Korva is a scroll of the internet’s best ideas.
No history to train on
Patterns without history
A big share of what AI does is knowing history and recognizing and extrapolating patterns. There will always be things with no history to train on, and some people have a better understanding than others of the probability distribution that defines future events.
When Lehman Brothers collapsed in 2008, there was no pattern for the end of the world. Decisions rest on data, analogies to past experience, and supposition, as a Harvard epidemiologist put it during the pandemic, and in 2008 there was no data and no prior experience, only supposition. Whether AI can think that way is an open question.
Indexing's lesson
Indexing put a lot of people out of the equity business by showing they couldn't do what they claimed, and most active equity investors underperformed the averages. AI may do the same to another group whose talents are not as great as they purport. Whether the list of things AI can do is limited or unlimited is a big part of the question.
A role for judgment
One thing experienced investors may still offer is avoiding bad people. Sometimes you talk to someone and for undefinable reasons it doesn't feel right, and the hair on your neck goes up. If AI has no hair on its neck, there may be a role left for judgment.
How do AI models learn from historical data?
AI models are trained on very large collections of examples, and training adjusts their internal parameters until their outputs match patterns in that data. Once trained, a model's outputs reflect the distribution of the examples it saw. Inputs far outside that distribution are a known weak spot, which researchers call out-of-distribution data. Events with no precedent in the training data therefore give a model little to work from.
How did index funds challenge stock pickers?
Economist Paul Samuelson argued in a 1974 article that someone should offer a fund that simply tracks the market. John Bogle's Vanguard launched the first index fund available to ordinary investors, which opened in 1976. Over the following decades, studies of active funds kept finding that most trailed their benchmarks after fees. Index funds now hold a large share of US stock fund assets.
Why do first impressions of people carry weight?
Psychologists Nalini Ambady and Robert Rosenthal reported in 1993 that ratings based on silent clips of a few seconds of teachers matched the end-of-semester evaluations given by their students. They called these quick judgments thin slices. Later studies found thin-slice ratings predicted outcomes ranging from sales performance to courtroom behavior. The researchers did not claim that such impressions are always right.
More from My First Million
