> There are a million guarantees that what is being done is thoroughly and totally meaningless.
There are a million subtle ways to influence behavior and thinking. It's very likely that the scientists aren't able to properly measure the effects (because it's hard to isolate the control variables), but the resulting knowledge won't be zero. I agree with you that many scientists overestimate their understanding of the underlying mechanisms - IMO it's still too much hand-waving - in reality our knowledge about social psychology is very limited.
The discomfort is partly because you can make yourself like or hate people you don't know personally and trying to come up with bad/good things to say about them is one approach how to do it. You start process insulting them and you can see yourself starting to feel differently and some might feel good about the process and where it leads.
You know how people close themselves into bubbles that says only bad things about some outside group and then become to genuinely hate that outside group? It is the same, but is small scale, temporary limited and easier to detect.
It certainly can be frightfully easy to lead people down a path into forming tons of negative associations about some group or topic, and that will result in a trained emotional response pretty reliably..... but it's just an emotional response. It doesn't mean anything, and can't be used to support or detract from anything in terms of what is true.
Emotional responses are more important for your belief system and vice versa than you seem to realize. "what is true" is based on your belief system. It may not seem visible, but all those small interactions shape our emotions and belief system.
It's like training data for a neural network which has functional components and meta-learning capabilities (this is essentially what a brain is). Saying "this training data doesn't affect my neural network" or "this part (emotions) of my neural network doesn't affect another part (reasoning)" is not correct when training neural networks. Why would you assume that this is any different for your biological neural network? Every interaction shapes your brain, however small it seems.
Children will see your negative emotional response to them and will react to it. The negative or positive emotional response on someone determines a lot about your interaction with them and is hard to hide for most people.
Asking a person to assign negative terms to photos is as consequential as asking them to whistle or drum their fingers on a desk. It is meaningless and utterly amoral in content. We're not talking about even applying the terms to actual people, but to images. And they're not being monitored or published. There are a million guarantees that what is being done is thoroughly and totally meaningless.