> This is not a new discussion, the research community has long ago established a near-consensus about what are the appropriate ethical boundaries, permissible tradeoffs, and non-negotiable safeguards when studying human subjects, and this goes waaaaay beyond that.
Unfortunately, our definitions are extraordinarily broad and technically almost anything is human research by the letter of the definition. In practice, we don't call everything that interacts with data about people "human experimentation", but figuring out the exact line is very hard.
E.g. I am a teacher. I do things differently from year to year. I do my best to gather the best metrics on this time series data that I can in order to inform my choices on what I consider best practices. Do I need to get an IRB involved? My institution doesn't have an IRB.
I really wish there were high quality controlled experiments in education that I could rely upon. But the vast majority of the educational research is crap.
Think about, say, Khan Academy. I have a hard time imagining a world where interacting with millions of kids on a protocol is OK, but the second we take some basic metrics to try and figure out how well it works, it's bad. I'd arguably ethically one needs to perform a certain basic level of monitoring and optimization of any intervention.