Кого ИИ сделает богаче, а кого оставит без работы | Нобелевский лауреат по экономике — Transcript
Full transcript
- 0:01I can easily imagine a grim future where
- 0:03many people will not have meaningful
- 0:05work.
- 0:06>> This is Daron Acemoglu, an economist
- 0:09from MIT and a 2024 Nobel Prize winner
- 0:11in Economics. He believes we are missing
- 0:14the main opportunities offered by
- 0:15artificial intelligence. What should be
- 0:17done? How can we remain valuable? I
- 0:19constantly see posts. Instead of hiring
- 0:22a person, I launched an AI agent.
- 0:24>> Automation is great, but we shouldn't
- 0:26focus solely on it.
- 0:28Otherwise, we will destroy too many jobs
- 0:30without creating new, meaningful,
- 0:32high-paying, productive ones that open
- 0:34up opportunities for people.
- 0:36>> So, what should people do? What skills
- 0:38should they develop?
- 0:39>> If you want to know my opinion, go and
- 0:42learn.
- 0:46>> In 2024, you said that within 10 years,
- 0:48AI would add about 1%to the US GDP, and
- 0:50only 5%of tasks would be fully automated
- 0:53.
- 0:54What do you think about that now, in
- 0:552026?
- 0:57First of all, I am very glad to be here
- 1:00.
- 1:01And I want to say right away that I do
- 1:03not claim to have any unique
- 1:05understanding of artificial intelligence
- 1:07.
- 1:08Artificial intelligence is a huge topic
- 1:10.
- 1:11Everyone looks at it from their own
- 1:13perspective and sees only a part of the
- 1:14bigger picture.
- 1:16I try to bring some basic economic logic
- 1:19into the discussion about productivity.
- 1:22I made those calculations more than 2
- 1:24years ago.
- 1:25There was already a lot of uncertainty
- 1:27back then, and today there is even more
- 1:30. But there have been very rapid changes
- 1:32, especially this year with the
- 1:34emergence of agentic AI.
- 1:36Perhaps today I would name a figure
- 1:38slightly higher than 5%.
- 1:41And yet, I still think most
- 1:43conversations exaggerate the speed at
- 1:45which we will see automation.
- 1:48Because automation is complicated.
- 1:51Integrating it into the work of
- 1:52organizations is difficult.
- 1:55The diffusion of new technologies,
- 1:57especially when they require large
- 1:59additional investments and serious
- 2:01changes, is a difficult and slow process
- 2:03.
- 2:04This does not mean that we are not
- 2:06looking at a technology capable of
- 2:07changing the world.
- 2:09And it does not mean that we cannot
- 2:11benefit from it. But when discussing how
- 2:14quickly and how deeply the economy will
- 2:17change, we must remain realistic.
- 2:20And we still need to discuss the same
- 2:23questions that arise with any new
- 2:24technology.
- 2:26How can it be used best? How can it be
- 2:30abused? And who will ultimately benefit
- 2:33?
- 2:33>> You mentioned AI agents. What would have
- 2:36to happen for you to say, "Now I am
- 2:38ready to raise my estimate of AI's
- 2:40impact to 10%or higher"?
- 2:42>> In programming, progress is indeed very
- 2:46rapid.
- 2:48Why is artificial intelligence
- 2:50developing faster here than in other
- 2:52professions?
- 2:54First, programming happens in an office
- 2:56.
- 2:57There is no need to interact with the
- 2:58physical world.
- 3:01The people who verify the code work are
- 3:03well-versed in both artificial
- 3:05intelligence and technology in general.
- 3:08But there is one more very important
- 3:11thing.
- 3:12In programming, like in mathematics or
- 3:14chess, there are clear rules.
- 3:17A program either works, or it doesn't.
- 3:21Reinforcement learning is especially
- 3:24effective when there is an unambiguous
- 3:26criterion that lets you say: "No, that's
- 3:29wrong."
- 3:30In many professions, there is no such
- 3:32criterion.
- 3:34Take customer support, for example.
- 3:37Most people probably expected that
- 3:39artificial intelligence would be the
- 3:41first to fully automate it.
- 3:44Something is already happening, but the
- 3:46process has proven to be very slow and
- 3:48difficult.
- 3:50In many companies I deal with, the
- 3:52artificial intelligence works just
- 3:53terribly.
- 3:55You call, you get a bot, and you think:
- 3:58"God, another 10 minutes of waiting
- 4:00until they connect me to an operator who
- 4:03can help, because dealing with people is
- 4:05what matters here.
- 4:07Their requests change over time and
- 4:09depend on the individual."
- 4:12People ask the same question in
- 4:14completely different ways.
- 4:16Plus a multitude of ambiguous and
- 4:19non-standard cases.
- 4:21All of this is much more complex.
- 4:23People must either control artificial
- 4:26intelligence or work alongside it.
- 4:29Ideally, we need to establish a
- 4:31cooperation where artificial
- 4:32intelligence complements the human.
- 4:36All these options are possible, and
- 4:38which one we choose will largely
- 4:40determine how many jobs disappear, how
- 4:43much productivity grows, and who
- 4:45benefits.
- 4:47>> In programming, the result is
- 4:48unambiguous and easy to check, while in
- 4:50support, as you said, everything is much
- 4:52more individualized. Can this problem be
- 4:54solved by a stronger model, or is there
- 4:56a fundamental limitation of artificial
- 4:58intelligence meaning it cannot work
- 5:00without a human?
- 5:02I definitely wouldn't say there's
- 5:04anything artificial intelligence will
- 5:07never be able to do.
- 5:12Over the last 10 years, we've seen
- 5:13enough examples.
