Yes. But also I think it's the sort of white elephant project that can only thrive in a ZIRP environment. Once money has an actual cost, and investors require

Yes. But also I think it's the sort of white elephant project that can only thrive in a ZIRP environment. Once money has an actual cost, and investors require actual returns, there are 1000s of better business models than this. Hopefully there is consolidation & regulation in the space before we have too much testing in public gone wrong. In my mind the entire self driving industry is trying to apply ML/AI practices from use cases that just require ~51%-ish "right more often than wrong" precision onto driving which needs 99.999+%. If some of your phones images are mis-tagged as a cat instead of dog, or some of your cat photos are not tagged as a cat, it's not a huge issue.. In quant trading as long as your trade makes money 51+% of the time, you are going to accumulate returns in the long run. Car license plate readers for speed tickets / tolls / etc have a similar "pretty close / pretty good" requirement.