00:00We have often seen, and this is a thought that you have all, you know, pretty openly shared, that AI
00:06is powerful, but not intelligent.
00:10When we make that distinction, and there are conversations around LLM, where do you see intelligence and AI-driven power?
00:19Yeah, I think there's a lot of confusion, really, because we tend to anthropomorphize systems that can reproduce certain human
00:29functions.
00:30So what's, I mean, LLMs are incredibly useful.
00:34There's no question about that.
00:36And they do amplify human intelligence, like computer technology, going back to the 1940s.
00:43But LLMs, to some extent, except for a few domains, are mostly information retrieval systems.
00:50They can compress a lot of factual knowledge that has been previously produced by humans and can give easy access
00:57to it.
00:58In a way, it's kind of a natural evolution of, you know, the printing press, the libraries, the internet, and
01:04search engines, right?
01:06It's just a more efficient way to access information.
01:09And there are a few domains where the intelligent capabilities of those systems actually is more than that.
01:18It's more than just retrieval.
01:20So for generating code, maybe doing some type of mathematics, you know, we're getting the impression that it's beyond this.
01:29But it's still, to a large extent, domains where reasoning has to do with manipulating symbols.
01:36The problem is that, you know, why do we have systems that can pass the bar exam and win mathematics
01:43olympiads, but we don't have domestic robots?
01:45We don't even have self-driving cars.
01:47And we certainly do not have self-driving cars that can teach themselves to drive in 20 hours of practice
01:53like any 17-year-old.
01:54So we're missing something big still.
01:57So what are we teaching a 17-year-old then?
02:00Well, so the question is, how does a baby learn?
02:05Or even an animal, right?
02:07Animals have a much better understanding of the physical world than any AI systems that we have today, which is
02:13why, you know, we don't have smart robots.
02:16And so, you know, we learn about the world, how the world works, mostly by observation when we are babies,
02:25a few months old.
02:26And then we learn by interaction.
02:28And we learn mental models of the world that allows us to apprehend any new situation, even if we haven't
02:35been, you know, exposed to it beforehand, we can still handle it.
02:39So a big buzzword in AI today is world models.
02:43And this is really this idea that we develop mental models of the world that allow us to think ahead,
02:53to apprehend new situations, plan sequence of actions, reason, and predict the consequences of our actions, which is absolutely critical.
03:02And LLMs don't do this, really.
03:03There is the sense, Professor, that perhaps AI will unlock an era of radical abundance.
03:14Will this abundance benefit us?
03:17Well, so, you know, if you talk to economists, they tell you, you know, if we can measure the improvement
03:27that AI will bring to productivity, which is, you know, amount of wealth produced per hour worked.
03:33It's going to add up to maybe 0.6% per year.
03:37Okay, this is from economists that actually have studied the effect of technological revolutions on the labor market and the
03:44economy.
03:45People like Philippe Aguillon or Eric Brinovson.
03:50And so that seems small.
03:52It's actually quite big.
03:53And, you know, it's certainly going to accelerate scientific progress, progress in medicine.
03:58I do not believe there's going to be, like, a singular identifiable point where the economy is going to take
04:06off and there's going to be abundance.
04:08And there's also the question of the, you know, policies surrounding this, you know, are those benefits going to be
04:14shared across humanity or, you know, different categories of people in various countries?
04:20That's a political question.
04:21It has nothing to do with technology.
04:22So if economists see this as boom, will then openness survive?
04:30It's not going to be an event.
04:32It's going to be progressive.
04:35There is this false idea that somehow, at some point, we're going to discover the secret of, you know, human
04:44-level intelligence.
04:44I don't like the phrase AGI because human intelligence is specialized, so I don't like the, you know, artificial general
04:50intelligence phrase.
04:52But it's not going to be an event.
04:54We're not going to discover one secret.
04:55We're going to make continuous progress, and we're not going to be able to measure that progress by just having
05:04a series of tests that are going to test, you know, whether a machine is more intelligent than humans.
05:09Because machines are already more intelligent than humans on a large number, a growing number of narrow tasks.
05:17And so, you know, it's not like a uniform, you know, scalar measurement of quality.
05:24It's a collection of quality.
05:26But what's more important is that intelligence is not just a collection of skills.
05:32It's an ability to learn new skills extremely quickly and even to accomplish new tasks without being trained to do
05:40it the first time we apprehend it.
05:42That's really what, you know, intelligence should be measured at.
05:48So we're not going to be able to just design a test that is going to figure out, you know,
05:52are machines more intelligent than humans?
05:53So if it's about upskilling and ensuring that you're relevant, then only perhaps you're intelligent, will that then mean that
06:01the countries that adopt AI and the, you know, the pace at which India and the scale at which India
06:09has adopted AI, the challenge would be to create talent which is upskilled and reskilled and have the required skills
06:18for this?
06:20Absolutely.
06:20So the relationship that we're going to have with intelligent AI systems is going to be similar to the relationship
06:29that a leader in business politics or academia or some other domain has with their staff.
06:40Okay.
06:41AI is going to be our staff.
06:42Every one of us is going to be a manager of a staff of, you know, intelligent machines.
06:49They'll do our bidding.
06:50They might be smarter than us, but, you know, certainly if you are an academic or a politician, you work
06:56with staff that are smarter than you.
06:58In fact, that's the whole point.
06:59You need to, you know, attract people who are smarter than you because that's what makes you more productive, you
07:05know, for an academic is students who are smarter than their professor and teach, you know, it's not the professor
07:11that teaches graduate students.
07:13It's the other way around, actually.
07:14And certainly we have a lot of examples of politicians who are surrounded by people who are smarter than them.
Comments