In their thought-provoking target article, Rahnev & Denison (R&D) argue that real progress in understanding perception could be achieved by observer models that account for optimal and suboptimal behaviour. We believe that such models could furthermore be very useful for characterising variations in perception across healthy participants and those affected by psychiatric disorders. Inter-individual variations in perception (e.g., Grzeczkowski et al. Reference Grzeczkowski, Clarke, Francis, Mast and Herzog2017; Partos et al. Reference Partos, Cropper and Rawlings2016; Schultz & Bülthoff Reference Schultz and Bülthoff2013; van Boxtel et al. Reference van Boxtel, Peng, Su and Lu2017) and perceptual decision making (Ratcliff et al. Reference Ratcliff, Thapar and McKoon2010; Reference Ratcliff, Thapar and McKoon2011; Schmiedek et al. Reference Schmiedek, Oberauer, Wilhelm, Süss and Wittmann2007) have been widely reported. An established approach to investigate these processes and their variations has been to model accuracy and response times using diffusion models (Ratcliff Reference Ratcliff1978; Ratcliff et al. Reference Ratcliff, Smith, Brown and McKoon2016). Comparing parameters of these models with personality traits across healthy participants, or between healthy participants and patients, provides insight into the origins of the variability. This has allowed researchers to relate individual differences in perceptual decision making to individual differences in IQ, working memory, and reading measures (Ratcliff et al. Reference Ratcliff, Thapar and McKoon2010; Reference Ratcliff, Thapar and McKoon2011; Schmiedek et al. Reference Schmiedek, Oberauer, Wilhelm, Süss and Wittmann2007) and to characterise deficits in participants with aphasia (Ratcliff et al. Reference Ratcliff, Perea, Colangelo and Buchanan2004), dyslexia (McKoon & Ratcliff Reference McKoon and Ratcliff2016), attention-deficit/hyperactivity disorder (Mulder et al. Reference Mulder, Bos, Weusten, van Belle, van Dijk, Simen, van Engeland and Durston2010), schizophrenia (Moustafa et al. Reference Moustafa, Kéri, Somlai, Balsdon, Frydecka, Misiak and White2015), depression, and anxiety (White et al. Reference White, Ratcliff, Vasey and McKoon2009).

As part of the Research Domain Criteria (RDoC) project (Insel et al. Reference Insel, Cuthbert, Garvey, Heinssen, Pine, Quinn, Sanislow and Wang2010) aiming to incorporate genetics, neuroimaging, and cognitive science into future psychiatric diagnostic schemes, applying neurobiologically plausible models of value-based decision making to characterise deficits observed in psychiatric disorders (Collins et al. Reference Collins, Albrecht, Waltz, Gold and Frank2017; Huys et al. Reference Huys, Daw and Dayan2015) has led to the development of computational psychiatry (Maia et al. Reference Maia, Huys and Frank2017; Wiecki et al. Reference Wiecki, Poland and Frank2014). This approach promises mechanistic explanations of how psychiatric symptoms such as cognitive biases may result from failures of decision variable evaluation (Huys et al. Reference Huys, Daw and Dayan2015). Bayesian models combining prior information with sensory evidence are particularly promising in yielding insight into pathophysiological mechanisms of perceptual distortions observed in schizophrenia. For example, information processing favouring prior knowledge over incoming sensory evidence can account for differences in visual illusion perception observed in early psychosis and schizotypy (Partos et al. Reference Partos, Cropper and Rawlings2016; Teufel et al. Reference Teufel, Subramaniam, Dobler, Perez, Finnemann, Mehta, Goodyer and Fletcher2015). The “jumping-to-conclusions” bias in event probability estimation typical of schizophrenia can be characterised by increased circular inference – that is, the corruption of sensory data by prior information, with feedforward and feedback loops of the model correlating with negative and positive symptoms, respectively (Jardri et al. Reference Jardri, Duverne, Litvinova and Deneve2017). In time, such approaches may allow us to develop specific therapeutic approaches, such as metacognitive training (e.g., see Moritz & Woodward Reference Moritz and Woodward2007).

