pfp Advisory: “Can AI play chess, but not the stock market?”

Fund manager Christoph Frank reflects on the AI hype, doubts that machines will soon replace human fund managers due to imperfect information and practical hurdles, and asks whether the investment industry is even experiencing its “Deep Blue moment.”

March 2, 2026. FRANKFURT (pfp Advisory). Almost exactly 30 years ago, in February 1996, another point broke off the crown of the “crown of creation.” For the first time, a machine won a game of chess played under tournament and competition conditions against a reigning world champion. “Deep Blue” was the name of the black computing monster created by IBM that left the world in incredulous amazement. A machine defeated the best human player of the time, Garry Kasparov, in the “royal game,” and did so in the opening game! Although the world champion made a comeback and ultimately won the entire match, he lost to an even more advanced computer in the rematch just one year later. From then on, all hell broke loose and the triumphant advance of chess programs was unstoppable. Today, no human player stands a chance against programs that are available for free on various platforms.

I still remember it vividly 30 years later because the defeat of the world's best player by a machine really unsettled me at the time. The year before, I had begun to systematize my investment strategy and put it on a stable data foundation. But after Kasparov's defeat, I seriously asked myself whether my efforts might be too late. Would all the work possibly be in vain and human fund managers become obsolete in just a few years, replaced by “Investment Deep Blues”?

As we know today, things turned out differently. Over the past 30 years, there have been repeated attempts to use automated trading systems (or, today, “AI”) to manage investment funds. However, I am not aware of any cases of lasting success, but rather of projects that were started and then abandoned shortly thereafter due to lack of success. 

That's another reason why I'm much more relaxed today than I was in 1996 when it comes to being replaced by “automatic trading systems” or, as we call them today, “AI,” even though computers are now many times more powerful than “Deep Blue” was back then. What I underestimated at the time was that chess is a game with perfect information. This means that players can see all the pieces at all times and thus calculate all possible moves at any time, unlike in many card games such as “Schafkopf” or “Skat,” where each player can only see their own cards, but not those of their opponents. This perfect information in chess made it possible for machines to outperform humans in the first place. Similarly, despite being much less complex than chess, the imperfect information in Schafkopf means that, at best, the programs can only compete with a good amateur player.

And when it comes to perfect versus imperfect information, the stock markets are more like Schafkopf than chess: in addition to objective data such as stock prices, there is an enormous amount of information that cannot be objectified, but at best estimated, such as future earnings or, even more complex, the behavior of all other players active on the stock markets. This overwhelms even the most powerful computers and the best AI – in my opinion, structurally and not just because of insufficient computing power.

Of course, AI can help make better investment decisions. If only because it never gets tired, is never in a bad mood, and never has stress with its family, which can lead to suboptimal decisions. What's more, it can sift through vast amounts of data at a much faster rate than any human analyst or fund manager. (At most, the notorious and not yet fully understood AI hallucinations could remain a permanent problem.) However, there are often practical hurdles to overcome: AI must be integrated into existing processes (unless a company is completely rebuilt around AI), and experience shows that this is much more challenging in practice than in the theory of model builders. It should have total access to all relevant data. However, this is often not permitted in the financial sector for regulatory or internal compliance reasons. Moreover, some data cannot be made available in a “machine-readable” form because it is stored in the minds of employees or in “lived” routines that are difficult to objectify.

In my opinion, it could therefore be a long time before human fund managers can be successfully replaced by machines on a large scale. Perhaps it will never happen. It is also not yet clear whether the large language models that are currently so hyped are fundamentally suitable for this purpose. In this respect, it remains to be seen whether the investment industry will ever experience its “Deep Blue moment.”