Muchos analistas predicen que la IA transformará la economía global, pero Daron Acemoglu, economista del MIT y ganador del Premio Nobel, ofrece una visión contraria. En este breve video, explica por qué la IA podría automatizar solo el 5 por ciento de las tareas y aportar solo un 1 por ciento al PIB mundial durante la próxima década, y comparte su perspectiva sobre cómo los líderes empresariales deberían abordar las inversiones en IA.
00:05From agentic AI to instant cures, the hype around AI
00:08can be deafening.
00:10But what's the real economic impact,
00:12stripped of the speculation?
00:14Today, we cut through the noise with MIT economist
00:17and Nobel laureate Daron Asimoglu, whose data-driven research
00:21reveals a surprising reality.
00:23Forget overnight transformation.
00:25Asimoglu's research projects that AI will automate just 5%
00:29of all tasks and add just 1% to global GDP this decade.
00:34So why the massive disconnect?
00:35And what should smart business leaders
00:37be doing with AI right now?
00:39I recently interviewed Asimoglu and asked him
00:42these questions and more.
00:50Thank you so much for being here with us today.
00:53I have a few questions for you about generative AI
00:56and AI in general and its impact on the economy.
00:59So ChatGPT came out in November 2022.
01:02And since then, we've seen generative AI go through a lot
01:05of developments.
01:06It has observers, I think, excited and a little bit worried
01:09about what it means for their jobs and for the economy
01:13in general.
01:14Last April, you published a paper called The Simple Macroeconomics
01:19of AI, in which you estimate that over the next 10 years,
01:23only about 5% of all tasks will be profitably automated by this
01:28technology and that it's only likely to contribute about 1% to global GDP.
01:33That's a stark contrast to what some other analysts have said.
01:38You know, people have been predicting that this will be a truly transformative technology
01:44to the labor force and to the economy in general.
01:48Can you explain why your estimates are different from these others?
01:51And since you published that paper last year, have you seen anything that either confirms
01:58or makes you question those estimates you made?
02:00Well, thank you, Kaushik.
02:02Well, look, I said one other thing in that paper.
02:05It's hugely uncertain.
02:06And these are just guesses.
02:08I think it's very difficult to know because it's a very rapidly changing technology.
02:12And over the last year, we have seen even more advances.
02:16So we don't know where we're going.
02:18But the basis of my prediction, uncertain though it may be, still remains.
02:25The industry has not produced applications that are critical for the production process
02:36or for generating new goods and services that are going to be hugely valuable.
02:40So if you compare AI to the internet, I think from the very early days of the internet,
02:47even when there was hype and a boom, it was clear how the internet was going to change everything.
02:54The way that we communicate has been completely transformed by the internet.
03:00It was very clear at the time.
03:01It was also very clear that the internet would introduce a lot of new goods and services
03:05and provide platforms for people to come together in various ways for production,
03:10for recreation and other things.
03:12I think those things are not clear yet for AI.
03:16Of course, if you're a believer that AGI is just around the corner,
03:20you think somehow in the next few years, somehow we're going to get such amazing machines that
03:29they can start performing all the cognitive tasks.
03:33But even that scenario is not so clear in how you're going to actually get
03:38AI tools into the production process.
03:41And I think the current approach is well targeted for dealing with cognitive tasks
03:51that are performed in predictable environments, in offices,
03:56and don't require much social interaction and very high levels of judgment.
04:00So if you are a software engineer that does some very basic routines for your work,
04:09or you're in IT security, or you're in accounting,
04:12those are things that I think there will be applications based on AGI
04:17and some other AI tools that will be able to perform these tasks.
04:21If you're a CEO, if you're a CFO, if you're an entertainer, if you're a professor,
04:27if you are a construction worker, or a custodial worker, or a blue color worker,
04:33I think those things are beyond what AI can perform, or AI can indirectly contribute to
04:42by being bundled with flexible robotics because we're not there in terms of those technologies.
04:47So when you do that calculation, you end up with about 20% or so of the economy
04:54that is either at the crosshairs of AI to be automated or could be majorly boosted by AI input.
05:02Things that are feasible, they take takes a long time.
05:04Many of them are performed in small companies.
05:07It's not going to be profitable to do them.
05:08So that's how I arrived to the 5% number.
05:10Based on these inputs and a lot of detailed material.
05:15But it may turn out to be wrong.
05:18Last year, I wouldn't have expected to see the kinds of leaps and bounds.
05:22I mean, the leaps and bounds are really inspiring at some level.
05:25So I'm pretty impressed by those.
05:30The question is, with these leaps and bounds,
05:34do you still think that in two, three, four years time, you can have an AGI with no human supervision that can do all of your accounting or all of your marketing?
05:51And I think that is a much higher bar.
05:54Why? First of all, because every single occupation has so many complex, tacit knowledge parts and requires a lot of checking and a lot of different types of intelligence being applied to it.
06:08And does that tie into the distinction you make in the paper between what you call easy to learn and hard to learn tasks?
06:14And should that distinction inform how executives study or decide what business processes are most amenable to automation?
06:26Look at the domains in which we have truly inspiring achievements from AI, such as AlphaGo, AlphaFold, or answering some complex but knowledge-based questions.
06:44Those are all domains in which there is a ground truth that everybody can agree on.
06:50You know, you either fold the protein or you do not.
06:53AI is capable, there's no doubt about that, that's why we're talking about AI, and it is capable of learning that knowledge if it's in its training data set.
07:02So once you provide AI with the right powerful algorithm, for example, reinforcement learning was very important for the Alpha series, maybe other things for generative AI, and the ground truth is there, AI is going to get there.
07:16But no task that we perform in reality is just recounting already established knowledge or playing a parlor game.
