Why do AI projects fail?
AA: I don’t know if you’re seeing the same thing, but I hear a lot of statistics around failure rates in AI – something like 80 or 85 percent. Assuming you agree that there’s some truth in that, what do you think is the main reason? Technology is cheap and easy to obtain. Data is everywhere – we’re drowning in it. Where are organizations going wrong?
KB: Well, I’d like to know the origin of that number, first of all. As I said earlier, I believe strongly in the “fail fast to learn fast” principle. If you fail four times out of five in order to get something right one of those five times, then you have an 80 percent failure rate, but I think that’s acceptable. If you just deploy the first thing you build, that’s probably not going to be a good idea. You probably want to find the failure cases, or what they call edge cases. A perfect example of that is self-driving automobiles. There...