Bayesian inference informed perspectives

Bayesian inference models (Friston, Reference Friston2008; Rao & Ballard, Reference Rao and Ballard1999) of brain function assume that the brain actively generates probabilistic predictions (i.e. prior beliefs) of its sensory inputs to infer their (probable) causes. Mismatches between these predictions and actual inputs (prediction errors) are used as ‘learning signals’ to gradually refine the internal model of the world, thereby improving subsequent predictions and reducing prediction errors. A central assumption is that the higher the relative reliability (termed ‘precision’) assigned to the sensory inputs compared to the prior belief, the more the prior belief is updated. The internal model of the world is assumed to be organized as a cortical belief hierarchy, with the bottom levels representing beliefs related to the more concrete features of the environment and increasingly abstract beliefs represented at higher hierarchical levels. Within this hierarchy, top-down prior beliefs are fed forward and integrated with bottom-up prediction errors signaled from the next lower cortical level (Friston, Reference Friston2008).

Delusion formation. Drawing on this model of brain function, delusions are proposed to arise from a deviation from optimal Bayesian belief updating (Hemsley & Garety, Reference Hemsley and Garety1986). In more recent predictive coding models (Adams et al., Reference Adams, Stephan, Brown, Frith and Friston2013; Harding et al., Reference Harding, Wolpe, Brugger, Navarro, Teufel and Fletcher2024; Sterzer et al., Reference Sterzer, Adams, Fletcher, Frith, Lawrie, Muckli and Corlett2018), this deviation has been characterized as aberrant precision weighting, that is, assigning too little precision to the prior belief or too much precision to the sensory input, or a combination of both. This is proposed to lead to aberrant prediction error signaling (Corlett, Honey, & Fletcher, Reference Corlett, Honey and Fletcher2016; Fineberg & Corlett, Reference Fineberg and Corlett2016; Fletcher & Frith, Reference Fletcher and Frith2009) that manifests as feelings of surprise and uncertainty that are resolved by adopting a higher-order delusional belief (Adams et al., Reference Adams, Stephan, Brown, Frith and Friston2013; Feeney, Groman, Taylor, & Corlett, Reference Feeney, Groman, Taylor and Corlett2017; Griffin & Fletcher, Reference Griffin and Fletcher2017; Miyazono & McKay, Reference Miyazono and McKay2019; Pfuhl, Reference Pfuhl2017; Sterzer et al., Reference Sterzer, Adams, Fletcher, Frith, Lawrie, Muckli and Corlett2018).

Delusion maintenance. Several ideas have been put forward to reconcile the paradox of fixed and stable delusional beliefs with the notion of a decreased prior belief precision. One is to differentiate mechanisms for the ‘formation phase’, characterized by an uncertain model about states in the world, from a ‘consolidation phase’, characterized by a strong impact of the adopted ‘extraordinary higher-order belief’ (Diaconescu, Hauke, & Borgwardt, Reference Diaconescu, Hauke and Borgwardt2019). Another is to assume a disbalance in the precision weighting of prior beliefs at different positions in the hierarchy. The basic idea here is a reduced weighting of prior beliefs at low hierarchical levels that is compensated by increased weighting of prior beliefs at high hierarchical levels (Sterzer et al., Reference Sterzer, Adams, Fletcher, Frith, Lawrie, Muckli and Corlett2018). These higher-order beliefs can include delusional beliefs which then ‘sculpt perceptual processing into conformity with delusions and foster their resistance to contradictory evidence’ (Petrovic & Sterzer, Reference Petrovic and Sterzer2023, p. 1425). Relatedly, Ashinoff, Singletary, Baker, and Horga (Reference Ashinoff, Singletary, Baker and Horga2022) propose overweighted high-level priors to induce a primacy bias in belief updating, such that early interpretations dominate subsequent evidence and beliefs become increasingly resistant to change over time. In a similar vein, Fineberg and Corlett (Reference Fineberg and Corlett2016) propose delusions to protect the belief system from further destabilization under persistent aberrant prediction error signaling by preventing escalating uncertainty.

