I have often addressed the distinction between atheism and agnosticism but I haven’t said a lot about what agnosticism involves, apart from it being a suspension of judgement about belief claims. So a few remarks are in order, prompted (but probably misreading) a recent paper by Jane Friedman, “Rational Agnosticism and Degrees of Belief“, forthcoming in Oxford Studies in Epistemology, Volume 4.
What is agnosticism? I do not mean agnosticism about just gods, but all kinds of agnosticism? Graham Oppy (1994) distinguished between two kinds:
strong agnosticism, i.e. the view which is sustained by the thesis that it is obligatory for reasonable persons to suspend judgement on the question of God’s existence. And, on the other hand, there is weak agnosticism, i.e. the view which is sustained by the thesis that it is permissible for reasonable persons to suspend judgement on the question of God’s existence.
I’m not sure about this. Oppy’s view is that one is either forced or permitted to suspend belief (in God’s existence, which we can replace here with any general claim p). I would rather say that weak agnosticism is the view that one finds it more reasonable than not to suspend belief in p. It isn’t an arbitrary choice, but a choice that is based on our best (fallible) reading of the evidence. But we can accept the definition that agnosticism is sustained by a thesis that reasonable persons should suspend judgement on p, whether or not that is a forced choice.
Friedman takes a much more agreeable approach. She considers the widely held view that agnostics weigh the evidence and find the “credence” (a degree-of-belief value given to beliefs p that either leads you to adopting p or adopting not-p) to be equivalent; that is, it neither supports nor rejects p. She calls this the Straightforward Reduction Thesis (SRT): that all judgements including agnosticism are reducible to claims about the credence of their beliefs p.
With a fair degree of technical discussion, Friedman argues that agnosticism cannot be reduced to credence claims, including in a vague manner or in the cases where one has some Bayesian priors. To illustrate this, consider how Dawkins discusses agnosticism in his The God Delusion:
There is nothing wrong with being agnostic in cases where we lack evidence one way or the other. It is the reasonable position. 
I’ll begin by distinguishing two kinds of agnosticism. TAP, or Temporary Agnosticism in Practice, is the legitimate fence-sitting where there really is a definite answer, one way or the other, but we so far lack the evidence to reach it (or don’t understand the evidence, or haven’t time to read the evidence, etc.). …
But there is also a deeply inescapable kind of fence-sitting, which I shall call PAP (Permanent Agnosticism in Principle). … The PAP style of agnosticism is appropriate for questions that can never be answered, no matter how much evidence we gather, because the very idea of evidence is not applicable. 
Notice that for Dawkins, agnosticism is the taking of a position regarding the degree of evidence available to us. PAP is Oppy’s strong agnosticism. It was held, for example, by the coiner of the term, Thomas Huxley, who held that the p in question (God’s existence) was forever unknowable. But is TAP the same as weak agnosticism? Are agnostics merely temporarily suspending judgement until degrees of belief ramp up and decide the matter?
Friedman’s argument is that they are not. Instead she rejects what she calls the “middling” assumption: that one suspends disbelief and belief when the credence is within some middling interval (say, between ⅓ and ⅔). The argument is technical but concludes that a Straightforward Reductionist should conclude the middling interval is between, but not when equal to, 0 and 1. In short, we should be on the SRT agnostics about anything that is neither known with certainty nor known to be false with certainty (which is, let’s face it, pretty well every belief).
This view, by the way, was discussed and not very clearly dropped by Dawkins. I say “not very clearly” because it looks for all the world that he wanted his readers to think the middling assumption was in fact the right way to think of agnosticism – the evidence hovers at 50% – even though he disavows it.
Let us, then, take the idea of a spectrum of probabilities seriously, and place human judgements about the existence of God along it, between two extremes of opposite certainty. The spectrum is continuous, but it can be represented by the following seven milestones along the way.
1 Strong theist. 100 per cent probability of God. In the words of C. G. Jung, ‘I do not believe, I know.’
2 Very high probability but short of 100 per cent. De facto theist. ‘I cannot know for certain, but I strongly believe in God and live my life on the assumption that he is there.’
3 Higher than 50 per cent but not very high. Technically agnostic but leaning towards theism. ‘I am very uncertain, but I am inclined to believe in God.’
4 Exactly 50 per cent. Completely impartial agnostic. ‘God’s existence and non-existence are exactly equiprobable.’
5 Lower than 50 per cent but not very low. Technically agnostic but leaning towards atheism. ‘I don’t know whether God exists but I’m inclined to be sceptical.’
6 Very low probability, but short of zero. De facto atheist. ‘I cannot know for certain but I think God is very improbable, and I live my life on the assumption that he is notthere.’
7 Strong atheist. ‘I know there is no God, with the same conviction as Jung “knows” there is one.’
The spectrum of probabilities works well for TAP (temporary agnosticism in practice). It is superficially tempting to place PAP (permanent agnosticism in principle) in the middle of the spectrum, with a 50 per cent probability of God’s existence, but this is not correct. PAP agnostics aver that we cannot say anything, one way or the other, on the question of whether or not God exists. The question, for PAP agnostics, is in principle unanswerable, and they should strictly refuse to place themselves anywhere on the spectrum of probabilities. The fact that I cannot know whether your red is the same as my green doesn’t make the probability 50 per cent. The proposition on offer is too meaningless to be dignified with a probability. Nevertheless, it is a common error, which we shall meet again, to leap from the premise that the question of God’s existence is in principle unanswerable to the conclusion that his existence and his non-existence are equiprobable.
