I think that the general advice of "find something that bugs you and then fix it" is a particularly bad way of building a business because you run a very high risk of building something that nobody else wants. In general, "you" != "your customer".
My advice would be to go out and find some group of people/companies and then ask them what problems are they currently facing that are not currently ideally solved (stressed out moms, small businesses, etc. It really doesn't matter, just a large accessible group who may have problems that could be solved via ML).
This way, you don't need to know really anything about the domain because the interviewees will know the space well enough and will tell you what's wrong with it (and will most likely list ideas on how to fix it.) This will also remove your worry about the space being crowded because even if it is, if your interviewees are saying that there is still an unsolved problem, then there is still opportunity for you.
Another idea is to attend a startup weekend and learn how to quickly test assumptions/prototype and then try and replicate the process during the week. (e.g. run one full validation sprint per week, every week on different ideas until something gains traction.)
My advice would be to go out and find some group of people/companies and then ask them what problems are they currently facing that are not currently ideally solved (stressed out moms, small businesses, etc. It really doesn't matter, just a large accessible group who may have problems that could be solved via ML).
This way, you don't need to know really anything about the domain because the interviewees will know the space well enough and will tell you what's wrong with it (and will most likely list ideas on how to fix it.) This will also remove your worry about the space being crowded because even if it is, if your interviewees are saying that there is still an unsolved problem, then there is still opportunity for you.
Another idea is to attend a startup weekend and learn how to quickly test assumptions/prototype and then try and replicate the process during the week. (e.g. run one full validation sprint per week, every week on different ideas until something gains traction.)