- 5:16If you invest enough resources and data
- 5:18, artificial intelligence can advance in
- 5:20many things.
- 5:23The question is how quickly and in which
- 5:25areas.
- 5:27If we look broader and ask why
- 5:30productivity is growing so weakly and
- 5:33what to do about it, my answer fits in
- 5:36one word: application. We need more
- 5:40applications, and they must be much
- 5:42better.
- 5:44One could even assume, though I don't
- 5:46know for sure and I might be wrong,
- 5:48since I don't work at Anthropic or
- 5:50OpenAI.
- 5:53I don't see the results of their
- 5:55experiments, but they have huge
- 5:57resources and very talented people.
- 6:00But one could say that America and
- 6:02Silicon Valley, in particular, have
- 6:04focused too much on base models and too
- 6:06little on applications and their
- 6:08implementation in companies.
- 6:11We are constantly told that America is
- 6:13ahead of China in large language models
- 6:15. And that, by all appearances, is true
- 6:17. Chinese models sometimes catch up, but
- 6:20largely at the expense of American ones
- 6:23. Most likely, they are partly learning
- 6:25from their responses.
- 6:27However, Chinese companies are doing
- 6:29something I see much less of in the USA
- 6:32.
- 6:33They are actively integrating robots and
- 6:36artificial intelligence into production
- 6:38and employing a huge number of engineers
- 6:41to do it.
- 6:42And it seems the U.S. just doesn't have
- 6:44that many engineers.
- 6:46And all that are available are going
- 6:48into model development.
- 6:49>> You've brought up a very interesting
- 6:50topic.
- 6:52I'm an immigrant, and when I see how
- 6:53things are set up in the U.S. and ask
- 6:54why, they tell me: "Well, that's just
- 6:56how it’s always been."
- 6:58>> Yes, yes. Or that there’s simply no
- 7:01other way.
- 7:04Here is something else I would like to
- 7:05say.
- 7:07I don’t want to overgeneralize, but in
- 7:09my view, the American AI industry is
- 7:12harmed by a belief in inevitability.
- 7:16The conviction that our way is the only
- 7:19possible one.
- 7:20And if you have doubts, then you're an
- 7:23idiot.
- 7:26I categorically disagree with that.
- 7:29Artificial intelligence is a very
- 7:31flexible technology, much like other
- 7:33general-purpose technologies.
- 7:35That’s what economists call, for example
- 7:38, electricity or the printing press—
- 7:40technologies that serve as a foundation
- 7:42for many other solutions.
- 7:45Both history and economic research show
- 7:47that they can be developed in completely
- 7:49different ways.
- 7:51And these paths are not equivalent; they
- 7:54yield different benefits and come with
- 7:57different costs.
- 7:59Therefore, believing that our way is the
- 8:01only one or inherently the best is, in
- 8:03my opinion, a mistake.
- 8:05>> And I think there is another problem
- 8:07here. How do you measure it? How do you
- 8:09assess the impact of AI on your personal
- 8:11productivity or on a company's
- 8:12productivity?
- 8:14We spoke with Erik Brynjolfsson in May.
- 8:16According to their research, the
- 8:18economic impact of AI has been virtually
- 8:21zero.
- 8:22The capabilities of models are growing
- 8:25rapidly, but the economic impact remains
- 8:27quite modest.
- 8:28And this gap between capabilities and
- 8:31reality, in my view, opens up huge
- 8:33prospects.
- 8:34>> Do you think the problem also lies in
- 8:36how we measure it?
- 8:37>> Erik is a very serious researcher.
- 8:40He knows this industry perhaps better
- 8:42than any other social scientist, and I
- 8:45think he is more optimistic than I am.
- 8:48If you asked him about the impact of AI
- 8:50on productivity by 2034-2035
- 8:54—which is exactly what I was writing
- 8:56about then—he would give a bolder
- 8:58estimate. But on the whole, I think his
- 9:00conclusion is: AI hasn't yet delivered
- 9:03the productivity growth we expect, but
- 9:05it will. My position is roughly the same
- 9:07. It’s not visible yet, but we will see
- 9:10it. It’s just, in my view, it will
- 9:12happen a little later than Erik thinks.
- 9:14I will be very glad if he turns out to
- 9:16be right.
- 9:17And he has a much better chance of being
- 9:19right than many people in the industry
- 9:21who view what is happening with
- 9:23excessive optimism.
- 9:25>> And how, in your opinion, should one
- 9:27properly measure the impact on a company
- 9:29or on personal productivity?
- 9:30>> Productivity has always been difficult
- 9:33to measure.
- 9:34And one of the problems is quality.
- 9:39Erik rightly emphasizes: "In the digital
- 9:42age, over the last 40 years, we still
- 9:44haven't learned how to measure quality.
- 9:47"
- 9:48Perhaps that is why we underestimate the
- 9:50benefits not only of artificial
- 9:52intelligence, but of computers in
- 9:54general.
- 9:55I agree with that.
- 9:58But we never really knew how to properly
- 10:00measure improvements in car or road
- 10:02quality before.
- 10:04So this problem has always existed.
- 10:06Maybe happiness levels?
- 10:08>> Yes,
- 10:09>> we should simply measure happiness
- 10:10levels and see how they change.
- 10:12>> That is also a good idea.
- 10:14Well-being is very important, but here I
- 10:17would rate AI and related technologies
- 10:20rather negatively.
- 10:23In my opinion, social media has caused
- 10:25us immense harm.
- 10:27It destroyed community and weakened our
- 10:29ability to cope with challenges because
- 10:32we stopped truly communicating with
- 10:35people.
