Monday, June 09, 2008

On Devil's Advocacy

Eliezer Yudkowsky has an odd post arguing against devil's advocacy, i.e., against playing the devil's advocate in argument. At least, that's what the post is supposed to be about; I'm not sure that it proves anything but that Yudkowsky himself would be an extraordinarily bad devil's advocate in any argument. Devil's advocacy doesn't involve finding and putting forward any argument, however absurd, for an opposing position. It's worthwhile to remember for a second the origin of the phrase. The Devil's Advocate was an informal name for the Promotor Fidei in a process of canonization. Canonizations are processes that can't be entirely isolated from human passions and enthusiasms, so it became clear early on that there need to be safeguards to prevent rash decisions motivated more by personal taste and interest than by concern for the Faith. Some of these safeguards are purely procedural, and from them we get the standard process Rome uses. The Promotor Fidei, however, is a special kind of safeguard: it is his job to guarantee, to the extent reasonably possible, that the Faith be protected from human rashness. This he does by raising any doubts and difficulties that a reasonable person might raise against the cause of a saint: he tries to find natural explanations for purported miracles, he tries to uncover selfish motives behind deeds of apparently great virtue, and so forth.

In argument someone playing devil's advocate takes on an analogous role. As the advocatus diaboli promotes the interests of the Faith in a field full of human passion and politics, so devil's advocates promote the interest of reason in a field full of human bias and oversight. Their role is not to rationalize opposing positions, any more than the role of the advocatus diaboli is to do the devil's work. Rather, the role they've taken on is to identify, clearly and straightforwardly, the attractions and strengths of the opposing arguments; and to identify, equally clearly and straightforwardly, the unattractive and apparently weak features of one's own argument. They are there to guarantee, to the extent possible, that bad arguments for correct positions do not slip through simply because they get the right answer (or what is thought to be so), and that good arguments for incorrect positions are not facilely dismissed simply because they get the wrong answer (or what is thought to be so). Yudkowsky is right that people who play games by thinking up arguments, however absurd, for a position, are simply being irrational; but this is to no point whatsoever: everyone knows that the devil's advocate is supposed to come up not with any old argument but with good or at least reasonably plausible arguments, arguments with at least some genuine strengths. People play devil's advocate for a reason, not simply in order to start making things up without any rational restraint. There are less elaborate and roundabout ways to play-pretend.

Yudkowsky does recognize that devil's advocacy can shake one out of a rut, but, oddly, his argument for the limitations of this seems to assume that people are only shaken out of a rut when they play devil's advocate for themselves. But this surely overlooks the fact that devil's advocacy is a social thing: one usually plays devil's advocate for another, to aid them in refining their argument, or to help them to fill in gaps that they might not be in a position to see; and when we do it for ourselves we are simply trying to imagine ourselves in the place of such a person. Nobody has exactly the same perspective on every argument, and it will often be the case that one person, while agreeing with the gist of the argument, sees a potential weakness or lacuna that another does not. Devil's advocacy is one way (only one of a great many, and not the most important, but certainly one way) that reason takes on a social aspect, breaking free of the confines of a single brain. Human beings are rational, and therefore social, animals, and devil's advocacy is an expression of the sociality of reason.

UPDATE: Yudkowsky has added a note indicating that he just hadn't considered the social aspect; that's certainly fair enough.