Epistemic Benefits of Delusions (1)

This is the first in a series of two posts by Phil Corlett (pictured above) and Sarah Fineberg (pictured below). Phil and Sarah are both based in the Department of Psychiatry at Yale University. In this post and the next they discuss the adaptive value of delusional beliefs via their predictive coding model of the mind, and the potential delusions have for epistemic benefits (see their recent paper 'The Doxastic Shear Pin: Delusions as Errors of Learning and Memory', in Cognitive Neuropsychiatry). Phil presented a version of the arguments below at the Royal College of Psychiatrists' Annual Meeting in Birmingham in 2015, as part of a session on delusions sponsored by project PERFECT.

The predictive coding model of mind and brain function and dysfunction seems to be committed to veracity; at its heart is an error correcting, plastic, learning mechanism that ought to maximize future rewards, and minimize future punishments like the agents of traditional microeconomics—so called econs (Padoa-Schioppa forthcoming). This seems at odds with predictive coding models of psychopathology and in particular psychotic symptoms like hallucinations and delusions (Corlett et al. 2010). Put simply, if delusions result from a noisy maladaptive learning mechanism, why do individuals learn anything at all—let alone the complex and strongly held beliefs that characterize psychotic illness? We know from behavioural economists like Kahneman, Tversky, and, Thaler that humans can depart from econ-like responding. Can predictive coding depart likewise? And does it depart in interesting ways that are relevant to delusions?

We think so. Bayesian models of cognition and behaviour need not necessarily optimize expected value. For example, Bayesian models of message passing in crowds can recapitulate the rumors and panic that characterize communication after a salient world event (Butts 1998). With regards to delusions, we would like to re-consider Daniel Dennett and Ryan McKay’s assessment of adaptive mis-beliefs (McKay and Dennett 2009). McKay and Dennett explored the existence of misbeliefs—incorrect beliefs that, despite being wrong, nevertheless confer some advantage on the adherent. They argued that only positive illusions—beliefs that one is more competent, more attractive, less biased than in reality, etc.—were evolved, adaptive, misbeliefs (McKay and Dennett 2009). We (and others) think delusions might confer such a function (Hagen 2008).

McKay and Dennett introduced an interesting concept to which we return presently—the doxastic shear-pin (McKay and Dennett 2009). A shear-pin is a concept from engineering. Shear-pins are built into systems in order to disable a machine in trouble so that continued functioning does not destroy the whole machine (an electrical fuse functions similarly). A broken shear-pin allows a machine to continue to function, albeit at an attenuated level. We argue that delusions form when a doxastic shear-pin breaks. The shear-pin has much in common with psychological defenses and biases—ways of believing that perhaps to do not fully reflect reality but rather the reality that an individual desires (McKay and Dennett 2009).

This sounds maladaptive, but there are situations in which wrong beliefs (like overconfidence in ones abilities) can confer an adaptive advantage e.g. when the benefits of contested resources outweigh the costs of competition (Johnson and Fowler 2011). We argue that a doxastic shear pin breaking to produce delusions is not adaptive in this sense (although we note that people with attenuated odd beliefs may be more fecund (Nettle and Clegg 2006)), but rather it allows some continued instrumental engagement with the world (and exploitation of resources) rather than no action at all (Mishara 2009).

We explain the shear-pin in terms of aberrant reinforcement learning and memory mechanisms. These mechanisms are themselves normally adaptive, allowing organisms to exploit the reward contingencies in their environment and respond with flexibility when those contingencies change. Over time (and with repeated exposure) this normally plastic learning mechanism gives way to an inflexible belief, akin to a stimulus response habit that, through overtraining, becomes resistant to contradictory evidence. In the context of the learning model, delusions are adaptive misbeliefs because they provide a means through which patients can continue to engage with their environment (Mishara 2009).

These ideas—of the doxastic shear-pin—raise the notion of the epistemic innocence of delusions; whether some delusions are imperfect cognitions that can be wrong but may still confer some epistemic benefit (Bortolotti 2015). We argue that at least some cases of delusion are epistemically innocent—because they provide an explanation for ineffable, salient, and frightening experiences (Corlett et al. 2010). In our next post we will outline our argument for this claim.