The way I see it, there are three main drivers of asset manager returns (excluding fees), and adding AI to any of them doesn’t help much in terms of improving outcomes for investors (non-institutional investors.)
Portfolio Construction: As long as your assets look like stocks or ETF funds, you don’t add much value by going beyond mean/var optimization with constraints. Most investors don’t need and can’t access exotic assets, and for those who can it’s less a question of optimizing the portfolio than it is of getting access to the investments in the first place. This is an easy problem in the first order approximation, and you don’t gain much by adding complexity.
Execution trading: Over time, trading adds value by 1) providing liquidity, and 2) being better than the other traders in the market. 1) Marginal returns to providing liquidity in already liquid markets are small, and in illiquid markets its not a problem for an algorithm but rather needs major innovation in how the market functions… 2) this is a zero sum game, and as the arms race of faster and smarter algorithms accelerates it converges to zero for all participants who survive.
Asset/Security selection: This is probably the area where AI could add the most value. It’s also the area that has proven most problematic, such that a huge amount of total assets have migrated away from active selection and into passive index investment. Vehicles that paid to take security selection risk, such as hedge funds, are already pretty actively using algorithmic strategies. Consensus thinking is that for the typical retail investor, it’s not worth the risk for the incremental return. Taking more security level risk, assuming the methodology is “smart”, will increase returns. It will also increase variance of returns and concentration within individual assets. For a very large chunk of total financial assets being managed, even an AI that was good at security selection would not be desirable.