Wittgenstein deploys the idea of criteria in a number of different contexts but the context which has been most influential is that of grounding knowledge of other minds. The influential Wittgenstein exegete PMS Hacker, writing in the Oxford companion to philosophy, defines a criterion thus:
A standard by which to judge something; a feature of a thing by which it can be judged to be thus and so. In the writings of the later Wittgenstein it is used as a quasi-technical term. Typically, something counts as a criterion for another thing if it is necessarily good evidence for it. Unlike inductive evidence, criterial support is determined by convention and is partly constitutive of the meaning of the expression for whose application it is a criterion. Unlike entailment, criterial support is characteristically defeasible. Wittgenstein argued that behavioural expressions of the ‘inner’, e.g. groaning or crying out in pain, are neither inductive evidence for the mental (Cartesianism), nor do they entail the instantiation of the relevant mental term (behaviourism), but are defeasible criteria for its application. [Honderich 1995]
Key features of this definition are that the criteria of, for example, an ‘inner’ state like pain are fixed by convention and are partly constitutive of what we mean by pain. Thus groaning and crying out are not mere symptoms but rather part of what we understand by pain. They are connected by definition not induction.
This account of the relation between relevant behaviour and an ‘inner’ state can be motivated as a response to the problem of other minds prompted by a Cartesian picture. If the relation between mental states and behaviour had to be established through inductive correlations then the only available correlation would be the one that governed one’s own case. This is the basis of the ‘argument from analogy’. But that inference seems not to licence claims about other people.
If I say of myself that it is only from my own case that I know what the word “pain” means - must I not say the same of other people too? And how can I generalise the one case so irresponsibly? [Wittgenstein 1953 §293]
In addition to this challenge, Norman Malcolm presses a further line of thought found in Wittgenstein. If one starts simply from a first person perspective, it is unclear what sense one can give to the thought that other people have pain.
If I do not know how to establish that “someone has a pain” then I do not know how to establish that he has the same as I have when I have a pain. You cannot improve my understanding of “He has a pain” by this recourse to the notion of “the same”, unless you give me a criterion for saying that someone has the same as I have. If you do this you will have no use for the argument from analogy; and if you cannot then you do not understand the supposed conclusion of that argument. [Malcolm 1958: 970]
Thus, unless there were a criterion to give sense to the judgement ‘he is in pain’ the argument from analogy could not be applied. But, if there is a behavioural criterion, that argument is unnecessary.
This is the general shape of the response to a Cartesian form of scepticism about other minds popular amongst Wittgensteinians in the 1960’s and 1970’s. Instead of being mere symptoms of pain, relevant behaviour (eg groaning and crying out) is linked to underlying pain through an a priori conceptual connection that predates empirical inquiry. At the same time, however, the criteria of pain are taken to be defeasible.
The reason for this qualification is the following intuition. Whilst, in general, pain behaviour is the expression of underlying pain on occasion behaviour which resembles pain behaviour in every detail is not the expression of pain. It may be the result of acting or pretence. (And equally, genuine underlying pain may sometimes be stoically kept from expression.) As a result, the criterial support that apparent pain behaviour gives for a judgement that someone is in pain is taken to be defeasible. That is, it can, on occasion, be overturned.
The idea that criteria give only defeasible support for a claim is combined with a further assumption which McDowell describes thus: ‘if a condition is ever a criterion for a claim, then any condition of that type constitutes a criterion for that claim, or one suitably related to it’ [McDowell 1982: 462-3; 1998a: 377]. In other words, criteria are types. Thus whilst on most occasions, when instances of some general type of criterion are satisfied the underlying fact for which those instances are criteria also obtains, on some occasions the type of criterion is satisfied (by some particular circumstances) but the fact does not obtain. In the latter cases, the criterion is satisfied but is nevertheless also defeated.
If the contrast is to be taken to be with the connection between symptoms and an underlying mental state, criterial support looks at first sight to be a very close connection, almost as close as direct access to the underlying state. McDowell’s criticism, however, is that it is not close enough to serve the role for which it is intended. As an exegetical device, McDowell suggests contrasting the idea of criterial support with an alternative: ‘the circumstance that someone else is in some “inner” state can itself be an object of one’s experience... [and that] one can literally perceive, in another person’s facial expression or his behaviour, that he is... in pain’ [McDowell 1982: 456; 1998a: 370]. In fact, the position which McDowell advocates is not quite as direct as this.