Observer models may allow similar progress in understanding the mechanisms underlying dysfunctions of social perception and interaction. Parameterizable social stimuli may prove very helpful in this regard; for example, point-light motion stimuli and tasks assessing different levels of processing have allowed researchers to better understand how autistic traits affect certain aspects of biological motion perception (van Boxtel et al. Reference van Boxtel, Peng, Su and Lu2017). The response to others’ gaze is also affected in autism (Leekam et al. Reference Leekam, Hunnisett and Moore1998; Wallace et al. Reference Wallace, Coleman, Pascalis and Bailey2006); here, a recently developed computational model of the perception of gaze direction (Palmer & Clifford Reference Palmer and Clifford2017) has yielded insight into the origin of those dysfunctions: It has been proposed that autism is associated with reduced divisive normalisation of sensory responses, attributable to an increased ratio of cortical excitation to inhibition (Rosenberg et al. Reference Rosenberg, Patterson and Angelaki2015). Interestingly, both divisive normalisation and sensory adaptation occur robustly in autism in the context of gaze processing (Palmer et al. Reference Palmer, Lawson, Shankar, Clifford and Rees2018). This suggests that the differences in response to others’ gaze may instead be related to differences in the interpretation of gaze direction or the spontaneous following of others’ gaze (Senju et al. Reference Senju, Southgate, White and Frith2009). Similar work could be undertaken for elucidating other essential social cognitive functions, such as face recognition. Face recognition capacities widely vary across healthy participants (Wilmer et al. Reference Wilmer, Germine, Chabris, Chatterjee, Gerbasi and Nakayama2012), ranging from congenital prosopagnosia (Behrmann & Avidan Reference Behrmann and Avidan2005; McConachie Reference McConachie1976) to “super-recognition” (Russell et al. Reference Russell, Duchaine and Nakayama2009). Although progress towards understanding the cognitive and neural underpinnings of congenital prosopagnosia is being made (Susilo & Duchaine Reference Susilo and Duchaine2013), the most widely used tests may not capture the alternative perceptual strategies adopted by people afflicted by prosopagnosia (Esins et al. Reference Esins, Schultz, Stemper, Kennerknecht and Bülthoff2016). Parameterizable face stimuli (Dobs et al. Reference Dobs, Bülthoff, Breidt, Vuong, Curio and Schultz2014; Esins et al. Reference Esins, Schultz, Wallraven and Bülthoff2014; Reference Esins, Schultz, Stemper, Kennerknecht and Bülthoff2016) may allow us to better characterise those strategies by allowing direct comparisons between human and ideal observer performance (Dobs et al. Reference Dobs, Bülthoff and Schultz2016; Reference Dobs, Ma and Reddy2017). Such approaches may be instrumental in identifying alternative heuristics used by participants with congenital prosopagnosia and other impairments of social perception.

Recent studies have demonstrated that exogenous administration of the neuropeptide oxytocin (OT) influences the perception of social stimuli such as facial emotions in a dose-dependent manner (Spengler et al. Reference Spengler, Schultz, Scheele, Essel, Maier, Heinrichs and Hurlemann2017b). Furthermore, OT modulates attractiveness judgements of faces (Hurlemann et al. Reference Hurlemann, Scheele, Maier and Schultz2017; Striepens et al. Reference Striepens, Matusch, Kendrick, Mihov, Elmenhorst, Becker, Lang, Coenen, Maier, Hurlemann and Bauer2014), alters the sensory quality of social touch (Kreuder et al. Reference Kreuder, Scheele, Wassermann, Wollseifer, Stoffel-Wagner, Lee, Hennig, Maier and Hurlemann2017; Scheele et al. Reference Scheele, Kendrick, Khouri, Kretzer, Schläpfer, Stoffel-Wagner, Güntürkün, Maier and Hurlemann2014) and body odours (Maier et al. Reference Maier, Scheele, Spengler, Menba, Mohr, Güntürkün, Stoffel-Wagner, Kinfe, Maier, Khalsa and Hurlemann2018), increases a tendency to anthropomorphise (Scheele et al. Reference Scheele, Schwering, Elison, Spunt, Maier and Hurlemann2015), or, in rats, boosts the salience of acoustic social stimuli (Marlin et al. Reference Marlin, Mitre, D'amour, Chao and Froemke2015). At present, it is still unclear whether the behavioural effects of OT result from perceptual changes, such as increased attention to the socially informative eye region (Guastella et al. Reference Guastella, Mitchell and Dadds2008), improved recognition of cues about sex and relationship (Scheele et al. Reference Scheele, Wille, Kendrick, Stoffel-Wagner, Becker, Güntürkün, Maier and Hurlemann2013), and/or facilitated sensing of and responding to emotional stimuli (Spengler et al. Reference Spengler, Scheele, Marsh, Kofferath, Flach, Schwarz, Stoffel-Wagner, Maier and Hurlemann2017a). Analysing these effects using observer models may help identify which aspect of the perceptual decision-making process is influenced by OT. As OT is also a promising therapeutic (Hurlemann Reference Hurlemann2017; Palmer & Clifford Reference Palmer and Clifford2017), understanding its mode of action may be informative in order to specifically target dysfunctional perceptual processes particularly amenable to OT treatment.