07:26They are much more complex, they involve interactions, they involve a lot of things that are based on tacit knowledge,
07:32or they are based on matching your contextual understanding of a problem with the specific task at hand.
07:40For example, diagnosing a difficult ailment, or finding the kind of product that's going to work well given the retirement planning that an individual is doing.
07:50With the current architecture, the best that we can do is we can copy human decision makers that make decisions.
07:55So we can load in a lot of data from doctors making diagnosis, or reading radiology reports, or from financial planners.
08:07And then AI, generative AI in particular, has a great way of imitating these human decision makers.
08:15But if you do that, you're not going to get much better than the human decision makers.
08:18And especially if you don't know who the very best human decision makers are, you may not even very easily achieve the best level human decision maker level.
08:26Places where we need a lot of judgment, or social interaction, or social intelligence, I think are still beyond the capabilities of AI.
08:34And on the basis of this, I would say, you know, my prediction, which, again, has huge error bands around it, so may well turn out to be wrong.
08:43But I don't expect any occupation that we have today to have been eliminated in five or 10 years time.
08:50So if you are an AGI believer that you think that generative AI and other AI tools are going to completely transform the economy within the next three or four years or five years,
09:01then you must have in your mind a list of occupations that will completely disappear.
09:05All of this that I have summarized briefly is predicated on the current approach to AI.
09:16And what I have been arguing, and this paper was a small part of that bigger edifice, is that we are not developing AI in the best possible way.
09:28And that best possible way is much more pro-human.
09:31It's much more targeted at working with human decision makers.
09:36It requires a bigger celebration of the places where AI is better than humans and the places where humans are better than AI.
09:44And once you take that approach, I think the biggest promise is using AI for providing new goods and services, new ways of doing things for humans.
09:55We are at the cusp of many major transformations.
09:59We are an aging society.
10:01There are going to be many, many more people over the age of 60, many, many, many more people over the age of 70 in the United States, many more in Europe.
10:09They are going to demand new goods, new services, new accommodations.
10:16Financial industry is at the cusp of big changes.
10:19Again, this is not going to be on cost savings.
10:21It's going to be, for example, what sometimes people call financial inclusion, meaning we provide new, better services for people who are not currently making enough use of financial services, including banking.
10:32Climate change, whether you mitigate it or not, is going to change many aspects of our lives.
10:38Again, new goods and services.
10:39And the entire production process requires new tasks, new ways of increasing the expertise and sophistication of workers.
10:48All of these, I think, are to play for.
10:50And those are the places where I think AI could make a big difference.
10:53So my recommendation to business leaders would be don't be taken by the hype.
10:57I think the hype is an enemy of business success.
11:00Instead, think where my most important resource, which is your human resource, can be better deployed.
11:09And how can I leverage that human resource together with technology, together with data, so that I increase people's efficiency and I enable them to create better and newer goods and services.
11:23Not just cutting costs, but doing new things that are so important in this changing world.
11:28Business executives should really be thinking about a much wider scope of possibilities than simply eliminating costs or finding roles that they can cut from their organizations.
11:39That's my perspective.
11:41Again, you will be hard-pressed to find many people in Silicon Valley who agree with this perspective.
11:47But I've been researching this for quite a while.
11:50I may be wrong, but at least I do have data, I do have historical knowledge, and I do have some theoretical understanding of these issues.
11:57And I would say, on the basis of those, that, of course, any business leader should be happy if they can reduce their costs even by 1%.
12:04That's great, 1% more profits.
12:07But the evidence, as far as I read, is quite clear.
12:13No business has become the jewel of their industry by just cost-cutting.
12:18All good business leaders are looking for that next big idea, that next innovation that can turn them into one of these stars of their industry.
12:30In the meantime, right now is when they are putting investments into AI, and they are starting to look for a return on that investment.
12:38What metrics do you think they should be paying attention to to know whether those investments are really paying off?
12:44Well, I'm not going to be able to provide a simple metric for you, but let me give you my perspective.
12:49And the reason why I wrote the paper that you started with is precisely because I'm worried about those investments.
12:54I think most business executives, not all, but most business executives, are investing in AI blindly.
13:02They are doing so without understanding how AI can be synergistically deployed with their workforce.
13:09And they are doing so because they are under tremendous pressure, because every day they hear from management consultants, from the newspapers, from podcasts, that your competitors are investing big time in AI, and if you're not, you're falling behind.
13:22That's not the way to create a successful business.
13:27You never create a successful business because you think your competitors are investing, and you should do it not to fall behind.
13:32And I think the recipe that I would suggest is start by thinking about where it is that you can make a big difference in terms of the new things that you do.
13:43I think for many financial industries, it's quite clear.
13:47New financial services are badly needed.
13:49I think if you are producing other services, health services, education services, I think a complete overhaul of these things is necessary.
13:57And that's not going to happen just by buying more cloud services from Amazon or just introducing some generative AI tools easily.
14:07It's going to happen by identifying, with the help of your most skilled employees, identifying where these new services can be introduced, what the demand for them is, and how that can be made possible.
14:21Well, then, AI would then be a great tool to augment the capabilities of your workforce and yourself in doing that.
14:28That's fascinating.
14:29Well, thank you so much for your perspective, Daron.
14:32You've given us a lot to think about.
14:34I hope you enjoyed my discussion with MIT economist and Nobel laureate Daron Asamoglu on AI's economic impact.
14:40The key insight for leaders, rather than following your competitors into blind AI investments, focus on how the technology can help you and your team deliver meaningful innovation.
14:51Are you seeing AI create new opportunities in your industry?
14:55Share your thoughts in the comments.
14:57For more research-based information from MIT SMR, check out this playlist.
Sé la primera persona en añadir un comentario