Unconscious Inference in Delusion Formation

In this post, Federico Bongiorno (now a Postdoctoral Researcher at the University of Oxford, funded by an Analysis Trust and Mind Association award) is summarising a paper he wrote with Lisa Bortolotti while a PhD student at the University of Birmingham. The paper is entitled: "The Role of Unconscious Inference in Models of Delusion Formation" and appeared in Inference and Consciousness, a volume edited by Timothy Chan and Anders Nes and published by Routledge in 2019.

Federico Bongiorno

Brendan Maher was the first to suggest that the formation of delusions involves an inferential transition—although he denies that the inference from which delusions arise is faulty (Maher, 1992; Maher, 1999). Maher defends a view known as ‘explanationism’ (Maher, 1974; Stone and Young, 1997), according to which delusions are hypotheses adopted to explain anomalous perceptual experiences and arrived at by inferential reasoning that is neither abnormally biased nor otherwise deficient (Maher, 1974, p. 180). In essence, delusions for Maher are the product of normal reasoning processes brought to bear on some experiential aberration. 

This means that the pathological nature of the delusion does not lie in the person’s inferential reasoning, but only in the experience that generates it. It also means that no additional abnormality is needed to explain delusional belief formation and maintenance, which is why Maher’s view has become known as a one-factor theory. An alternative to explanationism is the endorsement theory, according to which the delusional belief is an acknowledgement that the anomalous experience is veridical and no inference from experience to belief is required (Pacherie et al., 2006; Bayne and Pacherie, 2004).

Delusional inference has been increasingly understood in terms of Bayesian updating. In our paper we critically evaluate an influential Bayesian model of delusional inference put forward by Coltheart, Menzies, and Sutton (2010), which we call, for simplicity, ‘the Coltheart model’. The Coltheart model has been developed with specific reference to the Capgras delusion, the belief that a person or persons dear to the deluded individual have been replaced by identical or nearly identical imposters (Capgras and Reboul-Lachaux, 1923). 

The proposal is that in Capgras the delusional hypothesis is adopted via unconscious abductive inference from abnormal data, where ‘data’ is used to refer to explananda that are not available to personal-level consciousness. In the Coltheart model the inference involved in the formation of the delusion is Bayesian rational and does not involve a reasoning impairment, though a reasoning impairment, a second factor, is postulated to explain the maintenance of the delusional belief in the face of counterevidence.

In the last decade, many theorists have pointed to the use of Bayesian framework for modelling delusional inference, with the debate revolving around the number of factors necessary for delusion formation, and the similarities and differences between the available models (e.g., Coltheart et al., 2010; McKay, 2012; Bortolotti and Miyazono, 2015; Miyazono et al., 2015; Miyazono and McKay, 2019). 

By contrast, we focus on the question of compatibility between, on the one hand, the Coltheart model, and on the other hand, explanationist and endorsement accounts of delusion formation. Strictly speaking, the Coltheart model is not a Maher-type explanationism, at least if one treats experience as by definition conscious. It does, however, conceptualise delusions as hypotheses serving an explanatory function. 

Because of that, the Coltheart model has been interpreted as a modern version of explanationism (Parrott, 2019; Young, 2014). We argue, however, that an explanatory function as understood here is no less compatible with an explanatory picture than an endorsement one. If that is correct, the mere presence of inference in delusion formation is not sufficient to discriminate between explanationist and endorsement accounts.