- 10:36It brought out our worst instincts
- 10:39because algorithms bet on anger, emotion
- 10:42, and envy.
- 10:44So, I think it's terrible.
- 10:48I hope it will be different with AI.
- 10:51>> Let's talk about how to ensure that
- 10:53workers also benefit from new technology
- 10:55.
- 10:56>> My favorite topic.
- 10:58>> Yes, you talk about that a lot. You said
- 11:00the models are strong, but there aren't
- 11:02enough applications. What is the problem
- 11:03? Is it education, or do we need even
- 11:05more powerful models? What needs to
- 11:07happen for artificial intelligence to
- 11:09positively impact our daily lives?
- 11:12>> I think current models are already
- 11:14strong enough to build applications on
- 11:16top of them.
- 11:18You can do a lot with existing
- 11:20technology.
- 11:23Let me give a few examples.
- 11:26You can significantly improve quality
- 11:28control. With current models, it’s not
- 11:31difficult. Tools for nurses.
- 11:35Our country is in a real healthcare
- 11:38crisis. Almost every fifth dollar goes
- 11:41to medicine.
- 11:43And look at the results.
- 11:45Serious public health issues, many
- 11:48limitations within the system.
- 11:51Imagine if nurses could be more involved
- 11:54in treatment, helping with diagnostics,
- 11:56prescriptions, and patient condition
- 11:59decisions.
- 12:01One of the main barriers here is access
- 12:03to information and quality control.
- 12:07All this can be provided with existing
- 12:09models if you create suitable
- 12:11applications based on them.
- 12:13Yes, nurses will probably need to be
- 12:15trained a little differently.
- 12:17So, investment in human training will be
- 12:20needed in parallel.
- 12:22But the main obstacle is not training
- 12:24nurses. The main obstacle is the lack of
- 12:27applications and the fact that tech
- 12:29companies aren't building them. It’s the
- 12:32same with teachers.
- 12:34Electricians could also be given much
- 12:37more convenient tools. This is an
- 12:39obvious opportunity that is easy to
- 12:41capitalize on.
- 12:43In many trade professions, AI can
- 12:45noticeably increase efficiency.
- 12:48We are already seeing some examples in
- 12:50China.
- 12:51Why isn't this being done in the US?
- 12:53It’s not about the base models.
- 12:56For most such solutions, last year's
- 12:58models would suffice.
- 13:00>> Sounds like a great business opportunity
- 13:02.
- 13:02>> Yes, exactly.
- 13:05>> And for electricians, it’s a huge market
- 13:07. Only after moving here did I learn how
- 13:09much they earn and how much you have to
- 13:12pay for any little thing.
- 13:14>> And there is such a shortage of
- 13:15electricians that you have to wait 3
- 13:17weeks for even simple work.
- 13:18>> Yeah, yeah. Then the electrician comes
- 13:21and says, "No, I can't handle this new
- 13:23equipment."
- 13:24Why? Because I hadn't seen it before.
- 13:26>> That's where artificial intelligence
- 13:27comes in handy.
- 13:28>> Exactly. That's a very simple task for
- 13:31AI.
- 13:32>> And in medicine and education, is it a
- 13:33state task, or are there opportunities
- 13:35for companies there too?
- 13:36>> In medicine and education, two or even
- 13:39three parties must interact.
- 13:42Most healthcare spending is somehow tied
- 13:44to the state.
- 13:46It’s the same with education.
- 13:49So yes, the state can play an important
- 13:51role, but electricians, laborers,
- 13:53support staff, and scientists—that's
- 13:56primarily the private sector.
- 13:59Why did I mention three parties? Even
- 14:02without the state, there are two: tech
- 14:04companies and the business leadership
- 14:07itself. And both must participate.
- 14:10When I talk to CEOs and top managers of
- 14:12large companies about AI that will help
- 14:15their employees and make them more
- 14:17productive, they tell me: "That's
- 14:19exactly what we need.""We are all for it
- 14:21."
- 14:22Are they being sincere when they say
- 14:24that? I don't know. I can't verify it,
- 14:26because tech companies don't offer such
- 14:29tools yet.
- 14:32And when I talk about this to people
- 14:33from tech companies, they either nod and
- 14:35look away, or they reply: "You live in a
- 14:37different world."
- 14:39"We are creating AGI."
- 14:41"What are you even talking about?"
- 14:43>> They care about the distant future, not
- 14:44about what can be improved right now. We
- 14:46were talking about niches where routine
- 14:47tasks haven't been automated by anyone
- 14:49yet. For a long time, such a routine in
- 14:51my team was onboarding contractors. As
- 14:52you know, we have a distributed team all
- 14:54over the world. People work from
- 14:56different countries. With everyone, you
- 14:57need a contract and closing documents.
- 14:59And all of this must be stored so that
- 15:01before an audit or a meeting with an
- 15:02investor, you aren't scrambling to piece
- 15:03it together. You can onboard contractors
- 15:05in different ways. You can do it
- 15:06directly. Then the company manages every
- 15:08contract, background check, and document
- 15:10itself. You can use tools where
- 15:11contracts are just stored, but you
- 15:13remain the party to the contract. We
- 15:15have been working with Fodev.com for
- 15:16several years. The platform becomes a
- 15:18party to the contract with the
- 15:20contractor and manages their paperwork.
- 15:22With Fodev.com, documents for the remote
- 15:24team are generated in a unified format.
- 15:26Contractors can be hired from more than
- 15:28150 countries. And the rights to the
- 15:30work result can be secured immediately
- 15:31upon onboarding, if the business needs
- 15:33it. Contracts, closing documents, and
- 15:34the status of all people are stored in
- 15:36one place, and they can be shown at any
- 15:38moment.