In fact the middling assumption is not correct for his TAPs either. Dawkins assumes that there must be a credence for God’s existence, and that only in cases when the belief simply cannot have credence, that is, in his positivistic framework, when p is incapable of having meaning, can one be a PAP about it. But this cannot be the case, because it is simply not meaningless that you see green where I see red; at best it is false, and at worse it is something we cannot know for sure, but it isn’t meaningless (there is a fact of the matter). This kind of positivist view about meaning and verifiability was abandoned by philosophy some time ago for all kinds of reasons. I invite you to read up on the topic yourself.
Provisional agnostics hold that a question is not – yet – capable of being answered, not that the evidence is equivocal or evenly balanced. Friedman’s discussion is therefore of interest. If we are not, in being agnostics, attempting to find a middling credence, what is it we are doing?
We should at the outset note that there are cases like Dawkins’ nonsensical questions. For example, it makes no sense to ask whether green is red. It might make sense to ask if your experience of red is identical to my experience of green, but that is a question about experiences. Red is not green, and to ask if it can be known if red is green is nonsense. However, in the case of nonsense questions, one is not agnostic; one is simply bemused.
But if there simply are no credences, which is to say we cannot reduce the question that p to evidence for or against p, we should suspend any attempt to consider either p or not-p worth belief. In fact, as Rosenkranz (2007) notes, agnosticism is a third stance: there is belief that p is true, belief that p is false, and the opposition to both p and not-p. Rosenkranz analyses this in terms of realism and antirealism about p and out cognitive ability to discern which is correct. Agnosticism, as I have previously argued on this blog, is neither an existence claim that p (realism) nor the denial of an existence claim that p (antirealism), but the denial of the possibility of knowledge regarding p, either now (weak) or ever (strong).
What reasons for thinking that we do not know p might there be? I can think of a few possibilities; add more in the comments if you can think of any.
1. There is no evidence for p, nor evidence for not-p.
2. There are reasons for thinking there can not be evidence that bears on p.
3. There are reasons for thinking that, while there may be evidence bearing on p, it is inaccessible to us (now or always); for example, we may never have investigated p claims. Strong agnosticism here might involve contingent or necessary claims that we can not get evidence bearing on p claims. Contingent reasons might include economic claims (there may be a lake of helium-3 on Pluto, but we’ll never be able to afford a probe to check), technical reasons (we can’t detect the Higgs boson because it would take a hadron collider bigger than one we could build), or conceptual (we can’t predict the fine grain behaviour of a cell because the mathematical simulations are not developed yet). Necessary reasons might include logical or metaphysical necessities.
4. We do not care about p. Someone might have no opinion or stance towards a claim because it is simply outside their sphere of interest. In such a case a reasonable person might decide to have no doxastic stance whatsoever. I feel that way, for example, about the doctrine of double predestination in Calvinist theology. You could say I am agnostic about any of the options simply because I feel no motivation to take an opinion (this could be understood as a denial of the Calvinist theological presumptions that do motivate the debate, but that is different to denying double predestination). Of course, once the question p? is put, if it is a question that I take to be serious, then I will find a solution or be agnostic for one of the other reasons. This apathetic agnosticism is, I put it, the kind of agnosticism we have about the majority of possible beliefs.
[Sidenote: this goes to the question of dialectic logic, sometimes called erotetic logic. We are not forced to take a stance on all possible questions, but only those that are put to us by our community and its traditions. If there are creatures with religions on Alderbaran V, I am not required to take a stance to their beliefs if they are not identical to any of those in my community.]
So these are the conditions under which a reasonable person suspends both belief and disbelief. One is agnostic when credence cannot be assigned, not even vaguely or in a Bayesian fashion. How does agnosticism relate to skepticism?
A skeptic assigns belief only when there is warrant for that belief’s content. In any other case, the skeptic will reject that belief. If one is skeptical of p claims, a failure to assign a credence of 1 means one assigns a credence of 0 to p. In ordinary terms, if you have no positive reason to accept a claim, you reject it. This underlies some of the rhetoric regarding atheism: arguments that God’s existence is a hypothesis, and that the hypothesis is unsupported and so one should not believe it and deny that it is reasonable to believe it, is skeptical, but not agnostic. Of course a skeptic on some matters can be agnostic on others, but to achieve this one needs to have reason to treat some claims differently from others. This is not something one has by intuition, or else it ends up being special pleading for those beliefs we most strongly feel about.
As Rosenkrantz notes, a reasonable* agnostic does not stop at the decision not to make a judgement, but, if there are avenues to investigate, will continue to try to make one. We should be skeptical of claims that are extraordinary, that are either making out some great cognitive novelty (like moving from a geocentric to a heliocentric model of astronomy) or run contrary to too much of our existing beliefs.
* You’ll note I don’t say “rational”. This is a concept that has been wrung to death by people trying to advance all kinds of philosophical agendas. A reasonable person is somebody who takes reasons seriously and accepts the conclusions they license.
Oppy, Graham. 1994. Weak agnosticism defended. International Journal for Philosophy of Religion 36 (3):147 – 167.
Rosenkranz, Sven. 2007. Agnosticism as a Third Stance. Mind 116 (461):55-104.