If you dislike a _b_ version of an a/b tested site to the point that you feel deeply dissappinted by it, were the people who designed the experiment operating with a malicious intent?

I don't think facebook engineers or data scientists care what people feel when they're adjusting their models, they only care about time spent on their platform or some other overly reductionist metric. One could argue that chasing super-efficiency will always result in abominations, but I think its hard to say that someone is doing something imoral by just doing the human equivalent of gradient descent in the optimization problem of driving engagement. That's not to say that we shouldn't have a conversation about this and not figure out ways to make these super efficient and powerful companies find ways to be satisfied with organic engagement.

There's a difference between disappointment about a product (which is fine) and doing an experiment to test that might knowingly create bad feelings in someone's life in one cohort of your experiment.

What if an almost-suicidal person got the "b" version?

A/B testing should be used when you think, in good faith, that BOTH "a" and "b" versions could be good and want to know which one works better, not to confirm a hypothesis that "b" has a negative impact on users' personal lives.

Surely, people understand that Facebook makes changes to their website, and that these changes are based on data. That is all an experiment is. If Facebook wasn't allowed to experiment, they would never be allowed to change their interface or recommendation engine at all.

There's a difference between experimenting on the product, and conducting an experiment that deliberately puts one cohort of users in an environment hypothesized to create negative emotions.

A/B tests for the case where you honestly, in good faith, think both A or B could be good designs, but you don't know which one is better.

A/B testing is for product design changes, not psychology experiments. It is NOT for the primary intention of a psychology experiment with a group of subjects that didn't consent to a study, where either A or B has a suspected negative impact on emotions, and you're using A/B testing to confirm or deny that hypothesis.

It's still a product decision though, especially the way the top comment frames it. It's just one that happens to be centered around emotional language. I suspect that an experiment that deals with more or less political content would have a similar effect on people's mental state.