- 15:39If you have a distributed team of more
- 15:40than twenty people, sign up for a demo
- 15:42at Fodev.com using the link in the
- 15:43description.
- 15:44They will show you how to bring order to
- 15:46your contracting process. Now, let's get
- 15:49back to our conversation. You mentioned
- 15:51solutions for manual labor and China.
- 15:53What are some good examples there?
- 15:55>> I’ve spoken with several manufacturing
- 15:57companies in China.
- 16:00Foxconn, BYD, and other electric vehicle
- 16:03manufacturers.
- 16:05They are all actively implementing
- 16:07robots and
- 16:10They are technologically very advanced,
- 16:12but their experience shows that the
- 16:14process is slow and always requires more
- 16:17engineers than expected.
- 16:20Fortunately for them, there is a huge
- 16:22number of engineers in China.
- 16:25What advice would you give to
- 16:26entrepreneurs watching us who are
- 16:28choosing what to do? How can one find a
- 16:29suitable niche right now? Lately, on
- 16:31podcasts, I’m constantly told: "The
- 16:33money is in small niches, especially
- 16:35with artificial intelligence."
- 16:36Opportunities exist in boring businesses
- 16:38that no one has automated for 20 years
- 16:40and where everything is still recorded
- 16:42on paper. You go there, solve the
- 16:44problem, and there’s the opportunity.
- 16:45>> Absolutely. I just wouldn’t call it
- 16:48boring. In my opinion, it’s very
- 16:50interesting. Imagine,
- 16:51>> boring from the outside. Does AI sound
- 16:53more? exciting.
- 16:54>> I understand, but that’s a completely
- 16:57wrong perspective. First of all, AGI
- 17:00remains a dream for now. No one can even
- 17:03agree on its definition. It is always
- 17:05somewhere in the future. And honestly,
- 17:08we can't really imagine that future.
- 17:12If by AGI, and later ASI, artificial
- 17:14superintelligence, we mean machines that
- 17:17can do absolutely everything humans can
- 17:20do, and do it better, then that seems to
- 17:23be a long way off.
- 17:24And that sounds like a dystopia. Then
- 17:27there will be no work left even for the
- 17:30best physicists, mathematicians, or
- 17:32communication specialists, as everything
- 17:34will be done better.
- 17:36There are many contradictions in public
- 17:38discussions about AGI. People say: "AGI
- 17:41is wonderful, it will appear soon, but
- 17:43people will still have things to do."
- 17:46That’s not how it works. If you
- 17:48seriously mean AGI or AI, there won’t be
- 17:51many tasks left for humans.
- 17:54>> I want to talk more about practical
- 17:55applications. And today you mentioned
- 17:58opportunities for researchers. How do
- 18:01you use it yourself in your work, and
- 18:03what else could be done?
- 18:05>> I must admit, I’m not at the forefront
- 18:09here.
- 18:10I use artificial intelligence to gather
- 18:13information on a topic, and this year
- 18:15I’ve been asking it more and more to
- 18:17critique my texts and ideas.
- 18:21And it’s very interesting.
- 18:24When I tried this with ChatGPT 3.5, and
- 18:27later with ChatGPT 4, the results
- 18:30greatly disappointed me.
- 18:32Now everything is much better.
- 18:36I advise all my students and research
- 18:39assistants to build a system of AI-based
- 18:42tools for research.
- 18:45Specifically to help with work, not to
- 18:47replace oneself.
- 18:50I don’t advise letting artificial
- 18:52intelligence write for you, even if
- 18:54English is not your native language.
- 18:57A person must be responsible for their
- 18:59own text.
- 19:01However, for students for whom English
- 19:03is not their native language, I
- 19:04recommend using artificial intelligence
- 19:06as a proofreader.
- 19:08Check the text for phrases that are
- 19:11unclear, too long, or grammatically
- 19:13incorrect.
- 19:15For such daily tasks, artificial
- 19:17intelligence is already very good.
- 19:20But I am worried about a different loss
- 19:22of control.
- 19:25We all, especially students and those
- 19:27who are learning, delegate too much to
- 19:29artificial intelligence.
- 19:31As a result, we don't train our thinking
- 19:33and don't engage in the learning process
- 19:35.
- 19:36In my opinion, this is one of the main
- 19:39problems.
- 19:40I see it in students.
- 19:43And studies show that in school
- 19:45education, the use of artificial
- 19:48intelligence has led to very serious
- 19:50problems. We introduced it blindly.
- 19:53Without any protective measures.
- 19:56I was talking about this back in 2024 or
- 19:59even 2023.
- 20:02Simultaneously unleashing generative
- 20:04artificial intelligence into all schools
- 20:07without any restrictions is a huge and
- 20:09dangerous social experiment.
- 20:12We didn't know what it would lead to.
- 20:14I can't say my worst fears have fully
- 20:17materialized, but we've ended up quite
- 20:20close to that. And how do you know when
- 20:22it's time to tell yourself: "That's it,
- 20:24I need to think for myself here." And
- 20:25only use artificial intelligence as a
- 20:27helper?
- 20:28>> For me, things are a bit different.
- 20:30Firstly, unfortunately, I am from a
- 20:32different generation. And secondly, I
- 20:34just love to think. I like to take a pen
- 20:36, paper, and sketch something out.
- 20:39So, I actually have to remind myself:
- 20:41try checking this with AI, for instance
- 20:44, I've never used AI for math.
