Not so Ignorant After All: Metaknowledge Matters More than Knowledge

In one way or another, democracies put political power in the hands of the people.  A perennial worry about democracy, voiced almost as soon as it was invented, is that it places power in the hands of the ignorant. To some extent, this worry reflects an elitist condescension, but there isn’t much doubt that most potential voters know astonishingly little about policy and politics.

For example, less than half of all Americans can name a single Supreme Court justice. Even during election years, most people can’t identify the candidates standing in their district; most cannot name the party that controls congress; they routinely report the opinion that the US spends too much on foreign aid, instead suggesting as appropriate a percentage of the budget that vastly exceeds actual spending; 40% can’t name the United States’ enemies during World War II; at the height of the cold war, only a minority correctly reported that the Soviet Union was not a member of NATO. More than a third of actual voters are “know-nothings” Ilya Somin claims. Summing up the evidence, Brennan suggests that “most people know nothing, and many people know less than nothing.”

The world loves to mock Americans and their ignorance, but political ignorance seems just as widespread on the other side of the Atlantic. A survey of people who voted in the Brexit referendum showed that neither Leave nor Remain voters knew much about the EU; on some questions they performed worse than chance.

Widespread voter ignorance seems to be a serious challenge for democracy. Effectively, democracy leaves the most consequential decisions (how, and whether, to respond to a pandemic, for example) in the hands of an uninformed mob. Representative democracy mitigates the harm, we might think, by interposing representatives between ‘the people’ and decision-making, but this is protective only so long as the representatives are themselves informed. As Plato pointed out, democracies may elect people whose appeal to voters is not dependent on their competence or expertise. The Trump election and Brexit referendum seem illustrative.

Voter ignorance is the central plank of Brennan’s case for epistocracy. Decisions (whether the decision to implement a policy, or the decision to elect someone with the power to implement policy) will be better if they are informed. That fact gives us a reason to replace rule by ‘the people’ in general with some form of rule by the competent.

I suspect the case for epistocracy relies in important part on an individualistic conception of epistemology. I want to replace that conception with one on which knowledge and its production is socially distributed. I won’t attempt to spell out how knowledge production works on my picture, but a central plank of the overall account is flexible and adaptive epistemic deference. People – all people, including the experts – don’t and can’t know all that much. But they don’t need to: they rely on others to do the knowing for them.

Voters may not know much, but they have ‘metaknowledge’: they know who knows on their behalf, and/or they know how to find out. Metaknowledge is often better than knowledge: it updates flexibly and in realtime. The person who takes themselves to know may resist updating their beliefs in the face of new evidence, but the person who outsources their belief to the community of experts metaknows whatever the experts agree on.

The case for the existence and the adaptiveness of metaknowledge – or for the broad and pervasive outsourcing of belief to other agents and even to the environment – is extensive. It rests on experimental evidence (for instance, that people attribute knowledge to themselves when they know they can access the internet to retrieve the information, or when a partner knows the information), on work in the philosophy of science on how little individual scientists know about their own papers and on how the institutions of science are responsible for knowledge production, on work in anthropology and cultural evolution on how representations may guide behavior without being accessible to individuals and on work in political psychology on how attitudes to policy are often better predicted by knowledge about who endorses the policy than by its actual content, and much more besides. It is, I claim, the best explanation of an enormous range of data from many different areas.

Of course, contemporary democracies do have many ills, and the Trump election and the Brexit referendum are plausibly explained by some of them. In the US, and in many other countries, people have put spectacularly ignorant individuals in power. They haven’t deferred appropriately. But the problem may not be their use of epistemic deference. They may have responded appropriately to the higher-order evidence we use to guide deference. The problem is that this evidence is misleading. Given misleading evidence, rational agents respond by forming inaccurate beliefs.

If that’s right, the best response to the problems of democracy is neither to attempt to make agents more rational (they’re fine as they are) nor to limit the franchise. It is to engineer the epistemic environment, so that cues for epistemic deference better align with good reasons to defer. That’s not paternalism or manipulation: if these cues are genuinely (higher-order) evidence, its making sure that people are given good evidence and not bad. It is, in effect, to ensure that they’re not being lied to.

Of course, it’s much easier to say we should do this and much harder to come up with concrete proposals. The suggestion also leaves a great deal of further work to do on defending the pervasiveness and rationality of epistemic deference and making good on the claims that engineering the epistemic environment need not be paternalistic. But those are topics for another day (in fact, for the whole book, in which these ideas will be fleshed out somewhat more fully).

Picture: Vote by William Vandivert.