Facing AI, what "cards" do humans still have in their hands?
In the past two years, discussions about AI have almost revolved around the same word - unemployment. Which jobs will disappear, how many programmers will be replaced, and who will be the next white - collar worker to lose their job. This narrative is easy to understand and anxiety - inducing enough for everyone to relate to.
But the other day I re - listened to an episode of Dwarkesh Patel's podcast. Two economists - Alex Imas, a behavioral economist from the University of Chicago Booth School of Business, and Phil Trammell, an economist researching long - term growth at Oxford - turned this issue on its head.
They said, "What will AI replace?" is actually a question with little information. As machines become more powerful, they can do more and more things. This is a trend and there's no point in arguing about it. The truly difficult and important question lies on the opposite side:
When machines can do almost everything, what will still be scarce?
Because in economics, there is a near - ironclad common sense: value always lies on the side of scarcity. If something is available in abundance, it has little value; wealth and power will ultimately flow to places where "machines can't produce yet".
Thinking along this question, I found that the answers were more and more counter - intuitive. And the deeper I delved, the less it seemed like a story about "jobs" and the more it resembled a story about "who owns what".
01
AI Can't Take the "Blame" Yet
Let's start with a specific strange phenomenon.
In the past two years, AI's capabilities in many professional fields have approached or even exceeded those of ordinary practitioners. However, you'll find that the automation of positions such as lawyers, accountants, and senior engineers is much slower than expected. Why?
The intuitive answer is "AI is not strong enough". But Imas gave a completely different explanation: Many times, when you hire a lawyer, what you're really buying is not their ability to write documents or research cases, but a responsible entity that can "endorse" the results.
What you need is an "entity" that can have its license revoked, be sued, and truly bear the consequences when something goes wrong. You need someone to sign, someone to be fired or held accountable, and a compliant license on display. These have little to do with the lawyer's professional level and are purely institutional requirements that "someone must be in that position". Even if AI can do the job ten times better, the "responsible" position still has to be filled by a human for now.
Trammell added a more sophisticated model, from a recent study by economists Gans and Goldfarb, called "O - ring automation". It means that a job is often not a combination of nine independent tasks but a chain. You can automate 90% of it, but if AI performs worse than humans in the last 10%, the quality of the entire product will be dragged down by this weakest link.
So the rational choice is actually not to automate even the 90%. Conversely, if a person does a poor job in their 10% responsibility, they will also lower the 90% that AI has done well.
This model immediately explains the long - puzzling phenomenon: Why, even though AI's individual capabilities are sufficient, the entire position has not been replaced for a long time. Because what determines whether a job can be outsourced is not its strongest link, but its weakest and most error - prone link.
At this point, the story seems to have a heart - warming conclusion: There are always some "human parts" that AI can't replace, and humans just need to hold the line on responsibility, trust, and signing off.
But the two scholars immediately poured cold water . They said that these moats supported by regulations, licenses, and the requirement of "someone must be responsible" are probably "transitional".
In Trammell's view, legislation, being a judge, being a juror, and various license systems that lock certain professions to humans are all transitional arrangements. In human history, "what must be done by humans" and "how politics should be organized" have changed too many times, from small hunter - gatherer tribes to empires to modern bureaucracies. Once an AI - led arrangement completely outperforms the old organization in terms of efficiency, it will eventually squeeze out the old one. What we think "this must be the responsibility of humans" today may just be because we're not used to handing it over yet.
That is to say, the bottom line of "human responsibility" can hold for a while, but it's not the end - game.
So, if we can't even hold on to "responsibility", what will be truly and permanently scarce after AGI?
02
There Will Be More and More Robots, but Not More Ballet Dancers
Here, the conversation took a deeper and more interesting turn.
The two scholars believe that the truly non - disappearing scarcity is "the relationship between people" itself, which is called "relational goods" in economics. A wedding personally organized by friends, a real - life psychological consultation, a live ballet performance. The value of these things lies precisely in "being provided by a living person".
But what's really interesting is not this conclusion, but the way they argued it. They didn't appeal to emotions but brought up another concept:
Evolution.
