It's actually a great article with a HN title that misrepresents what it is actually about. Anyway, with that proviso, here is my take: So long as the bots are

It's actually a great article with a HN title that misrepresents what it is actually about. Anyway, with that proviso, here is my take: So long as the bots aren't conscious there will be no bot-dominated stock market. Too much information is wrapped up in conscious thought that isn't easily expressed in stock market pattern matching. The article mentions that the line between active and passive isn't as clear as people that make it out to be say it is, and that is true, but it's also incomplete. Certain events are conceivable to humans that are not conceivable to machines. Short of simulating humans, anyway. For example, I as a human know that there is a non-zero chance that the DPRK and USA get into a nuclear war. I can know that Trump defaulting on Chinese held US debt is something that is at least on the table. The quants can try to pull in percentages from experts all they want (and they were calling my nuclear weapons arms control friends a year ago asking for percentiles) but it isn't the same thing because the model is still fundamentally concerned with statistics and many of the events we understand as humans have implicit dependence to each other that is hard to model. It's far easier to employ a small number of smart people and have the models serve and be tuned by them than to have the models actually run everything. Plus you don't even really need them sometimes. Take Bitcoin, for example. The gains were obvious. The incentives were completely aligned and the friction was in the buy side and it was temporary. I didn't need to model out Bitcoin to buy it at $4 CAD. I just had to think about it from first principles, something that is currently not possible with ML.