- 20:46Although I should, it is much stronger
- 20:48than me in that, but I like solving
- 20:50things myself. When I write down models
- 20:53or do calculations, I do it by hand. I'm
- 20:56used to it, I'm fast at it, and in the
- 20:59process, I learn, including from my own
- 21:01mistakes.
- 21:03But looking from the outside, I could
- 21:05probably use AI more.
- 21:07>> And when you calculate yourself, don't
- 21:08you think about the history of
- 21:09calculators? Teachers were against them
- 21:11, but now we use them everywhere and in
- 21:13our spare time solve much more complex
- 21:15problems.
- 21:16AI might give us the same leap forward,
- 21:18or conversely, take away our critical
- 21:20thinking and leave us stuck in place.
- 21:23I'm worried about the latter.
- 21:26The comparison with the calculator is
- 21:28useful, but it must be made carefully.
- 21:31A calculator took on only a very small
- 21:33part of the work.
- 21:35I don't know anyone it did the thinking
- 21:37for, and I don't know a single school
- 21:39that said, "Now you have a calculator,
- 21:42so there's no need to learn
- 21:43multiplication."
- 21:45When I was studying, calculators already
- 21:48existed, but we all learned how to
- 21:50multiply and divide.
- 21:52I can't imagine how one could not know
- 21:54how to do that. It is the foundation for
- 21:56so many things.
- 21:58The same goes for derivatives. If you
- 22:01don't know calculus, how can you even
- 22:04take the next step?
- 22:06>> Yes, a calculator performs one specific
- 22:08function, while AI takes on the entire
- 22:10thought process,
- 22:11>> or a very large part of it.
- 22:14And the lines are blurred.
- 22:16Where should you stop? How do you work
- 22:19with artificial intelligence while still
- 22:21going through the key stages yourself?
- 22:24Coming up with something new, finding
- 22:26creative solutions.
- 22:27>> A short pause. Naron says we should
- 22:30delegate specific functions to AI, while
- 22:32keeping critical and creative thinking
- 22:34for ourselves. But how do you know
- 22:36exactly what to delegate?
- 22:38I compiled a list of thirty of my
- 22:40routine tasks—from sorting through
- 22:41hundreds of school emails and tracking
- 22:43my lab results to a dozen other mundane
- 22:45things I’ve handed off to AI.
- 22:49Thanks to this, I’ve gained time for
- 22:50much more important tasks and for my
- 22:52kids.
- 22:53This list is in my free newsletter.
- 22:55There, I share weekly the tools I
- 22:57personally use in my work and daily life
- 22:59. What I’ve tried, what I’ve noted, and
- 23:01what I’ve been disappointed in.
- 23:03Subscribe, the link is in the
- 23:04description. The newsletter is
- 23:05completely free. Let’s get back to their
- 23:08business. On X, I constantly see posts
- 23:10saying, "Instead of hiring a person, I
- 23:13launched an agent." What are such
- 23:15companies and people missing in the long
- 23:17run? I know you are against selling
- 23:19agents as a replacement for humans.
- 23:21>> Let me clarify my position.
- 23:24It is based on my research into robots,
- 23:2620 th-century technology, the Industrial
- 23:29Revolution, and theoretical works.
- 23:32I’m not against automation, and I don’t
- 23:34think it can be stopped.
- 23:36We need it, and I’m against trying to
- 23:40stop it. We certainly don’t want to go
- 23:42back to the days when people carried
- 23:44heavy loads on their backs.
- 23:46Automation is great. The problem is that
- 23:50we are focusing only on that.
- 23:52Firstly, we won’t get all the possible
- 23:55growth in productivity that way.
- 23:57Automating everything in sight isn't
- 23:59that simple. Secondly, we won’t achieve
- 24:02prosperity for everyone.
- 24:05Too many jobs will disappear, and in
- 24:07return, no new, meaningful, high-paying
- 24:10, productive roles that open doors for
- 24:13people will appear.
- 24:15That’s why I’ve always said: we need a
- 24:18balance.
- 24:19Use new technologies for both automation
- 24:22and other tasks.
- 24:24Digital technologies gave us many
- 24:26opportunities to create new occupations
- 24:28for people.
- 24:30And this applies twice as much to AI.
- 24:33I’m for AI-driven automation, provided
- 24:36that in parallel we use it to create new
- 24:39opportunities for nurses, electricians,
- 24:42and many professions we can't even
- 24:45imagine today.
- 24:47>> How do you explain this to a small
- 24:49business owner whose main goal is profit
- 24:51growth? Why is it bad from an economic
- 24:53standpoint?
- 24:53>> In productivity.
- 24:56In my experience, most companies
- 24:57understand this, although they might
- 24:59just be being polite to me.
- 25:01Or maybe I convinced them? Not all, but
- 25:04most understand that employees are your
- 25:07primary asset. Imagine if they could
- 25:09work more productively. For a business,
- 25:11that’s wonderful, and by the numbers,
- 25:14it’s more profitable. Let’s say you have
- 25:16100 employees.
- 25:18I offer a choice: automate 5%of the jobs
- 25:21and cut costs for them in half, or
- 25:23increase the productivity of all
- 25:25employees by 5%.
- 25:28The second option will bring in much
- 25:30more. they will automate the work. If
- 25:32you get rid of people, your productivity
- 25:34will ultimately decline.
- 25:36>> Exactly.
- 25:37Employees have a lot of experience, and
- 25:40many, if you have been a good manager,
- 25:42are also truly loyal to the company.
- 25:46If you help them become more inventive,
- 25:49better at solving problems, and better
- 25:51at doing new things, such a business
- 25:53will be successful.
- 25:55Let me ask you, which companies from the
- 25:58past do you remember?