Associative learning informed perspectives

The focus of associative learning has been on how animals, including humans, adapt to changes in the environment. A fundamental assumption is that they learn to anticipate and respond to significant events by associating them with cues that reliably predict their occurrence. Once such an association is formed, the presence of the cue activates the mental representation of the expected outcome, which initiates an appropriate response (Pearce & Bouton, Reference Pearce and Bouton2001). Of particular relevance here are the following three concepts: (1) classical conditioning – after pairing a neutral stimulus with an unconditioned stimulus (US) for a number of trials, the neutral stimulus comes to elicit a conditioned response in anticipation of the US, thereby becoming a conditioned stimulus (CS); (2) extinction learning – if the CS is presented repeatedly in absence of the US, the conditioned response gradually diminishes; (3) operant conditioning – behavior can be modified through its association with the consequences, such as addition or removal of rewards or aversive stimuli.

Delusion formation. Associative learning informed models propose that individual neurocognitive vulnerability (e.g., perceptual and semantic processing Rossell, Batty, & Hughes, Reference Rossell, Batty and Hughes2010; Vinogradov, King, & Huberman, Reference Vinogradov, King and Huberman1992 or attentional capacities, Roy, Reference Roy2017), or aberrant neurotransmission (e.g., dopaminergic, Corlett et al., Reference Corlett, Murray, Honey, Aitken, Shanks, Robbins and Fletcher2007) increase the likelihood of forming unusual associations through classical conditioning. They also hypothesize that perceptual processing difficulties (e.g., sensory gating difficulties, Vinogradov, King, & Huberman, Reference Vinogradov, King and Huberman1992), deficits in conditioned inattention (Roy, Reference Roy2017), or inappropriate mismatch signals (Corlett et al., Reference Corlett, Murray, Honey, Aitken, Shanks, Robbins and Fletcher2007) cause a person to pay attention to irrelevant stimuli, which also facilitates the formation of unusual associations.

Delusion maintenance. The hypotheses on delusion maintenance vary. Some have proposed that initially incidental associations are reinforced through operant conditioning, as feelings of relief or clarity strengthen these beliefs until they become increasingly rigid and shape the way further information is processed (Rossell, Batty, & Hughes, Reference Rossell, Batty and Hughes2010; Vinogradov, King, & Huberman, Reference Vinogradov, King and Huberman1992). Others highlight the amplifying role of fear-driven avoidance behaviors that prevent corrective experiences as the cues that were misinterpreted to signal threat continue to serve as a warning signal (Moutoussis, Williams, Dayan, & Bentall, Reference Moutoussis, Williams, Dayan and Bentall2007). Corlett, Krystal, Taylor, and Fletcher (Reference Corlett, Krystal, Taylor and Fletcher2009) proposed that deficits in extinction learning may underlie delusion maintenance. They drew on the internal reinforcement hypothesis by Eisenhard and Menzel (Reference Eisenhardt and Menzel2007) according to which the presentation of a CS reminds the organism of the US. This reminder is thought to trigger reconsolidation of the CS–US association, a process in which recalled memories become labile, integrate new information, and are then restabilized. This process is assumed to compete with extinction learning, in which a new CS–no US association is formed and consolidated. Corlett, Krystal, Taylor, and Fletcher (Reference Corlett, Krystal, Taylor and Fletcher2009) argued that delusions persist due to inappropriate activity in the midbrain ‘reminder system’ in which aberrant prediction errors might re-evoke the representation of the delusion without disconfirming it. This imbalance would favor reconsolidation over extinction learning, thereby strengthening the delusion rather than extinguishing it.

Neurobiological perspectives

Neurobiological perspectives emphasize the role of neuronal alterations in frontotemporal brain regions stemming from genetic predispositions, neurodevelopmental hazards, psychosocial adversity, and substance use. These alterations are proposed to disrupt neurodevelopment affecting neurotransmission systems in specific brain regions (the emphasis being on dopamine, but attention is also given to glutamate, norepinephrine, and serotonin), prefrontal cortex monitoring mechanisms, and balance of excitatory and inhibitory mechanisms that then impact perception, learning, and memory. Thus, many neurobiological models are tightly connected with Bayesian and associative learning accounts.

