By Zhiqian Wu
In mid-October, the people who manage the world's money gathered in Bangkok for the annual meetings of the International Monetary Fund (IMF) and the World Bank. The topic everyone kept coming back to this year was AI.
IMF Managing Director Kristalina Georgieva offered an analogy: AI is like a Formula 1 car. It goes incredibly fast, but one wrong move and it can fly off the track and hit the people standing beside it. If we don't put protections in place ahead of time, she warned, AI could end up causing more trouble than good.
The odd thing is that just a week earlier, she had said something far more optimistic: used well, AI could add 0.5 percentage points to global growth every year. Over time, that would be like conjuring up an extra economy the size of ASEAN.
So one week the upside is huge, the next the risks are. Which one is true?
Both, actually. AI comes with two ledgers. One records how big the cake can get. The other records how the cake gets divided. The first ledger looks reasonably good for now. Nobody has worked out the second one yet.
I. Is the Cake Getting Bigger? The Money Has Been Spent, the Results Haven't Arrived
Let's start with a fact that may surprise you: right now, AI's contribution to the economy comes mainly from building AI, not from using it.
What does that mean? Big companies are spending heavily on data centers, chips, and power lines. That spending itself counts toward economic growth. The IMF expects the global economy to grow 3.1% this year and 3.4% next year. With high oil prices and heavy government debt in many countries, that's not a bad result, and AI investment deserves much of the credit.
But the IMF itself admits that its forecasts don't include the productivity gains from people using AI to work more efficiently. The reason is simple: not many industries have truly put AI to work yet.
Think of it this way. A highway is being built. The construction alone has created jobs for workers and business for material suppliers, but there are hardly any cars on the road. Will the cars come once the road is finished, and when? Nobody can promise.
That's where the problem lies. Companies are willing to pay for the road because they believe the cars will definitely come. What if that confidence wavers and investment suddenly pulls back? The IMF has run the numbers: in that case, global output could be up to 0.4% lower by 2028.
More worrying still, some large companies are running short of their own cash and are increasingly borrowing to invest in AI. When borrowed money pays off, everyone is happy. When it doesn't, it can set off a chain reaction.
This brings to mind the dot-com bubble around 2000. The fiber-optic cable laid back then did eventually get used. But many of the companies and investors who laid it lost everything before that day came. A technology proving useful in the end does not mean all the money spent on it today is well spent.
II. What Happens to Jobs? The People Who Most Need a Chance May Lose It First
There's an awkward twist here: AI helps newcomers the most, yet it may lead companies to hire fewer of them.
The IMF has cited a study of customer service workers. After they started using AI, overall productivity rose by about 15%, while for newly hired workers it rose by more than 30%. Put simply, AI works like an experienced mentor you carry around with you, passing veterans' know-how directly to beginners.
That sounds like good news. But look at it the other way: if one experienced employee plus AI can now do the work that used to take three new hires, will the boss still hire those three?
Singapore's President Tharman Shanmugaratnam raised exactly this point at the meetings: AI is making it harder for university graduates to find jobs, and governments can't ignore it.
Picture a career as a ladder. Junior analysts, paralegals, entry-level programmers, customer service staff: these jobs are the first rung. It's where newcomers make mistakes, get corrected, and slowly build their skills. If the first rung is removed, where will the experienced professionals of ten years from now come from?
The usual answer is more training, so people can learn new skills. But the IMF has also found that AI's benefits have been slow to materialize precisely because training hasn't kept up and companies haven't reorganized themselves. What holds AI back is often not the technology itself, but people and organizations failing to keep pace.
Training also has an old problem: who pays for it? Companies worry that once they've trained someone, a competitor will poach them. Government programs often lag behind. And individuals may not know what they should be learning.
Georgieva has compared AI's impact on jobs, taxes, and education to "a slow-moving tsunami." A tsunami that arrives slowly doesn't mean you can wait until the water reaches your feet to build the seawall.
III. How Is the Cake Divided? The Money May Increasingly Flow to the Owners
Even if AI really does make the cake bigger, there's another question: who gets the extra slice?
Building AI is extremely expensive. Chips, data centers, and large models are a game only a few big companies can afford to play. So the money AI makes shows up first as these companies' profits and share prices, not as ordinary people's wages.
If companies use AI mainly to hire fewer people, rather than to help their employees do better work, we end up in an uncomfortable situation: the economic numbers in the news keep looking better, while working people's wallets don't get any fuller.
This also creates a headache for governments. In many countries, taxes and social insurance are funded mainly through deductions from wages. If more and more of the money is being made by machines and capital rather than by people, how will governments pay for training, unemployment benefits, and pensions? That's why the IMF chief talks about taxes and jobs in the same breath.
During the meetings, a few dozen people held a peaceful protest outside the venue, calling attention to the widening gap between rich and poor. Inside, people were discussing how many extra percentage points the economy could grow. Outside, people were asking: how much of those percentage points will end up in my hands?
How big the cake gets, the market will work on by itself. How the cake is divided, the market will not sort out on its own. That takes policy.
IV. What About Developing Countries? Some People Haven't Even Reached the Track
When people talk about AI, they think of Silicon Valley and Wall Street. But holding this year's meetings in Bangkok was a reminder in itself: most of the world's people live in developing countries, and AI is still far away from them.
The first hurdle is a very basic one. As the IMF chief put it, for AI to help more people, they first need electricity and internet access. For some countries, what needs filling in first is power lines and network cables, not large models.
The second hurdle is whether a country can join the AI supply chain. Right now, AI is built mainly in the United States, China, and India, and much of the hardware comes from Asia. According to IMF data, 62% of the AI computing hardware sold worldwide comes from South Korea and Taiwan. The host country, Thailand, was also named as one of the economies that has benefited from AI investment. Those who make it into the chain get orders and factories. Those who don't may be left only with the side effects.
There is good news, though. Some speakers at the meetings pointed out that for less wealthy countries, "small AI" holds real promise: AI on a mobile phone that helps farmers spot pests and crop disease, helps village doctors make an initial diagnosis, or helps shopkeepers keep their books. For these countries, the key is not whether they build their own race car, but whether they pave the road first.
Of course, not everyone agrees with the IMF's caution. A U.S. Treasury deputy secretary argued that the IMF focuses too much on risks and should pay more attention to the opportunities AI brings. The disagreement is telling: the people sitting in the race car and the people standing by the track naturally have different ideas about how fast is safe.
Final Thoughts: Four Things Ordinary People Should Watch
Will AI turn out to be "an extra ASEAN," or "more trouble than it's worth"? You don't need to wait for experts to decide. Just keep an eye on these four things:
Whether people are actually getting work done faster. If, a year or two from now, all we see is companies spending money with no visible efficiency gains across industries, it's time to worry about a bubble.
Whether new graduates can find jobs, and whether starting salaries are rising. This is the most direct signal of whether the first rung of the ladder is still there.
Whether training is real investment or just slogans. Look at how much money companies and governments actually spend, and how many people actually get trained.
Whether wages are taking a bigger or smaller share of economic growth. This tells us whether the cake is being shared by everyone, or ending up in the pockets of a few.
Whether AI truly benefits everyone has never depended only on how smart it is. Efficiency gains need to turn into real output, and ordinary people need a path forward, whether that means keeping the jobs that remain or finding the new ones that emerge. Neither will happen on its own.
How fast the race car can go is the engineers' business. How the guardrails along the track get built is everyone's business.
