The Microsoft AI for Good Research Lab published an article last year about how to do AI projects in the #AI4Good domain.
The current training academia gives to data scientists is not very practical. People learn data science with very sanitized datasets in very idealized problems, and when they meet reality, it’s usually a big shock.
I think the “Becoming Good at AI for Good” article is a good summary regardless of whether you’re doing AI for Good or more “commercial” applications of AI. A lot of the problems that AI for Good projects face (and are addressed in the article) are also problems that you’ll see if you’re doing AI projects for organizations that don’t have software engineering as their primary capability.
I especially like the part about getting things from model to impact. Before learning of this article, I’d usually send people to an article that was written at Google in the early 2000s.
This is a lot better, and likely to help you understand a lot about the real-world constraints of making AI happen.
The article can be found at https://www.microsoft.com/en-us/research/publication/becoming-good-at-ai-for-good