Delusion formation. A dopamine dysregulation is assumed to lead to the attribution of aberrant salience to usually irrelevant stimuli (Howes & Kapur, Reference Howes and Kapur2009; Howes & Murray, Reference Howes and Murray2014; Kapur, Reference Kapur2003; Spitzer, Reference Spitzer1995), or to an inappropriate attribution of certainty to sensory information (Broyd, Balzan, Woodward, & Allen, Reference Broyd, Balzan, Woodward and Allen2017). The experience of aberrant salience is thought to then lead to the formation of a delusional belief. Some models posit that the alterations of frontotemporal brain regions contribute to the altered evaluation of aberrant salience (Poletti & Sambataro, Reference Poletti and Sambataro2013) or other altered perceptions (Carr, Reference Carr2010). A Bayesian-informed neurobiological model (Denève & Jardri, Reference Denève and Jardri2016; Jardri & Denève, Reference Jardri and Denève2013) posits that an imbalance of excitatory and inhibitory (E/I) mechanisms in a hierarchical neural network causes a pathological form of causal interference in which bottom-up and top-down predictions are reverberated. This is thought to result in the predictions gaining precision – which manifests in erroneous perceptions and beliefs. Recently, it has been proposed that the E/I imbalance leads to abnormal neuronal representations (i.e. shallow attractor dynamics and disrupted generative replay) which manifest as maladaptive beliefs (Nour et al., Reference Nour, Liu, El-Gaby, McCutcheon and Dolan2024). Neurobiologically informed two-factor accounts (Coltheart, Reference Coltheart2007; Coltheart, Menzies, & Sutton, Reference Coltheart, Menzies and Sutton2010; Young, Reference Young2011) have been proposed for specific types of delusions (e.g., the Capgras delusion, i.e. the belief that a significant other is replaced by an imposter). They are assumed to be driven by a specific neuropsychological impairment (first factor, e.g., a disconnected face recognition system from the autonomic nervous system), triggering a delusional interpretation to explain the (constant) violation(s) of expectation this produces (second factor).

Delusion maintenance. Aberrant salience theories do not offer a unique neurobiological account of belief persistence, though some incorporate perspectives from other models (e.g., Bayesian models with the notion that the delusion then serves as a guiding cognitive scheme (or prior) for further thoughts and actions, Jardri & Denève, Reference Jardri and Denève2013; Kapur, Reference Kapur2003; see also extended two-factor account by Young, Reference Young2011). The two-factor model by Coltheart et al. posits that damage to lateral regions of right frontal cortex involves ‘a failure of the system … to consider new evidence … to revise the current belief’ (Reference Coltheart, Menzies and Sutton2010, p. 282). Spitzer (Reference Spitzer1995) hypothesized that the longer delusions persist, the more they become incorporated in the cortical maps encoding representations of events. Building on research showing anxiety to favor habitual response patterns and impair neuroplasticity, he argues that even if the neuromodulatory dysfunctions that led to the delusion (via aberrant salience) have ceased to exist, ‘the “normal” anxiety that accompanies any major changes in one’s world view may be enough to hinder the re-organization of cortical representations’ (p. 101).

Vulnerability–stress informed cognitive–behavioral perspectives

The core assumption of vulnerability–stress models is that symptoms result from interaction between predispositional vulnerability and external stress. The vulnerability is explained on different levels, including genetic, biological, or psychological. Here, we focus on models that highlight either cognitive vulnerability in form of biased information processing or emotional vulnerability in form of maladaptive schemas and emotions, or both.

Delusion formation. A core idea of these models is that delusional beliefs result from attempts to explain anomalous, aversive external or internal states triggered by stressors. The proposed stressors involve adverse childhood experiences and other adverse life events, daily hassles, or sleep dysfunction (e.g., Freeman & Garety, Reference Freeman and Garety2014; Garety et al., Reference Garety, Kuipers, Fowler, Freeman and Bebbington2001; Kesting & Lincoln, Reference Kesting and Lincoln2013; Morrison, Reference Morrison2001; Preti & Cella, Reference Preti and Cella2010). This search for an explanation is assumed to be influenced by a cognitive vulnerability in form of ‘biased’ information processing, such as selective attention to threat (e.g., Beck, Rector, Stolar, & Grant, Reference Beck, Rector, Stolar and Grant2009; Bentall, Reference Bentall, David and Cutting1994; Blackwood, Howard, Bentall, & Murray, Reference Blackwood, Howard, Bentall and Murray2001; Salvatore et al., Reference Salvatore, Lysaker, Popolo, Procacci, Carcione and Dimaggio2012) or a tendency to jump-to-conclusions (e.g., Garety et al., Reference Garety, Kuipers, Fowler, Freeman and Bebbington2001; Reference Garety, Bebbington, Fowler, Freeman and Kuipers2007; see Blackwood, Howard, Bentall, & Murray, Reference Blackwood, Howard, Bentall and Murray2001; Braun & Suffren, Reference Braun and Suffren2011 for neuropsychologically informed accounts of information processing deficits), or related biases leading to liberal acceptance of potentially false assumptions (Bronstein et al., Reference Bronstein, Pennycook, Joormann, Corlett and Cannon2019; Harrison, Shou, & Christensen, Reference Harrison, Shou and Christensen2021; Moritz et al., Reference Moritz, Pfuhl, Lüdtke, Menon, Balzan and Andreou2017). Along this vein, some models build on dual-process theories (Kahneman, Reference Kahneman2012) and suggest that deficits in the mobilization of neurocognitive resources that support analytic reasoning render the selection of implausible explanations more likely (Bronstein et al., Reference Bronstein, Pennycook, Joormann, Corlett and Cannon2019; Speechley & Ngan, Reference Speechley and Ngan2008; Ward & Garety, Reference Ward and Garety2019). Similarly, the erotetic theory of delusions postulates that people with delusions fail to endogenously generate questions (Parrott & Koralus, Reference Parrott and Koralus2015).

