Saturday 4 January 2014

Epistemic Innocence (part 4)

This is the fourth in a series of posts Lisa and I are writing on Epistemic Innocence. So far Lisa has introduced the two conditions we think characterize epistemic innocence, I have written about the Availability Condition, and Lisa has written about the Epistemic Benefit condition. In this post I want to outline some reasons for thinking that delusional beliefs, at least sometimes, meet these two conditions. In her next post, Lisa will apply the notion of epistemic innocence to memory.

The No Relevant Alternatives Condition

The claim that delusions meet the No Relevant Alternatives Condition is supported by Daniel Freeman and colleagues' study on alternative explanations. Subjects with delusions were asked about the evidence for their delusional beliefs and the ‘availability of alternative explanations’ (Freeman et al. 2004: 677). Freeman and colleagues conclude from the study that:

‘we have found evidence that (at least at the height of the episode) individuals do not have accessible alternative explanations for their experiences. This seems to be caused by (a) a lack of awareness and knowledge about internal anomalous experiences, (b) reasoning biases, and (c) the difficulty of conceiving alternatives that are as compelling as the delusions in explaining personal experiences’
(Freeman et al. 2004: 679, my emphasis).

Coltheart and colleagues' (2009) work on abductive inference and delusional belief also supports the claim that delusions meet the No Relevant Alternatives Condition. Coltheart and colleagues argue that subjects with delusions form beliefs in line with a Bayesian model of abductive inference, according to which ‘one hypothesis 
H1 explains observations O better than another hypothesis H2 just in case P(O│H1) > P(O│H2)’ (Coltheart et al. 2009: 276).

Coltheart and colleagues think that the delusional hypothesis is more probable than the non-delusional hypothesis, given the observed phenomenon. Considering a case of the Capgras delusion, the two hypotheses in play are the stranger hypothesis (the woman who looks like my wife is not my wife) and the wife hypothesis (the woman who looks like my wife is my wife):

‘the observed data are clearly much more likely under the stranger hypothesis than under the wife hypothesis. It would be highly improbable for the subject to have the low autonomic response if the person really was his wife, but very probable indeed if the person were a stranger’
(Coltheart et al. 2009: 277).

Coltheart and colleagues' work here suggests that non-delusional alternative hypotheses are unavailable to subjects insofar as they are explanatorily unavailable. The delusional hypothesis just does a much better job of explaining the observed data, if what we mean by doing a better job is cashed out in terms of Bayesian probabilities.

The Epistemic Benefit Condition

What kinds of epistemic benefits might delusions enjoy? One benefit might be the filling of an explanatory gap. The subject may have an anomalous experience which they cannot explain, this may create in the subject ‘puzzlement, anxiety, and a search for an explanation’ (Maher 2006: 181). Perhaps the delusional hypothesis which makes some sense of the experience is better than no hypothesis at all (given the No Relevant Alternatives condition, there is no alternative). Maher suggests that delusional hypotheses are best thought of as like scientific theories—both ‘serve the purpose of providing order and meaning for empirical data obtained by observation’ (Maher 1988: 20). If this is right, the delusion is epistemically beneficial in providing an explanation of an as yet unexplained phenomenon.

Delusions might also be epistemically beneficial insofar as they make other perceptual beliefs coherent. A subject’s belief that she sees a second head on her shoulder, or that the person in the mirror does not look like her, might be made sense of by a delusional belief.

Delusions may also have indirect epistemic benefits. Maher suggests that for subjects with delusions, coherent explanations of their experiences are often accompanied by a ‘strong feeling of personal relief’, as well as the excitement a scientist might get from intellectual insight (Maher 1974: 104). The explanation for the anomalous experience that the delusional subject comes to may provide ‘enough relief from anxiety that it becomes difficult for the individual to abandon it and return to the initial state of confusion and distress’ (
Maher 2006: 182). This relief from anxiety may have indirect epistemic benefits if the anxiety felt before coming to the delusional explanation for the anomalous experience was such as to adversely affect wellbeing and compromise the gaining of knowledge.


