a little bit north of Chennai,
07:53about a hundred kilometres north of Chennai is a town called Atiram Pakkam.
07:56They recently found these Stone Age tools there from 250,000 years ago.
08:02Dhai laak saal purane.
08:03We were making tools 250,000 years ago.
08:06You know, this is a new tool that has been built.
08:08And the question is, what will you do with this tool?
08:11It is, the whole thing about AGI, and this is all nonsense,
08:14that we should abandon that.
08:16You look at it as a powerful tool that is kind of an approximate condensation
08:21of all the bunch of human knowledge that has been known.
08:25What will you do with this?
08:26Okay, Dr. Sikha, you know, in 2015, you invested in OpenAI.
08:32What did you see that most people did not?
08:35Well, it was a donation, because OpenAI was a pure non-profit at the time.
08:40We gave them, I think, three million dollars or something like this.
08:44Yeah, it was 11 years ago.
08:46It was, so I had just started at Infosys,
08:52and the AlexNet was already two years old.
08:57AlexNet was the computer vision system that had beat human performance
09:01on a benchmark called ImageNet that was built by Fei-Fei Li.
09:06So, all of a sudden, there was a neural network system
09:09that was able to do vision tasks even better than humans.
09:13So, the trajectory was kind of clear of where this was headed.
09:18And we had to have an AI platform that was, that we could use.
09:24And so, here was Sam who was looking to build an open AI platform.
09:30So, to me, it was kind of a, it was obvious.
09:33It seemed straightforward that we should do this.
09:36Okay.
09:37And since you're talking about Infosys, back then, you wanted Infosys to be India's AI first company.
09:47As someone who has the knack of spotting transformative technologies,
09:52why couldn't you implement it then?
09:54Why didn't we take that leap through Infosys or similar companies,
09:58what perhaps United States and other countries did?
10:03Well, we did a lot.
10:04You know, we built our own platform back then.
10:06We had a large collection of efforts that we were doing,
10:09which I thought was remarkable for the time.
10:12But I, you know, like I said, I tend to look forward and look at what is possible now with
10:18AI.
10:19And I think, you know, at the time, we, what we had, we did what we could.
10:23And it's a different time now, so.
10:26Okay.
10:27In terms of its limits, like hallucinations in particular, which is being repeatedly spoken about,
10:32where do you think role of companies and also countries come in with regards to regulation?
10:38And in summits such as this, which is happening in India,
10:42do you think India will show the way for regulations?
10:45So, you know, the idea of regulation, it is, we are sitting here in this room
10:51and we are comfortable that this roof is not going to fall on our heads
10:55because of regulation.
10:57Okay.
10:57You know, even like barbers who cut our hair are regulated.
11:02So, what is the problem with regulating AI?
11:05It is an unbelievably powerful technology.
11:08You better believe it should be regulated.
11:09And I am very proud that the Indian government actually is taking a very leading
11:13and uncompromising stand on regulation, on safety, on ethics.
11:17It is very important to do that.
11:20The point about hallucinations,
11:23AI today, generative AI, is inherently hallucinatory.
11:29It is not a deterministic system.
11:31It is a system based on probabilities.
11:32You pick the most likely next token when you give it a prompt.
11:37And so, that hallucination creates the need for regulation, for safety and things like that.
11:44It also creates an opportunity.
11:47People and organizations who can understand how to tame the hallucination,
11:52how to close that last mile.
11:54By the way, it is much worse than the last mile.
11:57It is more like last hundred miles at enterprises, for example.
12:01Those organizations are going to do very, very well.
12:04And so, hallucination is an inherent attribute of AI.
12:08It creates the burden of regulation, the burden of safety and duty of care and things like that.
12:16At the same time, it also creates an opportunity for companies to mitigate it.
12:20And my company works in this area in order to deliver valuable solutions.
12:27Dr. Sikha, when we started this conversation, you said that there was an AI bubble in the 1980s.
12:33Right?
12:34Yeah.
12:34So, is this also a bubble?
12:36Dr. Sikha, I mean, the thing is that there is a whole stock market and valuations, all of that.
12:43I don't want to get into all that.
12:45But there have been hype cycles around AI multiple times before.
12:50When Frank Rosenblatt built the Perceptron in the 1950s,
12:54the New York Times had a headline that the scientist has built a system that mimics the human brain.
13:00And you know, so there has been hype around AI many times before.
13:04In the 1980s, there was a whole massive hype around AI.
13:08There were VCs that were exclusively AI VCs.
13:11There were hardware companies.
13:12There were all kinds of services companies focusing on AI.
13:16People find it hard to believe now, but that happened.
13:21If you look at the math today, there are a few gaps.
13:26Let's put it like this.
13:28For example, Nvidia, which is one of the incredible companies,
13:33they have done extraordinary engineering.
13:36If you look at their revenues are in the neighborhood of $60 billion a quarter.
13:41So that's about $250 billion a year, let's say, give or take.
13:45They make a real product.
13:47You know, they make a GPU.
13:49It's about the size of a toaster oven.
13:53People buy it, $250 billion a year.
13:56This is real.
13:57On top of the GPU, cloud companies add another 2x the amount to make that available as infrastructure.
14:07So that's around $750 billion, $800 billion a year of real spend.
14:11So, so far we are in the real world.
14:14And then people expenses on top of it and all that.
14:17If you amortize this $800 billion over, let's say, 4 years,
14:20that means you need $200 billion a year in revenue just to recover these $800 billion in spend.
14:28So where is that $200 billion?
14:31Open AI is at what, $15-$20 billion revenue, something like this.
14:34So there is a huge gap.
14:36And that gap, you could say that gap is a bubble.
14:39You could say that gap is a gap in the kind of applications that people have not been able to
14:44produce yet.
14:45But there is, it is very clear that yes, there is a severe shortage of GPUs and all of that.
14:51But at the same time, valuable applications of AI that enterprises or other people find worthwhile are not there yet.
15:01And we need to deal with that.
15:03So as a visionary, what will be your cautionary tale on AI?
15:08Because there has been similar bubbles multiple times in the past.
15:14You have to invest carefully, you know, make sure there is ROI and things like this.
15:20And on the other hand, if you are an application builder, right now, sky is the limit for you.
15:25Sky is the limit.
15:27And AI is not future, AI is present.
15:29AI is here, it is in the present.
15:32These things are not, you know, someone saying,
15:34Oh my God, in 10 years, this is what is going to happen.
15:37This is happening right now.
15:39Alright, Dr. Sikha, really appreciate your time.
15:42Thank you for joining us.
15:43Thank you so much.
Comments