- 26:00You will surely name familiar names:
- 26:03Ford, General Motors, General Electric,
- 26:07IBM, Microsoft; what unites them all?
- 26:10They did something new. They didn't just
- 26:13cut costs and repeat after others.
- 26:16They invented new markets, technologies
- 26:18, and tasks.
- 26:20Henry Ford put it this way: "If I had
- 26:22asked customers what they wanted, they
- 26:25would have asked for faster horses."
- 26:28That is what we expect from successful
- 26:30companies. And cutting costs by 5%is not
- 26:32enough to go down in history.
- 26:35>> You said that 5%of tasks can be
- 26:37automated. But how do you understand
- 26:39which ones are worth automating and
- 26:40which are not? Do you have any kind of
- 26:42rule?
- 26:43>> If you cut costs by 5%, then GDP usually
- 26:46grows by those 5%multiplied by your
- 26:49company's share of the GDP.
- 26:52Everything is simple here. Complications
- 26:54begin if the quality changes.
- 26:57And even to cut costs by 5%over several
- 27:00years, you need to work very hard. I see
- 27:03how some companies I know rush into
- 27:06automation.
- 27:08Some think it will be easy, others feel
- 27:10the pressure, and then nothing works out
- 27:13because automation is a difficult task.
- 27:16It needs to be integrated into the
- 27:18company's work. Many are saying now that
- 27:20AI has a jagged frontier of capabilities
- 27:23. It is strong in some tasks, but
- 27:25unexpectedly fails in others. I like
- 27:27this image, but in essence, it is a
- 27:30polite way of saying: "We overestimated
- 27:32it."
- 27:33AI turned out to be much more complex,
- 27:35and we thought we could cut staff thanks
- 27:37to AI, but that is not the case.
- 27:40And if you think about it seriously, it
- 27:42turns out that I will have about the
- 27:44same number of employees left.
- 27:46AI is added, and tasks are mixed between
- 27:48them.
- 27:49>> Yes, there is only more work. Let's move
- 27:52on to education.
- 27:53You work at a university.
- 27:55Will higher education still be worth the
- 27:58money in 10 years?
- 27:59>> I think so.
- 28:02But with a few caveats. Firstly, I have
- 28:05never believed that everyone should go
- 28:08to university.
- 28:10And now, for economic and social reasons
- 28:13, I am even more convinced of this.
- 28:16People have different abilities and
- 28:18interests.
- 28:20You need to learn your whole life, but a
- 28:22broad university education does not suit
- 28:24everyone.
- 28:26And we don't need everyone in society to
- 28:28look at the world in the same way.
- 28:30Secondly, a university diploma is, in
- 28:33many ways, also a signal to others. By
- 28:37enrolling at Harvard, MIT, or Stanford,
- 28:39you are buying the name, among other
- 28:41things.
- 28:42That is good for you, but not
- 28:45necessarily for society.
- 28:47Third, all universities now realize they
- 28:50need to change their business model and
- 28:53the way they teach.
- 28:55They are searching for which skills to
- 28:58focus on. From my rather superficial
- 29:00point of view, universities realize this
- 29:03. But, you know, the same applies to
- 29:06schools.
- 29:07And they don't seem to understand it
- 29:09there. Schools are lagging even further
- 29:11behind.
- 29:13In the US, I see that many students
- 29:15arrive at university poorly prepared.
- 29:18The situation will worsen, and it will
- 29:20become even harder for universities.
- 29:24I wrote an economics textbook for
- 29:25freshmen and I see how often they lack
- 29:27basic knowledge, and it’s getting worse
- 29:30over the years.
- 29:32This started even before the arrival of
- 29:34artificial intelligence.
- 29:35Universities are having to do more and
- 29:38more of what should have been taught in
- 29:40school because schools are failing. We
- 29:43are still at the very beginning;
- 29:45everything could still change.
- 29:47But the available data shows that
- 29:49artificial intelligence has complicated
- 29:51learning in middle and high school.
- 29:53This means this problem will only
- 29:56intensify.
- 29:57>> Yes, if AI solves all the problems, it
- 29:59will only be harder at university. So
- 30:01what should be done? Which skills should
- 30:02be developed? And do they require even
- 30:04more academic knowledge? I am thinking,
- 30:06for example, about taste and strategic
- 30:08thinking.
- 30:09>> Absolutely. I do not believe that true
- 30:12AGI will appear any day now.
- 30:15Therefore, human creativity is needed
- 30:17more than ever, and creativity without a
- 30:19solid foundation is simply impossible.
- 30:24If you want to know my opinion—though
- 30:26I’m not the best person for such advice
- 30:29, I’m not a mathematician or a computer
- 30:32scientist—go study mathematics.
- 30:34Don't be discouraged by the fact that AI
- 30:37is already solving complex mathematical
- 30:39problems. We need human knowledge, a
- 30:42deep understanding of mathematics. This
- 30:45will be very important.
- 30:47Study physics, economics, engineering.
- 30:51All of this will be useful.
- 30:53But also learn flexibility. We will
- 30:56really need it. Even while staying in
- 30:59the same profession, we will often
- 31:01change our work following technologies
- 31:03and societal changes. This is truly
- 31:05important.
- 31:08But in the next 10 years, in my opinion
- 31:10, human ingenuity and creativity will
- 31:12play a key role.
- 31:14>> Yes, and it’s time to stop considering a
- 31:15non-linear career something strange.
- 31:17That resonates with me. I have changed
- 31:19my field of activity more than once
- 31:20myself. I think this will soon become
- 31:21the absolute norm.
