What motivates me to write this post about "What is Science"? -- The industry, particularly in the area of AI, is fairly confused about what "science" or "research" is. My observation is that it's a pervasive idea that "scientist" or a "research" is someone who can do any of the following: 1) train a model, 2) evaluate a model, 3) write pytorch code. In short, anyone who can do AI/ML models would have the "scientist" or "researcher" title. I think it's very wrong, and I actually agree with Elon Musk's X message: https://x.com/elonmusk/status/1950254103474446728 (or see screenshot below):

I would say majority of people who are doing the "AI modeling" work is actually engineering work. It's just a special kind of engineering work dealing with AI modeling, which sounds fancy but are just copying from existing approaches from published papers.

For example, I got a question from an "applied scientist" job interview about "how to solve the imbalanced class binary classification problem". Perhaps the expected answer is to increase the loss for the under represented class during training. There are other answers too and AI can answer it in a split second. --> The interviewer's question was about AI modeling, but the question was not about science, because the answer to the question is well-known available in many sources, and anyone (including AI) can answer it easily. I don't think being able (or not able) to answer that question can determine whether you can be a good scientist.

For things with known answers that can be easily found (such as the "imbalanced class binary classification problem"), it's not about whether you know the answers or not. It's about how you do it fast, efficiently, securely, on scale -- and that's the engineering domain. I think that's what Elon Musk is trying to get at -- that his employees should focus on the engineering aspect of making things run fast, efficiently, secure, on scale, etc, instead of pretending to do "science" work with known answers already.

To me, what count as "science" has to do with 2 things:

1. To ask an important and worthwhile research question that nobody has answers to.
2. To apply scientific methodology and find the answers to the research question and be able to reproduce the results.

For example, measuring the value of AI that's deployed to an organization, in terms of productivity gain, time saved, etc -- no one has the answer or has a tried-and-true runbook to follow. And if one can find the answer to the question and be able to defend it and reproduce it, then that's what science/research is about, even though it doesn't involve building AI models.

Here's article: https://matt.might.net/articles/phd-school-in-pictures/ I recommend to end the post.