https://en.wikipedia.org/wiki/Inductive_reasoning

Or, to mirror your tone:

Being paranoid does sound like an issue! However, just because other people's personal incredulity hasn't been a great form of evidence in the past, that doesn't mean it isn't a good form of evidence at the moment. Got any reason to believe it isn't?

Perhaps someone can find some evidence that allows for deductive reasoning on this subject. I would really like it. Otherwise, all we've got is competing lines of perfectly valid incompatible inductive reasoning. Then people choose the one they like or fear the most.

That's a scenario that my priors suggest is not likely to be useful for producing valuable models of the future. Instead, it's a scenario where the primary outcome I expect is for confirmation biases to rule the day.

Such evidence would be admissible in a court of law, and would probably lead to the breaking up of Google (and Alphabet, and various holdings) under antitrust. Yes, that'd be nice, but you're asking for a lot; even given the premise, it's unlikely to materialise until the statute of limitations has passed, if that.

> Otherwise, all we've got is competing lines of perfectly valid incompatible inductive reasoning.

Not quite. Because you can ignore the lines of the reasoning, take into account the evidence, and then come to a conclusion yourself. If you don't quite trust yourself to do it properly (and who would, honestly? the bias you've highlighted is enough on its own, and it's just one), there's mathematics to do so: Bayes' theorem. (Of course, you have to estimate the conditional probabilities properly, but I find it easier to notice when I'm leaning on the scales when there's a level of indirection like that.)

Most predictions are made with inductive reasoning. If your comment doesn't contain some flaw (even if not the one I pointed out), then it suggests that predictions should be much harder than they actually are. Super-predictors should do not much better than chance. And yet, we find that, in reality, even the little-trained masses do pretty well: https://predictionbook.com/predictions You can see confirmation bias in the graph there, but it's not as significant as my model of you expects.