  1. Dear Ema,

    I've very much enjoyed this series of posts on Epistemic Innocence.

    I was hoping you could help clarify the notion of 'availability' you are using. Consider the Bayesian framework, if a particular hypothesis is less probable, does that make it 'unavailable'? Or is it 'unavailable' because the cognitive mechanism responsible for updating one's beliefs is unconscious - i.e., the hypothesis is not available to consciousness? Or is there some other notion of what makes an alternative hypothesis available?

    I think there is one way of understanding what is happening in the Bayesian framework as modeling how a rational agent compares rival empirical hypotheses. The one that better explains the data, should be adopted over the less probable rival, but it doesn't seem like that makes the less probable option 'unavailable'.

  2. Hi Matthew,

    Thanks for this, glad you enjoyed the series!

    So you ask: if a particular hypothesis is less probable, does that make it unavailable? Or is it unavailable because the cognitive mechanisms responsible for updating one’s belief is unconscious – i.e., the hypothesis is not available to consciousness?’ This is a really useful way of putting the question, because it roughly matches with what I have called 'explanatory unavailability' and 'strong unavailability' (I sketched a bit of this in Epistemic Innocence (part two)). I’m inclined to think that, if we’re thinking in terms of the Bayesian framework, that non-delusional hypotheses are explanatorily unavailable, which is to say that they strike the subject with a delusion as poor (or at least, less good) explanations of their experiences than the delusional hypothesis. (Though there might be stronger unavailability going on—something like the hypothesis not being available to consciousness, and that might be why 76% of Freeman and colleagues’ participants did not come up with an alternative hypothesis *at all*.)

    So you worry that one hypothesis better explaining the data than a rival doesn’t make that rival ‘unavailable’, so maybe that’s an unfortunate way of putting it (because I agree, ‘unavailable’ here seems a bit strong!). But what I mean by explanatory unavailability is cashed out in terms of the implausibility or poor explanatory power of alternative explanations (by the subject’s lights). A subject may come to have a cognition which explains some experience she has. If alternative cognitions which might also be candidate explanations for her experiences are such that they strike her as seriously implausible or explanatorily inadequate, I want to say that these alternative cognitions are explanatorily unavailable.

    Take a non-pathological case: suppose there are bite marks in my cheese, I hear scratching at night, and my cat is agitated. I come to the conclusion that I have mice in my house. An alternative explanation might be that cheese-eating, cat-hating, noisy fairies are infiltrating my home at night. This explanation is unavailable to me in the sense I have in mind here due to the incredulity I would feel towards it. It is either not considered by me, or it is such that I rule it out on grounds of implausibility or poor explanatory power, relative to the preferred and adopted cognition.

    So it is in that sense that I think rival non-delusional hypothesis might be ‘unavailable’ to the subject, though I agree, that might be a slightly misleading way of characterising what’s going on.

    Thanks again for this, it's really useful for us to think about what we mean by unavailability of alternatives!

  3. Hi Matthew

    Thank you for your comment.

    One way of understanding the notion of epistemic innocence is to say that it applies to those cognitions that are imperfect (false, irrational, factually inaccurate, groundless) but that have some significant epistemic benefit that could not be easily obtained otherwise. This is quite broad. Then, we need to unpack the reasons why that benefit may be unique to the cognition. It may be that a cognition that is less imperfect is not accessible to the agent at the time (the other cognition is unavailable) or that there is an available cognition that is less imperfect but it does not carry the same epistemic benefit as the imperfect cognition.

    That is why I prefer to call the second condition for epistemic innocence "no relevant alternatives". It leaves it open whether alternatives are strictly speaking unavailable or whether they just fail to have the relevant epistemic benefit.

    Hope it makes sense! Still very "work-in-progress".


Comments are moderated.