I feel like I only understood half of the video, but here's my interpretation as far as how it related to the stock market. I believe that it's not that people don't understand what the algorithms do - after all, someone had to design them. The potential problems come up from the fact that we don't necessarily know how it will react in various situations. For example, it may work perfectly well for certain kinds of companies, but if something different comes up out of left field, the algorithm may not apply as well to it, especially if there aren't any safety guards in place. Maybe all of a sudden, certain criteria crop up in some junk bond and the algorithm purchases tens of thousands of dollars of it when in reality, it's ridiculously risky and the company's position quickly slides and they become worthless.
Still, they're so useful just because decisions can be made so quickly. If you can be the first to discover some worthwhile stock, you can buy it cheaper than anyone else, as after you've bought it, demand has gone up and its stock price has risen. Micro-seconds could make huge differences here. I'm sure that given enough public knowledge of a company, a human could make a more informed decision, but that's in a matter of minutes at best.
The Fly book example was interesting. Looking into it, two vendors had algorithms based on each others selling prices without any safety guards. So one would always underprice the other by something like 0.5%, while the other, not having stock, would base their price on the others' price times 127%, figuring that given a sale, they can then afford to buy the book cheaper and then sell it. This gave rise to an infinite feedback loop.
