Wednesday, November 23, 2022

Social Obfuscations

 Hrishikesh Joshi has an interesting paper, Debunking Creedal Beliefs (PDF). The title is entirely misleading: nothing in the argument really requires that we are talking about beliefs (rather than, say, public claims proposed for belief or action, which are not the same thing); the argument is not actually debunking anything but raising a 'debunking challenge' that can often be answered, and that he explicitly gives a recipe for answering; and 'creedal' is just taken to mean here something like 'strongly influenced by social considerations'. Even sorting out the odd terminological choices, there is reason to be skeptical going into the argument. Actually debunking anything is extremely difficult; debunking an entire field of claims is almost always, and perhaps always, a category mistake -- debunking arguments are just not the sort of argument that can address an entire field of claims. It's actually very difficult to establish strict statements about broad fields of claims, in general; every claim has its particular quirks.

The 'beliefs' that Joshi is considering have three characteristics:

(1) any costs to individuals of their being wrong is negligible;
(2) they "fall under intense social scrutiny";
(3) in terms of evidence available, they are hard to verify.

One reason I think it's important to note that, despite Joshi's framing, there's nothing about this that has to do with believing is that everything here is actually public; the kinds of things that have all three of these characteristics are all claims that are publicly supported, whether they are believed or not. Society cannot directly incentivize (or disincentivize) believing; to the extent it does so at all, it does so environmentally, by making it easy to say such-and-such and therefore getting people used to such-and-such as something everyone says. Joshi also often tries to talk about his argument as downgrading certainty, but I think when we realize that the argument doesn't even directly apply to believing at all, what we are really looking at is a causal question of whether the social environment is making social testimony actively obfuscating (as opposed to things like 'merely approximate' or 'functioning as a loose practical heuristic'). That is to say, it's really an argument about the preparatory environment for belief, rather than belief itself. But Joshi insists on putting it in terms of beliefs, so we are stuck with talking about beliefs.

The 'debunking' challenge Joshi takes to be a 'blocking debunker', which prevents one from ever being justified in holding something. The paper is remarkably obscure about how this is supposed to work. Joshi says:

Beliefs formed with the goal (despite this goal being unconscious) of reaping social rewards and avoiding social costs are presumably not justified to begin with. Importantly for the debunking story, beliefs formed in this way are causally influenced by processes that are not robustly truth-tracking.

That is one whopping 'presumably'! Presumably how? All of our beliefs (and our public claims) are influenced by processes that are not robustly truth-tracking, so the claim has to be that beliefs with the above three characteristics are extraordinarily influenced by such processes. It's controvertible whether we always need the relevant influences to be robustly truth-tracking; if a process is weakly truth-tracking, as many social processes seem to be, it's unclear why one gets 'not justified to begin with' rather than just 'minimally justified but defeasible'. Almost every schoolchild forms beliefs with the goal "of reaping social rewards and avoiding social costs" (teaching itself is often structured on this principle), including a large number of claims that we don't generally regard as unjustified to hold, so 'presumably' here can't mean 'it fits our typical intuitions' or 'it makes sense of our usual behavior'. Indeed, that's not surprising, given the argument, since our social experience of learning things in general is unsurprisingly favorable to the value of social experience for learning, but then it's unclear where we get this 'presumably' at all. Endless numbers of broadly reasonable people, and many very reasonable people, act as if this 'presumably' is not at all 'presumably' true.

When one recognizes that the matter is really a causal question of the social environment, it becomes clear that what really matters is the causal story in each particular case. To be sure, one might still hold that claims with the above three characteristics are still cases where the causal story is often an obfuscating rather than an intellectually helpful one, but that has to be proven causally, and is not a matter of general debunking. One of the common problems with attempts to debunk is that debunking is not very discriminate; when not tightly constrained, it tends to spread very easily to other things like a fire. Joshi tries to argue that the debunking here is tightly constrained:

Many of our beliefs simply do not meet these conditions: either they are not incentivized and scrutinized by our communities, or they’re a priori obvious or easily verifiable, or the costs of being misinformed about the subject matter are sufficiently high.

This is certainly true, taken flatly; at any given time, a very great many of our beliefs (/public claims) are missing at least one of the characteristics noted before. The problem is that none of these characteristics are things that beliefs (or claims) have in and of themselves. What a community incentivizes or intensely scrutinizes is absolutely not stable; a claim that lacks this characteristic today might well have it next Wednesday. Whether something is easily verifiable depends on how clean or polluted one's evidence pool is (and we know that social influences sometimes incentivize tampering with the evidence pool), how easy it is to make certain arguments (and we know that some kinds of arguments are actively punished as a social matter), how accessible the means of verification are (and we know that this is highly influenced by social incentives to make them accessible). And the same goes for costs. (Indeed, with costs, Joshi later quite clearly understands this to be relative costs, because he talks about the first characteristic in terms of having costs that are greater than those that are the norm. But the relevant costs that are the norm change wildly over time and across cultures.) If any claim can be debunked in the way Joshi suggests, there is not a single claim that could not, at least in principle, become debunkable just by a change of the society around it. To be sure, many such changes are perhaps unlikely, and some might be actively detrimental to the society to which they would occur; but none are actually impossible. While most of our beliefs are in fact missing at least one of the characteristics at any given time, we have no beliefs that are such that they could never gain such a characteristic over time. Society can incentivize or disincentivize pretty much anything that can be put forward in public. It is highly suspicious to say we have a debunking argument, or in this case, a 'debunking challenge', that depends entirely on notoriously variable extrinsic circumstances, particularly given that we are supposed to be talking about a 'blocking debunker'. I could form a belief one week and you could form the same belief with the same evidential grounds next week, and if society had had a revolution between, my belief might not have been 'debunkable' but your belief might be 'debunkable'. Does that make sense? I don't that makes much sense. This does not seem to be a robustly truth-tracking form of 'debunking'.

In any case, Joshi gives a recipe for meeting the challenge, which is why the paper is not really a debunking argument at all. The recipe goes:

(1) Identify the groups that influence your social and professional success.
(2) Determine which claims are incentivized or rendered costly by these groups.
(3) Reduce confidence in any claim with the three previously noted characteristics.

Notice that the third step is reducing confidence, not rejecting; this is a sign that nothing has actually been debunked. What the argument really is, is an argument that everybody should be a contrarian and depreciate claims that are appreciated in a certain way by their society. While it's amusing to think of how that would work, this doesn't seem actually to get anyone closer to anything true; it, remarkably, requires us to be even more influenced by the society around us than we already are, since it makes a system of determining our own beliefs (/public claims) with an eye to what society is saying.

When we see this, it seems clear that Joshi is getting things the wrong way around. What's actually important are specific and definite processes of social obfuscation, which we should look for and not be tripped up by; which claims happen to be entangled in such processes at any given time is largely irrelevant. The beliefs are innocent, or at least not able to be proven guilty by an association that they themselves do not determine. They should just be assessed as they otherwise would be, by evidence and argument. Not they but the obfuscating processes are what need to be regarded with suspicion.