Imas's reasoning is as follows: Suppose there are two types of people in the world. One type doesn't care and will use whoever can better simulate companionship. If an AI psychological counselor is cheaper and more useful, they'll use AI. The other type has a near - moral aversion and thinks it's wrong to outsource interpersonal communication to machines.
So, which type of person is more likely to find a partner, get married, have children, and pass on their genes? The answer is quite clear: the latter. So the preference for real people will be strengthened by natural selection over generations, rather than diluted. Imas also mentioned that geneticist David Reich said in the same podcast that humans are still being strongly shaped by natural selection. In other words, even if some people don't care about AI companionship now, the pressure of choice will push the overall preference towards "being more dependent on real people".
This is a rather sharp perspective: Our preference for real people may not be because we're noble, but because those who don't prefer real people didn't leave descendants in the long process of evolution.
Then, a particularly wonderful picture emerged in the conversation, which explained in one sentence "why relational goods will become more and more expensive".
Trammell mentioned a concept ignored by most macroeconomic models, called "investment - specific technical change". In simple terms, in the future, what will become extremely cheap are mainly "capital goods" - machines, computing power, robots; while the prices of the parts of consumption provided by real people will hardly change.
His analogy is: One robot product this year can become a hundred robots next year - manufacturing and computing power are expanding exponentially. But the number of ballet dancers will remain the same next year. The marginal utility of a ballet performance is basically the same as it is today; while the marginal utility of a robot is much lower than it is today. So, if you use "robots" as a measure for that ballet performance, our desire for it will be much stronger than it is today.
This is the magic of scarcity. When robots become almost free, if you use "robots" as currency to measure a live performance, it will be incredibly expensive. It's not that the ballet itself has improved; it's that everything around it is depreciating, so it appreciates relatively and almost crazily. Just like in a world full of only gold, a glass of clean water is the real hard currency.
At this point, "humans" seem to firmly stand at the top of the value chain: machines are responsible for producing everything, humans are responsible for providing the bit of warmth that machines can't give, and then reap the benefits.
But if you really look at where the "money" is flowing, this warm picture immediately shatters.
Imas and Trammell asked us to look at the wealthiest people in the world and see in what form their wealth exists.
Most of Zuckerberg's wealth is in Meta's stocks. As the controlling shareholder, he could completely let Meta distribute all the profits as dividends and pocket the cash for consumption - hire MMA coaches for his wife's birthday, hire dancers, and buy all kinds of relational goods. But he didn't. He'd rather let his wealth continue to grow and let Meta use the money to build more data centers.
Elon Musk is even more extreme. He's seriously talking about building an "electromagnetic catapult" on the moon. He's the richest man on earth, but he clearly doesn't care whether the researchers working for him in the future are humans or AI.
The two scholars pointed out a characteristic: The wealthiest people have an "insatiable" appetite for capital.
Ordinary people will turn to consumption and enjoy relational goods after earning enough, but these people won't. They have the highest savings rate, so over a long enough period, in the end, most of the wealth will belong to them. And what they want is not ballet, but more machines, more computing power, and more machines that can produce more machines.
So there's a cruel misalignment here: Even if "human value" really becomes scarcer and more expensive, those who get this dividend may not be "humans". What's scarce is the relationship, but those who hold the wealth are precisely the ones who don't want relationships and only want more machines.
What about ordinary people? What can ordinary people rely on to get a share in this feast?
03
Is AI Like Electricity or Social Media?
This is the most memorable question in the entire conversation, in my opinion.
When the host asked what countries not in the AI industry chain - India, Nigeria, Uganda - should do now, Imas didn't give the standard answers (improve education, build data centers, train local engineers). Instead, he asked a rhetorical question:
In the end, will AI be more like "electricity" or "social media"?
Think about power supply companies. They're almost monopolies, and everyone needs electricity. But do we think power companies hold great political and social power? No. Because most of the benefits brought by electricity flow to "electricity users" - factories, shops, and households all benefit from it, and power plants only earn a stable profit.
The dividends of electricity are spread out.
Social media is the opposite. Everyone uses it, and it seems free on the surface, but all the "rents" - your attention, your data, advertising fees - are all taken away by the platforms. Although both are "used by everyone", one spreads the benefits to everyone, while the other siphons the benefits into a few companies.
Which path AI will take almost determines the fate of ordinary people.
