You kinda just did. You explained "that". You didn't explain "how".

Have you ever actually had trouble describing these sorts of things to people who display enough technical ability to understand what you're saying?

The basic claim is incredibly simple, and can be explained with an analogy: suppose you have mugshots of people from a jail in a poor black community, and also pictures of people from the Rotary Club in an affluent and mostly white neighborhood, and you ask any sort of algorithm to predict violent tendencies on the basis of picture. Obviously, given that the data was chosen to have very strong correlations between race and criminality, it hardly even matters what the algorithmic details are: if it can pick out skin color at all, that would be a good way to solve the task.

It doesn't require detailed knowledge of convolutional networks, multi-head attention, objective functions, or anything of the sort. The logic is purely an example of "correlation does not equal causation". You can simply assume that a ML architecture that's even worth using on the data has the capacity to pick out blatant correlations like that. Then you can reason accordingly.

And point out that a correlation doesn't have to be chosen in order to be present in the data.

Likewise, if someone is claiming that human societies did not engage in war prior to 15,000 BC, it's not terribly difficult to understand the types of evidence that would be relevant: skeletal injuries from diverse communities around the world. If someone points out a mass grave with grievous injuries from long before that, the response ought not be "shut up, the experts know what they're talking about, I've studied this for decades?" but rather "yes, but the rate increases by at least so-many-fold" or "I don't know that paper but I don't think it's the norm" or "this doesn't meet the definition of war, for such-and-so reasons".

Anthropology is a comparatively shallow science. Unless you start getting into the weeds with radiocarbon dating confounds or arguing about the degree to which arrowhead style reflects an invasion of people rather than an invasion of culture on the basis of quantitative linguistic analysis, it's pretty comprehensible with at most a little primer here or there to orient people.

Some sciences are deep. If you're going to talk about resolutions to the disparity in cosmic expansion, though, it's another matter--then, yes, you need a lot of background. But even then, someone who actually knows the subject can still point at the relevant literature. Or to articles written by someone like Ethan Siegal, who does his best to at least explain the issues for bright, motivated people to understand at least the shape of the problem.