Past and present research
Most of our past work has been on characterizing the perceptual, cognitive, and neural basis of judgments from facial appearance. For a review of some of this work, see Todorov & Oh (2021). One of the important contributions of our lab was to introduce data-driven computational methods for visualizing the perceptual basis of social judgments from faces. These models capture systematic biases in judgments. See demonstrations for some of the resulting models of judgments. Recently, using generative adversarial networks we have built models capable of generating and manipulating the appearance of hyper-realistic faces.
Present and future research I
People effortlessly and spontaneously evaluate stimuli in their environment. These evaluations are statistically reliable – consistent within people – suggesting that there is a consistent mapping from perceptual features to evaluations. But to what extent is this mapping shared across people? We use statistical modeling to estimate the idiosyncratic contributions to complex evaluations and preferences (e.g., they account for more than 80% of the variance in the case of perceived facial “trustworthiness” and basic color preferences). We further develop machine-learning methods for building models of individual evaluations and preferences (e.g., for colors, faces, consumer objects).
Present and future research II
We are working on a wide range of projects on how people make decisions in noisy, uncertain environments. Even in environments where the outcomes are transparently randomly determined, people appear to misinterpret lucky or unlucky predictions as indicative of skill (Roberts, Hastie, & Todorov, 2025). We are studying the particular obstacles to learning in noisy, unpredictable environments by building normative benchmarks and comparing human performance to these benchmarks. Current projects investigate strong inferences from small samples, the role of performance feedback, and sensitivity to the redundancy of predictive cues.