If AI is like electricity: In the future, every company in the S&P 500 will be there because it uses AI well, and the benefits of AI will be spread throughout the economy. Then, as long as you buy an index fund, you'll get a share of the AGI dividend. In Imas's words - Nigeria only needs to "buy the index" to own AGI.
If AI is like social media: All the money will be taken away by companies like OpenAI and Anthropic, and you can't buy them - they're not listed yet, and the profits are highly concentrated in private equity that ordinary people can't reach. Then, ordinary people and poor countries will be left far behind.
What can determine which way it goes?
Imas said it's open - source models. If open - source models always lag behind the cutting - edge by six to nine months, then once someone achieves AGI, everyone will be able to use the same capabilities in a few months, and AI will be more like electricity. That's why open - source is not just a technical route dispute; it's actually the master switch for "whether wealth will be spread out or concentrated".
And here, there's a historical depth that I didn't realize before.
The host pointed out a sharp fact: Why don't the descendants of tycoons like Rockefeller and Carnegie from a hundred years ago rule the world today? A often - overlooked reason is - For a long time in history, ordinary people simply couldn't "own the entire economy".
Before the emergence of index funds, if you wanted your wealth to grow with the economy, you had to personally pick the few companies that would skyrocket in the future. If you picked the wrong ones, your wealth would stagnate. Over the past hundred years, most of the value created by the economy has actually been highly concentrated in a very small number of companies - if you missed them, no matter how much capital you had, it would just stand still.
It wasn't until the 1970s when John Bogle founded Vanguard and launched the first index fund for ordinary people that humans finally had a handy tool: without having to pick companies, you could buy "the entire market" at once and hitch a ride on economic growth. Trammell said that there was probably a "golden window" after that - ordinary people could finally let their wealth grow at a rate similar to that of the overall economy.
But this window may be slowly closing.
Today, the most valuable assets are increasingly accumulating in unlisted private companies - OpenAI, Anthropic, SpaceX - these are precisely assets that ordinary people can't buy. And what's the biggest "capital" ordinary people have? A house. Unfortunately, a house is the least suitable asset in the world to "complement AI": its value lies in "being close to others", but humans may not be an important production factor in the future. When the center of production shifts from "people gathering together" to "machines gathering together", an asset like a house that bets on "people" becomes awkward.
Of course, there's also an optimistic side. Developing countries are not without precedents of "leapfrogging" - mobile payment in Africa directly skipped an entire generation of infrastructure such as credit cards and bank branches, and the popularity of M - Pesa in Kenya outpaces many developed countries. Imas said that a sufficiently powerful technology can indeed allow people to skip the intermediate step and move forward.
So, the question of "what poor countries should do", which seems like a development - economics question, is reduced to a very simple and sharp question: Do you have a "ownership" ticket in the wealth that AI is about to create?
After listening to the entire conversation, my biggest feeling is that our collective anxiety about AI may have been misdirected from the start.
We're all worried about whether our "jobs" will be taken away. But these two economists used a whole set of reasoning to show that jobs are just the surface. Behind a job is an income; behind an income is your "little bit of ownership" of this economy. What AI really shakes is not whether you'll be unemployed, but - When machines can produce almost everything, the only thing still scarce is the "qualification to own those machines".
Another thing Imas said really struck me. He said that the current negative narrative about AI is not because bad things are more likely to happen, but because "it's much harder to imagine a good thing that doesn't exist yet than to miss something that's being lost".
It's easy to describe unemployment. You just point to a specific person and say "your job is gone"; but it's difficult to describe a future where everyone benefits because it doesn't exist yet and there's no picture.
Fear always runs ahead of hope.
His last words sounded light but carried a lot of weight - "There's no one in the world who opposes electricity". Electricity also took away some people's jobs back then, but no one stands up against it today. What's the difference? It lies in the fact that the benefits of electricity were ultimately spread to every electricity user.
Will AI one day become something that no one wants to oppose? This probably doesn't depend on how powerful the model is, but on a simpler thing: When machines can produce everything, whether the value that can't be produced and the "qualification to own these machines" are taken into the pockets of a few companies or spread to every ordinary person.
There's no answer to this question yet. But at least, it's more worthy of our anxiety than "will AI take away my job".