Some of these models also consider an independent emotional vulnerability in the form of negative schemas about the self and others, along with their associated emotions, proposing that such beliefs and emotions shape the appraisal of anomalous or ambiguous experiences (e.g., Freeman et al., Reference Freeman, Garety, Kuipers, Fowler and Bebbington2002; Freeman & Garety, Reference Freeman and Garety2014; Jorovat, Twumasi, Mechelli, & Georgiades, Reference Jorovat, Twumasi, Mechelli and Georgiades2025; Kesting & Lincoln, Reference Kesting and Lincoln2013; Preti & Cella, Reference Preti and Cella2010; van der Gaag, Reference van der Gaag2006).

Delusion maintenance. Models that focus on cognitive vulnerability generally assume that the same type of reasoning relevant to delusion formation also accounts for delusion maintenance. Some hypothesize additional biases that come into play once a hypothesized explanation is judged as valid: the ‘bias-against-disconfirmatory evidence’ (BADE), where people tend to discard disconfirming evidence altogether, or the ‘confirmation bias’ that describes selectively looking for evidence supporting one’s hypothesis (Moritz et al., Reference Moritz, Pfuhl, Lüdtke, Menon, Balzan and Andreou2017).

Models that also emphasize emotional vulnerability point to the effect that either the delusion or its consequences have on emotions and distressing beliefs. For instance, it has repeatedly been suggested that coming up with an explanation, independent of its content, provides initial relief (‘insight relief’) or satisfaction and thereby reinforces the delusional belief (Freeman et al., Reference Freeman, Garety, Kuipers, Fowler and Bebbington2002; Roberts, Reference Roberts1991). Some suggest that the content of the delusional belief (e.g., paranoid or grandiose) is negatively reinforced by attenuation of self-doubt or emotional pain associated with alternative beliefs that pose a threat to the persons’ already low self-esteem (Kesting & Lincoln, Reference Kesting and Lincoln2013). Others propose that avoidance and safety behaviors prevent corrective experiences and thereby contribute to belief maintenance (Beck, Rector, Stolar, & Grant, Reference Beck, Rector, Stolar and Grant2009; Freeman, Reference Freeman2016; Freeman et al., Reference Freeman, Garety, Kuipers, Fowler and Bebbington2002; Morrison, Reference Morrison2001; Newman-Taylor et al., Reference Newman-Taylor, Richardson, Sood, Sopp, Perry and Bolderston2020).

Finally, some models suggest additional cognitive processes relevant to maintenance: These include the role of maladaptive metacognitive processes, such as positive and negative beliefs about one’s beliefs (e.g., their controllability, Morrison, Reference Morrison2001; van der Gaag, Reference van der Gaag2006), excessive worry or rumination (Freeman, Reference Freeman2016; Isham et al., Reference Isham, Griffith, Boylan, Hicks, Wilson, Byrne and Freeman2021), the reactivation of negative schemas about the self and others that sustain vulnerability beliefs and negative affect (Jorovat, Twumasi, Mechelli, & Georgiades, Reference Jorovat, Twumasi, Mechelli and Georgiades2025; Kesting & Lincoln, Reference Kesting and Lincoln2013), and the relative absence of competing beliefs about safety that could constrain or override the threat belief (Freeman, Isham, & Waite, Reference Freeman, Isham and Waite2025).

