Why AI projects fail
AA: We’ve talked about developing these products, but we also should mention that, often, we read articles that cite a Gartner report in which it states that supposedly 85 percent of projects fail in the broader data science and AI space. Do you agree with that, and if so, what is happening?
JTW: That’s a great statistic that you can just riff on for quite a while. I believe Gartner has one of those statistics, but also I’ve seen a statistic in the context of MLOps platforms, basically complaining that models don’t make it into production. To me, those are different things because just because a model doesn’t make it into production doesn’t mean the project is a failure.
As an analogy, I work for a drug company. Over 80 percent of vaccines fail, but I don’t hear people complaining about how 80 percent of vaccines fail, because that’s the nature of science. You do your best with the data. You do your...