From Deep Blue to Deep Value

Last time I posted one of these deep dives I got really good feedback. Please let me know your thoughts :) Thanks in advance! Also, sorry for the bad formatting and lack of footnotes, originally was made for substack.

“Still no chess program can play even amateur chess.”

So claimed Hubert Dreyfus in 1965 about artificial intelligence. Dreyfus was extremely certain that AI lacked the “perception” needed to beat humans at Chess. Two years later, Dreyfus was defeated by an AI. But Dreyfus was a professor and philosopher, not a chess grandmaster. In 1989 World Chess Champion Garry Kasparov made that differentiation. “"A machine will always remain a machine, that is to say a tool to help the player work and prepare. Never shall I be beaten by a machine! Never will a program be invented which surpasses human intelligence.” Just eight years later in 1997 Garry Kasparov lost a match to Deep Blue, the leading supercomputer at the time. After Kasparov’s match people started to understand that artificial intelligence (AI) had begun to surpass human intelligence (HI). Now almost thirty years later, even a low quality computer on my phone can defeat the best chess player in the world. Before the rest of the world, chess slowly began to understand that AI will be better than humans. Today, we still don’t know how and exactly where AI will disrupt human life, perhaps chess might have some answers.

Garry Kasparov, the world #1 chess player at the time taking on Deep Blue, the #1 computer model in 1997. Deep Blue ended up winning the match 3.5-2.5, confirming that for the first time in history, AI had surpassed humans.

Today, in many areas of life including chess, AI is already vastly superior. Just to stress the difference, today, the best AI is 800 points better than the best human in world. That is about the difference between the best chess player in the world and a very strong amateur player. Or in basketball terms, the difference between prime Lebron James and a pretty good three point shooter at the park. With this difference in mind, you would think that as AI has pulled ahead, the viewership would shift from human games to a better product, a game between two AIs. And yet the 2025 AI chess grand final, (in which OpenAI defeated Grok,) had only 30,000 viewers at its peak, around 10% of the viewers that tuned into the finals of the Norway Chess tournament in the same year. Even more shocking was that in 2021, an amateur chess tournament of people’s favorite streamers peaked at over 375,000 viewers. Despite AI producing a vastly superior product, we humans enjoy watching humans make moves and mistakes, not the best objective moves.

But there is a key difference between chess and real life. Chess is a board game, played for entertainment purposes, AI has not replaced humans because the end goal of chess is entertainment, not productivity. But what about situations where there is more at stake than pure entertainment.

What about health? If I go to a doctor to get a brain scan done, I want the results to be as accurate as possible, not as human as possible. And there are massive stakes on the line, medical malpractice is currently the third leading cause of death in America. The most recent studies indicate that in certain fields, artificial intelligence can actually more accurately read brain scans than doctors. Not just that, in certain time sensitive situations, such as when a patient is having a stroke, artificial intelligence can interpret the scans in seconds, a serious advantage over doctors in life or death situations. In addition, AI can detect tiny discrepancies that are undetectable to the human eye. But there are some things that human doctors currently still do better than AI. They are better at contextual understanding and using a patient’s past history to form a diagnosis. They are also better at identifying rare cases that the AI was not trained on. Today, the prevailing verdict among the medical community is to use both AI and human intelligence in tandem. Two weeks ago, Dr Robert Wachter, a leading academic physician and author of the book “The Digital Doctor” posted a Substack titled "Why do you still have a job?” In it, he identifies several key reasons why AI has not replaced humans in radiology... yet.

Currently, it seems like we are somewhere between 1990-1995 in the medical/chess world equivalence. At this point, AI is already objectively better at certain areas of medicine than humans are, but humans are still needed as a support system around it. To quote again Garry Kasparov “Never shall I be beaten by a machine! Never will a program be invented which surpasses human intelligence.” It seems that medicine and AI are currently at this junction. If we can learn from what happened in the chess world, AI will continue rapidly improving and soon be objectively considered significantly better. Does this mean that in ten years radiologists will be obsolete? Almost certainly not. Humans will still desire human doctors (likely working together with AI) in order to give them a feeling of empathy, trust and accountability. It is almost impossible to imagine a world where you get a diagnosis and treatment without a human doctor checking to make sure no mistake has been made. Proof of lack of trust in AI is the fact that Google’s Waymo is already objectively safer than human drivers and yet many are still hesitant to take rides with driverless cars.

There are a number of factors that differentiate chess from medicine, most notably that chess is a closed system while medicine is an open system. In chess, there is an objective best move in any specific position, the pieces are static and the best move is one that AI is able to find. In medicine by contrast, there is not always a specific “right move” that can be made in any given scenario. That being said, in many places, there may be a “best diagnosis” or treatment, one that humans are not able to always correctly identify. The advantage LLMs have here is high level pattern recognition, the ability to process thousands of different complicated data points in order to give the best diagnosis, something that a human brain is incapable of doing. In an open system, there is no “perfect” diagnosis, but there is one that is the best according to our current knowledge. Doctors here have the advantage of contextual understanding. AI might know the symptoms, but doctors know the patient.

What about investing?

If chess is a closed system, medicine an open system, the market is a reflective system. The market reacts and changes according to current data. The “best investment” today can change quickly depending on new information. And because investing is not a closed system, there is no obvious consensus that AI is better at investing than the best humans or even the S&P. However, this can quickly improve as AI improves. Looking forward, the important question is whether AI will be more able to more accurately identify stocks that are winners and if so, can they do it consistently enough to manage to not only be smarter than the average person but also a better investor?

In a bull market everyone is a genius.

One thing that is important to note is that since 2009, we have had almost 17 years of consistent growth. Even with the covid “crash” in 2020, within six months the market rebounded to its previous highs. This is one of the reasons so many people who invested and are currently invested in high potential growth stocks during this bull market outperformed the index. If we enter into a ten year bear market, where the market stays flat for years on end, it is very possible that the same people who had incredible returns during our current bull market will begin to underperform the market consistently. While some have claimed that they have invested using AI in the last few years and have managed to consistently outperform the market, this is logical as many growth stocks have outperformed the market in the same time period. The Nasdaq, focused on tech and growth has also outperformed the S&P in 14 of the last 18 years including significantly outperforming the S&P since 2022.

So can AI outperform humans when it comes to the stock market. The first thing to note is that the S&P 500 is really good. It routinely manages to outperform 90% of hedge fund managers over a ten year time horizon. This is often attributed to something called the “Efficient Market Hypothesis” which states that stocks are always traded at fair value which makes it impossible to correctly beat the market. For example, if Google is mispriced because people undervalue the potential of Waymo, the market will immediately correct as more information comes out. If the EMH is correct, AI has a limit of how effective it can be as a stock picking tool. The market not only is efficient, but it is becoming more efficient using AI tools and institutional algorithms. While many disagree with the EMH, the market speaks for itself and aside from a few very famous investors (Warren Buffett, Peter Lynch, John Neff etc) the market usually outperforms even the very best super investors. The fact that the names of Buffett, Lynch and Neff are so famous is due to the fact that they are the very few who did manage to beat the S&P.

Another thing to note about the market is that it