June 2026 Newsletter

Man vs. Machine

In 1996, Garry Kasparov defeated IBM’s Big Blue in a championship chess match. Garry won 4–2 in a regulation 6-round match. IBM did not give up. The next year, in 1997, armed with the computing ability to analyze 200 million positions per second, IBM’s Big Blue won the rematch. It was close: 2 wins for the machine, 1 win for Garry, and 3 draws.

In the summer of 1956, John McCarthy held a workshop at Dartmouth University on Artificial Intelligence. The name stuck, and we have been using it ever since. The timeline for developing AI follows the development of computers and the internet. AI development stalled in the 1970s and 80s. Then came the internet and graphical interfaces, and all of a sudden, every book ever written—along with research papers and more—was available on the internet.

Now, with ever-increasing compute power and speed, AI can find connections and sources in seconds that help solve problems. Many times, AI will try unique ways to solve problems that have stumped our best minds.

The Bitter Lesson

Rich Sutton, a Canadian computer scientist, wrote a paper in 2019 titled “The Bitter Lesson.” It turns out our minds often get in the way of letting AI develop and solve problems. Rich uses several examples of this. Big Blue won the chess match, but chess researchers were not good losers. They said that “brute force” may have won this time, but it was not a general strategy and not the way people play chess anyway. Researchers wanted methods based on human input to win and were disappointed when it did not.

He uses the example of computer Go. Researchers delayed the program’s ability to win by training it based on how researchers’ brains worked. Once they had the compute power, letting the computer search and learn won the day.

In the 1970s, DARPA sponsored an early competition in speech recognition. Two main schools of thought evolved to solve the problem. One relied on human knowledge—words, sounds, etc. The other method was statistical and looked for hidden patterns using much more compute power. Statistics and compute won. What we have today in speech recognition is all compute-based AI. Computer vision faced the same issue: researchers tried to teach the program based on how humans do it. In the short term this can work well and show progress, but it will always plateau.

“Gargantuan amounts of compute with search and learn will always win.”

This is a “Bitter Lesson” because our collective egos have to get out of the way. No matter how clever and imaginative our brains can be, we will never beat brute force. I was wrong. I thought that AI would need less compute on a percentage basis because, as it has been in the past, more efficient programming and programmers would make it more efficient. No, it won’t—at least so far.

There is a reason trillions of dollars are going into chips and computer AI farms. The smart money figured it out.

Thanks, Andy McClung, CFP®

Source: Tableau.com; Google.com/search

2026 Market Results – 05/30/26