The level of intelligence that artificial intelligence (AI) has so far demonstrated has shocked most observers. AI seems to be able to answer any questions we ask and write better essays than most people today can ever write.

This sensation has caused many to think whether human intelligence may soon be rendered obsolete. Some reactions are outright extreme, such as “we are all going to die by 2030.”

What is the nature of “machine intelligence”? How is it different from “human intelligence”?

I have been puzzled by these questions for a while. About a year ago, in the middle of 2025, I wrote a blog piece on this subject (link).

In recent weeks, slowly and then suddenly, a form of mental clarity came upon me. An answer landed in my head. I see two words: “Pure Thinking.”

Pure Thinking is a term that I use. It describes a way of thinking that is disconnected from worldly reality and only exists within an abstract pre-defined box.

It is highly likely, I believe, that “machine intelligence” is “pure thinking.”

If my perspective is right, it will have major bearings on the further development of today’s AI and its potential impact on the world.

Coding

It has become a broad consensus that coding is largely “solved” by AI. I’m a self-taught coder. I used to build financial software. I generally agree with this consensus.

Software codes are abstract, digital, readily available. In coding, objectives are clear—”I want my software to do xyz.” Feedback cycles are short—the codes run or they produce error messages.

Computer codes live within computers. They follow a set of rules that is clearly defined. They will succeed or fail based on those rules.

Being good at computer programming requires a person to be strong in abstract thinking, logical reasoning, and problem solving. It does not require a person to be social, outgoing, and presentable. You do not need to have broad business experiences to be a good coder. You do not need to have a broad personal network to be a good coder. What you need is your brain. That is one reason why computer programmers have this notorious reputation of being “nerdy.”

Computer coding, because it does not involve any meaningful input from the worldly reality, is very much self-contained abstract thinking—i.e., pure thinking.

Coding is pure thinking. AI is pure thinking. That is why coding is largely solved by AI.

Mathematics

“Mathematics,” to a certain degree, is similar to software programming. Coders manipulate codes. Mathematicians manipulate symbols and equations. Neither require close contacts with the real world. Both offer tight and fast feedback loops: you will know whether your codes, your math equations work or not, immediately, with no delays.

Both coding and math are more about pure thinking, less about interactions with the real world.

To quote an AI researcher at OpenAI who reflected on solving Millennium Prize Problems, “for the most part [in math,] it is just really bottlenecked by thinking really hard, and the [AI] models are really good at that” (link).

So math, at least parts of it, is just pure thinking. AI is pure thinking. Therefore, math is partially solved by AI.

Investing

There are major differences between “investing” and fields like “coding” or “math.”

One, the amount of information investors are dealing with is infinite: from financial data to country geopolitics, from weather patterns to human cultures. Investing is not a confined box.

Two, some of the information is available now (e.g., company historical financial data). But much of the information is unavailable (e.g., How will consumers react to the company’s next product?).

Three, in investing, feedback cycles are long. It can take years to determine whether an investment thesis is right or not. So, how can we train an AI model for investing, when answers are not immediately verifiable?

As such, unlike coders and mathematicians, investors deal with a non-abstract reality: A world that is real, physical, human, cultural, nondeterministic, with infinite possibilities.

Will American consumers like this new product?

Is the company leadership team trustworthy?

If we do Plan A, what would competitors do next?

Pure thinking cannot help with any of these questions.

If pure thinking can help, why can so few (if any) consumer companies reliably predict whether their next products will be a big hit or not? Meta did not foresee the sudden popularity of Muse. Pop Mart did not foresee Labubu becoming a global phenomenon. The Coca-Cola company did not foresee the disastrous outcome of “New Coke.”

If pure thinking can help, how on earth does misjudgement of people occur all the time? Investors can misjudge the integrity of company management teams. Company management teams frequently make wrong hires. Frontline salespeople often misread customers.

No pure-thinking models can solve any of these questions. But those are the exact type of questions that investors grapple with.

And Beyond

The current environment we are in is producing stronger and stronger AI models.

But, after all, it is still just pure thinking, more of it, more powerful versions of it.

The thinking is extremely powerful, but it only works in a confined, abstract way. Because it is disconnected from the worldly reality, it faces a chasm that it may not be able to cross for a very long, long time.

If my understanding is correct, it leads to several immediate interpretations, such as the following:

• Investing (i.e., stock picking) is highly unlikely to be solved by today’s AI.
• Most white-collar jobs, as long as they involve interaction with the real world, are unlikely to be replaced by AI anytime soon.

If we extend this line of thinking further, we can make the following argument.

Major human breakthroughs, such as the discovery of electricity and the discovery of penicillin, did not come from pure thinking (someone sitting in their study, just thinking, thinking, thinking).

They come from iterative interaction with worldly reality.

Iterative interaction with reality, by its very definition, is constrained by human experience of doing them. It means, it is constrained by time (i.e., the clock on your wall).

No matter how fast AI models are, or how powerful AI models are, we face a bottleneck, which is the very ticking of the clock. We experience the world through the passage of time. We cannot speed it up nor experience multiple instances of time at once. We only live in one moment at one time.

If this extended line of reasoning stands, it means today’s AI will have limited impact on the world. The bottleneck is not the power of AI. The bottleneck is time.

Unless we can make time go by faster, AI’s impact on the world would be constrained.

AI is real. AI is big. AI’s impact will be huge. But the realization of AI’s huge impact is likely to take way, way longer than people expect today.

A few years out from today, there is a real possibility that we might be shocked again, not by how powerful AI is, but by how extremely inflated our expectations are today about how AI might change the world.

That is why I argue: There are limits to pure thinking.

(END)