At The Garden of Forking Paths, Manuel Vargas has a -- at this point, rather old -- post on philosophical methodology, i.e., on what is the appropriate measure of success for philosophical argumentation. Given the nature of the blog, he's interested in applying it to free will debates. I'm not -- my interest is in showing how all the models he describes rest on a failed understanding of acting for reasons. Philosophers should, instead, adopt a fourth model, which I will try to sketch in some sparse detail.
Here are the models, as Vargas describes them:
Van Inwagen begins by rejecting a model of philosophical argumentation quite similar to what Robert Nozick has called "coercive philosophy"; on this model, one seeks to provide knockdown arguments from indisputable premises. As regards this sort of model, Van Inwagen points out (p. 36) that Nozick said that when he was young he thought that a philosophical argument is adequate only if anyone who understood the premises but not the conclusion would die! ...
On [the second] ... view, philosophical argumentation can be thought of as a kind of debate between two parties who have opposite views about the issue under consideration, where the goal of each is to convince the other to give up his position and adopt the competing view. (A weaker requirement would be that the other party to the debate switch from accepting the competing view to agnosticism; Van Inwagen doesn't explicitly consider this possibility. An even weaker requirement--or family or requirements--would be that the other party decreases to some degree his confidence in his view.) ...
Van Inwagen prefers a third model, according to which we understand philosophical argumentation as like an idealized debate between proponents of competing positions, where the goal is not to convert the other debater, but to convince an idealized "agnostic"--a person who is "neutral" in the sense that he has no particular antecedent inclination to accept the relevant position. ...
Vargas favours the third model, but I'll start by considering why the first two models go badly wrong, solely from the perspective of philosophy of action. First, it's important to note that coming to form a belief on the basis of some evidence is a perfectly respectable instance of acting on the basis of reasons. Evidence is the theoretical gloss on practical reasons generally; and belief-formation is a mentalized form of action. Although this is not philosophical dogma or accepted wisdom, I have never quite understood why -- it just seems obvious to me.
So, given that, on the first model, argument is supposed to give reasons that are so overwhelming strong that no one can possibly disagree with them or with their conclusions. But there is, in principle, no action which works like this. That is, there is no action the reasons for which cannot be defeated by some, even hypothetical, competing reasons. It is very nearly a conceptual truth that reasons for action are only reasons for doing that action (rather than some other) insofar as the reasons are stronger than the reasons for doing other available possibilities. But, in the actual world, there are always fewer available possibilities than possibilities in general -- the word "available" signals a host of pragmatic considerations which drastically limit the available avenues an agent can pursue. This actually does quite a lot of work in practical reasoning, as it allows an agent to narrow his focus on only a handful of possible actions, and thus a handful of sets of reasons. But, as should be clear, there is a multitude of sets of reasons which are never even considered, because the actions for which they could be reasons are never considered. It is thus possible that the reasons which back up the chosen course of action could be defeated by one of these sets of unconsidered reasons.
To make matters worse, it is again very nearly a conceptual truth that reasons only favour one course of action over another in a continuous way, not in a dichotomous "all or nothing" way. That is, when I choose to phi rather than psi, I do so on the basis that the reasons for phi (call them Rphi) are stronger than the reasons for psi (Rpsi). But this in no way implies either that Rphi annihilates Rpsi such that Rpsi are no longer reasons at all, or that Rphi is such a powerful set of reasons that no other set of reasons could defeat it.
So, applying these considerations to belief-formation on the basis of evidence, it follows that (1) some possible beliefs and their accompanying sets of reasons have not been considered at all; (2) no belief could ever be supported by enough reasons that the reasons for other beliefs are annihilated; and (3) no belief could ever be supported by reasons that would overwhelm any other possible set of reasons. Thus, arguments can never be all or nothing, accept them or face the 'splodey-head consequences, endeavours.
Oddly enough, it is exactly these three problems with the first model of argument which can be deployed to undercut the second model of argument. That is, someone who believes that p (Bp) is arguing with someone who believes that not-p (Bnot-p). Bnot-p can always, however, invoke exactly the above considerations against Bp, i.e., that Bp has failed to consider some other beliefs and their accompanying sets of reasons, that no belief (such as that p) is ever supported by enough reasons to annihilate another set of reasons (such as for the belief that not-p), and that no belief is supported by reasons that would automatically overwhelm any other possible set of reasons. So, without pain of any irrationality, Bnot-p could stick to his guns and refuse to accept any reasons offered by Bp, and not shift his confidence in his opinions one iota.
This brings us to the third model, which Vargas accepts. And the third model falls prey to exactly the same objection as the second: any agnostic, who simply did not believe that p, could give the same response to Bp as Bnot-p.
So, what has gone wrong here? My sense is that the problem is a failure to appreciate that belief-formation is an action, the same as any other, and thus is, in some sense, underdetermined by the available considerations. That is, when faced with a choice between phi and psi, I can, as long as I consider Rphi and Rpsi fairly and equitably, always choose to favour phi or psi without pain of irrationality. (I suppose this makes me some sort of libertarian with respect to free will?) And the reason is, as said, that practical reasons always underdetermine actions. It is choice, stemming from an agent's character, that completes the link from reasons to action.
Now, of course, choice can go wrong and character can be bad. But that's not the issue at stake. The issue at stake is what effect argument can have on a particular kind of action, and the answer is, unfortunately: not much. So, why argue? We need a fourth model to defend the point of argument at all.
This model, as I conceive it, holds that the purpose of offering reasons for a contrary belief or course of action is to get the opponent to take those reasons seriously and consider them. Simultaneously, of course, the purpose is to make clear to oneself the reasons for what one believes or does, a purpose which neither Vargas no Van Inwagen seem to consider. That is, reasons are given in order to engage with the opponent, test his reasons, and also test one's own. Persons of different, but equally good, characters may diverge in what they believe and how they act, but I think this is a consequence that we have to accept, given the underdetermination of actions by reasons.
Therefore, argument is not hopeless, but it is only part of the story. The role played by character highlights the importance of adequate education and intelligence for not only practical rationality but for reaching agreement on appropriate beliefs and choices. As that's a huge project, I'll leave it for another time.