Predictions Fail Spectacularly When Their Assumptions Collapse: Sayonara, Asia and the US-China Duopoly in the AI Era

Recently, I watched a video in which an economic commentator discussed the future of the world.

To summarize his argument very roughly, it went like this:

The United States and China will become the world’s two dominant powers.

Japan and Europe will gradually lose their influence and become something like “former advanced economies.”

And the two greatest sources of wealth in the future will be AI and energy.

It made sense.

Looking at the world today, the argument is fairly convincing.

The United States has its enormous technology companies. China is advancing at tremendous speed in AI, electric vehicles, batteries, solar power, and other fields.

Japan and Europe, by contrast, carry the weight of regulation, established industries, social security systems, and other massive structures. When it comes to adopting something new all at once, they inevitably appear slow.

When someone says, “The world will be dominated by the United States and China,” it is easy to think, “Yes, perhaps that is what will happen.”

But as I listened, I remembered a book I had read many years ago.

It was called Sayonara, Asia.

“Only a Few Asian Countries Can Become Advanced Economies”

I think I read the book sometime in my teens or twenties.

I no longer remember the details.

But one claim left a strong impression on me.

Only a limited number of Asian countries would be able to follow Japan in becoming advanced economies.

It would be difficult for China and many other Asian countries to develop as Japan had.

The reasons given involved democracy, social institutions, and fair rules.

For an economy to develop beyond a certain point, it is not enough simply to manufacture industrial products with cheap labor.

There must be fair competition.

The law must function.

The rules must not suddenly change at the whim of political power.

A society needs that kind of foundation.

At the time, I read this and thought, “I see. So that is how it works.”

The argument was logical.

Several decades have passed since then.

As the author predicted, South Korea and Taiwan were the only places to become advanced economies broadly resembling Japan.

But what happened to China?

At the very least, the future did not follow the simple scenario that a country cannot become an advanced economic and technological power without first becoming a Western-style democracy.

China became one of the world’s largest economies without becoming a democracy.

And its strength is not limited to manufacturing.

It is now among the world’s leaders in advanced technologies such as electric vehicles, batteries, telecommunications, drones, and AI.

Singapore also became wealthy while retaining a political system quite different from Western-style democracy.

The single path to becoming an advanced economy assumed by the old prediction never actually existed.

The Forecast Was Not So Much Wrong as Its Assumptions Were

This is the interesting part.

The expert was not foolish.

Given the evidence available at the time, the prediction was probably quite reasonable.

The problem came before the calculation itself.

It was the assumptions.

An economy needs certain institutions in order to become more advanced.

A country must pass through certain stages in order to become wealthy.

Late-developing countries will follow roughly the same path that Japan followed.

But reality offered another route.

The moment an assumption collapses, even the most precise forecast built on top of it becomes meaningless.

This is similar to what happens in IT system design.

The number of transactions will grow by 10 percent a year.

There will be roughly this many users.

Each transaction will generate this much data.

Therefore, in five years, we will need this many servers.

The calculation itself may be flawless.

But if someone announces the following year, “We are moving this service to the cloud,” the entire calculation becomes irrelevant.

The calculation was not wrong.

The foundation beneath it disappeared.

This problem follows every attempt to predict the future.

Reconsidering the US-China Duopoly

Now let us return to the present.

The United States and China will dominate the world.

AI and energy will become the next great sources of wealth.

Given the situation today, this forecast is also quite logical.

AI, in particular, consumes enormous amounts of electricity.

Companies build huge data centers.

They fill them with GPUs.

Then they secure the electricity required to run them.

Countries capable of securing vast amounts of capital, semiconductors, data-center capacity, and electricity will therefore have an overwhelming advantage.

Looking at AI today, that is a natural conclusion.

But this is where I remember Sayonara, Asia.

Will that assumption really remain true twenty years from now?

Generative AI has been widely available for only a few years.

Even within that short period, model performance, computing requirements, inference methods, semiconductors, model sizes, and AI running directly on devices have all changed at extraordinary speed.

Can we really move directly from the fact that “today’s AI requires vast amounts of electricity” to the conclusion that “the countries with the most energy will therefore control AI in the future”?

That argument contains a major hidden assumption:

Future AI will continue to consume computing resources and electricity in much the same way as AI does today.

What If AI Itself Changes?

I do not know what kind of systems AI will use in the future.

I am not an AI researcher.

So I am not claiming that AI will definitely become energy-efficient.

My point is the opposite.

Because we do not know, we cannot assume that today’s approach will simply continue.

Algorithms may become dramatically more efficient.

Semiconductors may change.

The idea of using one enormous model to handle everything may disappear.

On-device AI may become far more important.

Or an approach that none of us can currently imagine may emerge.

We do not know which, if any, of these changes will happen.

But a single major shift could break the neat chain of reasoning we use today:

“AI requires immense computing resources and electricity.”

Therefore:

“Countries with vast capital and energy resources will win.”

Structurally, this resembles the old prediction:

“Advanced economic development is difficult without democracy and fair institutions.”

Therefore:

“China cannot become an advanced economy.”

The logic itself is not absurd.

The danger lies in assuming that the initial premise will remain true forever.

Experts Calculate the Future Using Today’s Rules

When experts predict the future, they often use numbers.

Population.

GDP.

Energy consumption.

Investment in AI.

Semiconductor production capacity.

Data-center construction plans.

They add these figures together and announce what will happen in 2035 or which country will be dominant in 2050.

When a forecast contains enough numbers, the future begins to look predetermined.

But every one of those forecasts comes with an enormous condition:

As long as the rules of the game do not change significantly.

Some factors, such as population, move relatively slowly. Population forecasts are said to be fairly accurate because their assumptions are less likely to change abruptly.

Technology, however, can change the game itself.

Before smartphones appeared, even the most precise twenty-year forecast of the mobile-phone market would have been meaningless.

Before cloud computing became widespread, extending existing trends to calculate how many servers companies would own in the future would have produced the wrong answer.

AI will probably be no different.

And right now, the technology itself is changing at extraordinary speed.

It seems far too early to declare the winner twenty years in advance.

Today’s Winner Is Not Necessarily Tomorrow’s Winner

I am not saying that the prediction of a US-China duopoly is wrong.

It may indeed come true.

The United States and China may secure control of AI and energy while Japan and Europe gradually lose influence.

Twenty years from now, people may look back and say, “That commentator was right.”

But that would not necessarily mean that he had seen the future. It could also mean that the assumptions we make today happened to remain largely intact for twenty years.

When I encounter a prediction now, I pay less attention to its conclusion and more attention to this question:

“What does this prediction assume will not change?”

The United States and China will dominate.

Why?

Because they control AI and energy.

Why does AI need so much energy?

Because today’s AI does.

Will AI still work the same way twenty years from now?

Once we trace the argument back this far, a prediction that initially looked solid turns out to rest on a surprising number of assumptions.

The greatest danger in forecasting is not a mathematical error.

It is the sudden disappearance of an assumption that no one thought to question.

When I read Sayonara, Asia years ago, I thought, “Perhaps China will not become wealthy so easily.”

Today, we live in an age when a Chinese AI model can run on a Chinese-made smartphone.

The future has a cruel sense of humor when it reveals the answers.

So when someone tells us what the world will look like in 2050, it may be more useful to remember the assumptions than the conclusion.

Twenty years later, those assumptions will probably be the most interesting part to revisit.