Agreed, this is a good point. Here are some thoughts my contrarian comment generator had in response to this:
It’s also not a particularly lucrative place to apply the upper end of powerful agent intelligence. While ultimately everything boils down to algorithmic trading, the most lucrative trades are made by starting an actual company around a world-changing product. As a trader, the agent would commonly want to actually start a company to make more money, and that’s not an action that is available until you go far enough down the diminishing returns curve of world modeling that it starts being able to plan through manipulating stock price to communicate.
also, high frequency trading is not likely to use heavy ML any time soon, due to strict latency constraints, and longer term trading is competing against human traders who, while imperfect, are still some of the least inadequate in the world at predicting the future.
Regarding ML in high frequency trading, I’m not sure there is a significant impediment. What one would do there (and maybe someone already does?) is use ML to control the parameters of, and ultimately design from scratch, the algorithms that do the trading itself (so that the ML runs with high latency in the background while the algorithms operate in real-time).
Agreed, this is a good point. Here are some thoughts my contrarian comment generator had in response to this:
It’s also not a particularly lucrative place to apply the upper end of powerful agent intelligence. While ultimately everything boils down to algorithmic trading, the most lucrative trades are made by starting an actual company around a world-changing product. As a trader, the agent would commonly want to actually start a company to make more money, and that’s not an action that is available until you go far enough down the diminishing returns curve of world modeling that it starts being able to plan through manipulating stock price to communicate.
also, high frequency trading is not likely to use heavy ML any time soon, due to strict latency constraints, and longer term trading is competing against human traders who, while imperfect, are still some of the least inadequate in the world at predicting the future.
These are interesting contrarian comments.
Regarding ML in high frequency trading, I’m not sure there is a significant impediment. What one would do there (and maybe someone already does?) is use ML to control the parameters of, and ultimately design from scratch, the algorithms that do the trading itself (so that the ML runs with high latency in the background while the algorithms operate in real-time).