It’s a common observation in psychology and behavioral economics that normal human reasoning often violates formal norms. Many people might, for example, judge it more probable that a librarian will be an introvert and wear glasses compared to a question about librarians just wearing glasses. It should be the other way around, since among librarians that wear glasses, some will not be introverts. Vivid conformity to a stereotype, compared to a more generic situation, can make a description seem more representative or more likely.

In at least some contexts, though, I am not sure that describing this as a violation of norms of reasoning is completely accurate. Just the fact that somebody is giving you a more detailed description can be relevant information to take into account in assessing probabilities.

Consider the claim that nature is shaped by an Intelligent Designer. We can take a generic version of this, where all that is claimed is intelligent design—without specifying anything about the identity of the designer, the purpose behind the design, or anything. This would include aliens as designers as well as gods. Evil designers as well as benign ones, incompetent designers, designers who optimize parameters no human would ever care about. Designers who work by special creation, designers who guide a process of evolution. (Intelligent design proponents who want to hide their entanglement with religion often go in this direction.)

Alternatively, we could have a more specific designer. For example, a biblical God. We would presumably know something about the character and purposes of such a designer, since we can flesh these out with reference to certain religious traditions and claims of revelation. This knowledge about character and purpose would also allow us to infer some things about the designed world. (We would not necessarily know a lot, but certainly more than with the completely generic designer.)

Now, here’s an interesting observation. Clearly the specific designer claim picks out a subset of the generic designer claim. Therefore the generic designer is more probable. On the other hand, as Elliot Sober has long argued, there is something pathological about a completely generic designer claim, when we can say next to nothing about what sort of design they would create. That sort of designer cannot be part of any genuine explanation of phenomena. We need some independent evidence that, among other things, makes the designer more specific, in order to even begin taking the claim of a designer seriously. (I’m not entirely comfortable with Sober’s approach, and still less comfortable with Bayesian approaches in general. But this is fine for my purposes here.)

You don’t have to be a Bayesian or a Popperian or adopt any detailed philosophy of science to agree that it is usually a good thing for scientific claims to be specific rather than generic—even setting aside the relevance of specificity for future tests of the claim. Specific claims say more, and we want our science to have more explanatory power, not get lost in almost-meaningless generic realms. And yet, getting more specific only reduces the likelihood that we are correct.

This might not always be a problem for everyday reasoning. If somebody makes a specific rather than a generic claim, this is good because there is less of an issue about whether the claim is sufficiently meaningful. Moreover, we expect that if someone is in a position to know about something like an intelligent designer, this is because of some specific information obtained, whether through divine inspiration or empirical testing or whatever means. So revealing specific details, while it may seem to diminish the probability of the claim, can also be read as a sign that the claimant has access to the sort of specific information that would help make the claim credible. In other words, the mere fact that extra details are offered can itself be relevant extra information, tipping the scales in favor of the more specific claim.

So perhaps our inbuilt cognitive bias in this case is not all bad. (Mind you, it can easily be exploited. Consider how the tales that ended up in the New Testament seem to have had a history of pre-canonical embellishment, with more specific details added to later versions of the stories.)

But what, then, about more formal contexts of reasoning? Say someone says that she believes in a God. But this is not necessarily a fundamentalist kind of God, just that there is some generic personal supernatural purpose behind existence. The details of this God are vague to nonexistent, and purposefully so. (I’m sure many of us have run into such God-claims.) Now is such a person, with her generic God, making a respectable claim?

Possibly. There is a certain cognitive modesty here—she disavows the sort of knowledge Bible-thumpers claim, though even they may well be right in the end. Her generic God has to be more likely than the biblical God.

And yet, I’m sure skeptics also feel uncomfortable with such a generic God-claim. It’s mush. There’s very little content to it. It presents next to nothing to get a handle upon. How the blazes could you even know that such a generic God could exist? If you like Bayesian approaches, this kind of generic claim makes prior probabilities do all the work. But the arbitrariness of priors is a common (and I would say often fatal) flaw of Bayesian approaches.

So I’m not sure where we end up. I don’t know of any formal way to resolve the tension between high-likelihood genericity and low-likelihood specificity here. It think we have to strike a balance, and that balance is shaped by the case at hand and our history of cognitive performance. Personally, being a stereotypical science-type, I lean toward favoring specific claims. If somebody wants to argue for a God or a Designer or whatever, fine. We need some degree of specificity before we can even properly get started on criticism.