- 31:22>> And two more principles that, in my
- 31:24opinion, should be instilled starting
- 31:26from middle school.
- 31:28Of course, learning to work with AI, but
- 31:31placing even greater emphasis on civic
- 31:34responsibility.
- 31:37In our country, and indeed almost all
- 31:39over the world, we have lost a real
- 31:41sense of community, care for the common
- 31:44good, and willingness to work for a
- 31:46common cause.
- 31:49In the era of AI, this is especially
- 31:52important, and it all starts with
- 31:54understanding what is happening and
- 31:57participating in society.
- 31:59>> In fact, we need to start as early as
- 32:01elementary school.
- 32:02>> Yes, of course, we can start in
- 32:04elementary school as well.
- 32:06Perhaps we have encouraged selfishness
- 32:09too much, pushing people to retreat into
- 32:11their own cocoons, on social media and
- 32:14beyond. We have devalued the important
- 32:16aspects of community life.
- 32:19This is one of the main themes of my new
- 32:21book about what happened to liberal
- 32:23democracy.
- 32:25At the center of a liberal democratic
- 32:28society, there must be community.
- 32:31I am neither a pessimist nor an optimist
- 32:34, but I am definitely not a pessimist. I
- 32:37try to remain hopeful.
- 32:40And two things give me that hope.
- 32:43First, people are capable of adapting to
- 32:46anything.
- 32:48Second, even in such a divided world,
- 32:51our ideas of justice are very similar.
- 32:55There are shared moral values on both
- 32:57sides of the political spectrum. We just
- 33:00need to articulate them more clearly and
- 33:01stop seeing the other side as the enemy
- 33:03.
- 33:05Even with the deepest disagreements, it
- 33:07is important to remember that we are all
- 33:09in the same boat.
- 33:11Oh, right, we discuss all this, and then
- 33:13Dario says, "There is a 25%chance that
- 33:15things will go very, very wrong with
- 33:17artificial intelligence."
- 33:19What do you think when you hear that?
- 33:21Especially from a person who will soon
- 33:23take his company public and prove that
- 33:25his technology will truly change the
- 33:26world.
- 33:27>> If people in leading AI labs are worried
- 33:30about the consequences of this
- 33:33technology, of course, they should talk
- 33:36about it.
- 33:38I welcome and respect their courage.
- 33:41But I am much more concerned by the loss
- 33:44of societal control over AI than by the
- 33:47existential threat to humanity that
- 33:50Dario and Elon Musk talk about. I mean
- 33:53that ordinary people—workers, students,
- 33:56parents—are losing the ability to make
- 33:59decisions for themselves and manage
- 34:02their own lives.
- 34:05Not necessarily because AI has become
- 34:08superintelligent and turned into an
- 34:11autonomous entity, but partly because of
- 34:14how companies and leading labs develop,
- 34:17use, and market artificial intelligence
- 34:20.
- 34:22This loss of control is what worries me
- 34:25deeply. We must talk about the
- 34:26existential threat to humanity, but we
- 34:29cannot ignore other risks in the process
- 34:31.
- 34:31You rated artificial intelligence on a
- 34:34scale from -10 to +10, and you gave it a
- 34:36-6. Has it gotten worse since then? I
- 34:38believe you said that in 2025.
- 34:41>> No, I don't think so.
- 34:43In 2024 and 2025, I thought that AI
- 34:46could cause great harm, but I didn't yet
- 34:50understand what obstacles would arise in
- 34:53its development path. Progress has been
- 34:56rapid, so the risks have increased as
- 34:59well.
- 35:00But now it is also clearer how much good
- 35:02can be done with AI.
- 35:05We just need to change direction. We
- 35:07need to guide the development of AI
- 35:09ourselves.
- 35:11And its fate cannot be decided by a
- 35:13handful of people, even if they are very
- 35:15smart and largely have the best of
- 35:16intentions. This is too important to
- 35:19leave to just a few people.
- 35:21>> And what can each of us do to achieve
- 35:23this?
- 35:24I think we need to start with
- 35:27information and with how artificial
- 35:29intelligence is being discussed.
- 35:33I’m about to say something risky about
- 35:35those who largely decide what we even
- 35:37learn about the world.
- 35:40But, in my view, the media has failed us
- 35:43the most.
- 35:45The media itself.
- 35:47They have done a very poor job of giving
- 35:49people the information needed for a
- 35:51meaningful conversation about artificial
- 35:53intelligence.
- 35:54For 10 years, there were two extremes.
- 35:57It's changing a little now, but very
- 36:01slowly and unevenly. Either Elon Musk
- 36:04and his ilk could do no wrong, and any
- 36:06criticism only elevated them.
- 36:09And everything related to artificial
- 36:11intelligence was delightful, amazing,
- 36:13and wonderful.
- 36:15Or artificial intelligence was portrayed
- 36:18as an enemy, either as a parrot just
- 36:21repeating words or as a killer robot.
- 36:24There was no room for a balanced
- 36:26conversation so people could understand
- 36:29, get involved, and influence our future
- 36:32as consumers and citizens of a
- 36:34democratic society.
- 36:36The media didn't create that space, and
- 36:38this applies to almost every news
- 36:40organization.
- 36:41>> Let's finish with a few practical
- 36:43questions. I want the audience to be
- 36:45left with a clear plan of action. What
- 36:47do you do with your savings when you
- 36:48hear this news? Or has nothing changed
- 36:50for you?
- 36:50>> Almost nothing. I keep a balanced
- 36:53portfolio. Beating the market is very
- 36:55difficult.
