There is tension between foundational AI work where we interested in well-defined computational problems or even prototype let’s-try-this-out type research, versus products/ techniques that everyone is eager to commercialise.
Mistakes and bias will creep into both, and regardless of the intentions of the fundamental-research scientists, transfer of knowledge and technology will happen. And there’s also the danger that projects with dubious motivations will be attempted in both.
But in the academic setting, if done properly, there’ll be a clearly defined set of assumptions under which the results are to be tested and interpreted. Anything beyond those, all bets are off. Not that this absolves anyone of accountability, just worth noting the differences from an epistemic viewpoint.