Motivational perspectives

This category summarizes perspectives that emphasize a motive for developing or maintaining a delusion or emphasize a functional aspect. In many cases this is proposed as an additional factor alongside other explanations (e.g., McKay, Langdon, & Coltheart, Reference McKay, Langdon and Coltheart2007; Rigoli, Martinelli, & Pezzulo, Reference Rigoli, Martinelli and Pezzulo2021; Westermann, Gantenbein, Caspar, & Cavelti, Reference Westermann, Gantenbein, Caspar and Cavelti2018). These perspectives build on psychodynamic theories (e.g., Garrett, Reference Garrett2025), theories on the structure of motives (e.g., plan analysis, Caspar, Reference Caspar and Eells2011), philosophy (Ritunnano & Bortolotti, Reference Ritunnano and Bortolotti2022), or come from a utility-oriented perspective and tend to view both delusion formation and maintenance as a motivated process.

Delusion formation. A core assumption is that delusions arise as a way of dealing with an aversive state that leaves the person feeling vulnerable, such as a threat of loss of control (Melges & Freeman, Reference Melges and Freeman1975), or a threat to self-esteem (Bentall et al., Reference Bentall, Corcoran, Howard, Blackwood and Kinderman2001), or anxiety arising from ambiguity and uncertainty (Garrett, Reference Garrett2025; Houran & Lange, Reference Houran and Lange2004). All theories converge on the idea that the delusional experience serves to reduce the aversive state. For instance, preserving self-esteem by attributing the causes of current state to external factors (Bentall et al., Reference Bentall, Corcoran, Howard, Blackwood and Kinderman2001).

Delusion maintenance. The same assumption is made for delusion maintenance. The core postulate here is that maintaining delusional attributions serves to satisfy psychological needs, such as (re)gaining control (Melges & Freeman, Reference Melges and Freeman1975), or achieving a sense of orientation, coherence, directedness, and belonging (Ritunnano & Bortolotti, Reference Ritunnano and Bortolotti2022; Westermann, Gantenbein, Caspar, & Cavelti, Reference Westermann, Gantenbein, Caspar and Cavelti2018), or reducing negative affect by externalizing overwhelming affects into coherent narratives (Garrett, Reference Garrett2025). It is assumed, for instance, that to maintain a sense of rightness and control, individuals will be motivated to actively seek evidence supporting their delusion (Melges & Freeman, Reference Melges and Freeman1975). Some models also describe self-maintaining cycles. For example, persecutory delusions have been proposed to be maintained by the motivation to resist others influence, which increases the chance of actual rejection from others (Melges & Freeman, Reference Melges and Freeman1975). In addition, delusional beliefs have been postulated to become integrated into personal identity, further strengthening the motivation to maintain them (Isham et al., Reference Isham, Griffith, Boylan, Hicks, Wilson, Byrne and Freeman2021; Melges & Freeman, Reference Melges and Freeman1975). Rigoli, Martinelli, and Pezzulo (Reference Rigoli, Martinelli and Pezzulo2021) postulate that some types of delusional beliefs are maintained due to cost–benefit calculations (e.g., if the cost of incorrect rejection of a persecutory belief is deemed higher than the cost of incorrectly maintaining it).

Social psychological perspectives

Model perspectives from social psychology broadly build on self-categorization (Turner et al., Reference Turner, Hogg, Oakes, Reicher and Wetherell1987) and social identity theories (Tajfel & Turner, Reference Tajfel and Turner1979).

Delusion formation. These models posit that individuals’ memberships in social groups are integrated into their self-concepts and that social exclusion or interpersonal adversity threaten a person’s self-concept and perceived control, creating a basis for interpersonal mistrust that – in interaction with deficits in social perception and social cognition – increase the likelihood of misinterpretations and give rise to paranoid beliefs (Fuchs, Reference Fuchs1993; Green & Phillips, Reference Green and Phillips2004; Greenaway, Haslam, & Bingley, Reference Greenaway, Haslam and Bingley2019; Houseman, Reference Houseman1990; Raihani & Bell, Reference Raihani and Bell2019).

Delusion maintenance. Continued lack of corrective feedback by interaction partners and threat avoidance is hypothesized to maintain and exacerbate the beliefs. Along a similar vein, it has also been suggested that paranoid beliefs are strengthened by mutual misunderstanding between interaction partners (Hajdúk, Sasson, Park, & Pinkham, Reference Hajdúk, Sasson, Park and Pinkham2024).