- 36:56To do that, you either need to be an
- 36:59expert in investing, spend a lot of time
- 37:01, or have insider information. I don’t
- 37:04have that much time. I don’t have
- 37:06insider information either, and I
- 37:09wouldn't want it, so I stick to a
- 37:11balanced portfolio.
- 37:13Let’s fast-forward to 2035.
- 37:15If everything goes well, what will a
- 37:17typical workday look like for a
- 37:19specialist who spends 8 hours at a
- 37:20computer today?
- 37:21>> Great question.
- 37:24I don't know.
- 37:27I can easily imagine a future where
- 37:29people work a little less and remain
- 37:32very productive, but I can also imagine
- 37:35a scary future where many won’t have
- 37:38meaningful work.
- 37:41And I can imagine people still going to
- 37:44the office or combining office and
- 37:46remote work, but working differently
- 37:48because artificial intelligence truly
- 37:51augments their capabilities, with human
- 37:53ideas and talent at the center.
- 37:57Even leaving artificial intelligence
- 37:59aside, we live in extraordinarily
- 38:01turbulent times.
- 38:04Climate change, an aging population,
- 38:06geopolitical shifts, and the huge middle
- 38:09class emerging in developing countries.
- 38:13All of this requires new goods, services
- 38:15, tasks, and new approaches to markets
- 38:18and company organization.
- 38:20And who will handle that? Not artificial
- 38:23intelligence, but people.
- 38:26And it would be wonderful if we helped
- 38:28these people think creatively.
- 38:31What would you advise those who see the
- 38:32massive wealth being created by
- 38:34artificial intelligence?
- 38:36At least judging by company valuations,
- 38:37while real economic returns are yet to
- 38:39be seen. How can one practically
- 38:41participate in creating all of this?
- 38:44>> Honestly, despite all the investment in
- 38:46leading AI labs, I still don't
- 38:48understand how they plan to make big
- 38:50money on foundation models.
- 38:52>> So, we are in for some failed IPOs.
- 38:54>> Not immediately.
- 38:55For now, there is enough enthusiasm, but
- 38:58I really don't see how to make trillions
- 39:00of dollars from base models.
- 39:03Their development or distribution may
- 39:06face serious obstacles.
- 39:08And even if that doesn't happen, the
- 39:10next generation of open models will be
- 39:13very good, so the potential to make
- 39:15money from the models themselves will be
- 39:18limited.
- 39:19The money, in my opinion, is in quality
- 39:21data and applications.
- 39:23That is where you can gain an advantage
- 39:25.
- 39:26If your company or industry has a lot of
- 39:29data, work with it, use it, improve its
- 39:32quality, find ways to collect new data,
- 39:35and pay the employees who create it.
- 39:39This is both fair and good for business
- 39:42. Think about applications. That is what
- 39:45will provide the advantage.
- 39:47If an application is truly original and
- 39:49solves a problem in a specific niche, it
- 39:51will be much harder to replicate it
- 39:53using open models.
- 39:55That is where I see the opportunities.
- 39:57And for workers, it is important to stay
- 39:59flexible, understand AI, but not lose
- 40:01their creative thinking and fundamental
- 40:04knowledge.
- 40:05>> It is also important for employees to
- 40:06understand that they are the source of
- 40:07this data, having worked with clients
- 40:09and products for years.
- 40:10>> By the way, this is also important for
- 40:12those building applications. A lot of
- 40:14practical knowledge exists only within
- 40:16the employees. It is not written down
- 40:18anywhere.
- 40:19You can't just sit at a computer in an
- 40:20office, take a powerful AI, and invent
- 40:22an application. You need to go on-site,
- 40:24understand the work, and talk to people
- 40:27.
- 40:27>> I hope this conversation gives ideas to
- 40:29the entrepreneurs watching us who want
- 40:31to create useful AI applications. And
- 40:33one last question.
- 40:35You said that students joining AI
- 40:37companies need a reliable moral compass
- 40:39. I think everyone needs one,
- 40:41entrepreneurs and every member of
- 40:42society alike. What do you mean by that
- 40:44? Could you please explain that in more
- 40:46detail?
- 40:47>> Let me give an example from another
- 40:49field. Gene editing.
- 40:52It is also an amazing technology that
- 40:55has seen major breakthroughs.
- 40:58But most people going into medicine or
- 41:01related fields understand that there is
- 41:04a lot of ethical ambiguity here.
- 41:07And many regulatory bodies, from the
- 41:10National Institutes of Health to the
- 41:12U.S. Surgeon General and the American
- 41:15Medical Association, have issued ethical
- 41:18guidelines for gene editing.
- 41:21For example, germline gene editing,
- 41:23where changes are passed on to
- 41:25descendants, is currently not permitted
- 41:27.
- 41:27It is too dangerous. We don't understand
- 41:30the consequences well enough. There are
- 41:32too many risks.
- 41:34We need such moral guidelines in
- 41:36artificial intelligence as well, but we
- 41:38don't have them.
- 41:40People say, "I am doing this for the
- 41:42greater good," and I believe them.
- 41:45Almost everyone I know in this field
- 41:47sincerely wants to benefit not just
- 41:49themselves, but society as a whole. But
- 41:51there are no clear moral guardrails.
- 41:54That is what I am talking about.
- 41:55>> Another interesting topic. Perhaps we
- 41:57will continue it separately.
- 41:58>> I hope there will be another opportunity
- 42:00to continue. Thank you so much for an
- 42:01amazing conversation.
About this transcript
This page contains the full transcript of Кого ИИ сделает богаче, а кого оставит без работы | Нобелевский лауреат по экономике by Marina Mogilko, generated from the public captions YouTube serves with the video. The transcript has 6,456 words across 1,133 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.