Social cognition is typically evaluated against normative models, and these evaluations reveal that human inference is marked by several distinctive features: the use of shortcuts or heuristics to make judgments and decisions in a complex and often rapidly changing environment; the role of affective and motivational considerations in inferential processes and outcomes; and the importance of prior theories and expectations for guiding inferential search and outcomes.
Although each of these shortcuts leads to inferences that often approximate those made by normative models, each is susceptible to potential biases. Deciding what data are relevant to a judgment is often marked by prior expectations or theories, sampling is often biased, and biases in already existing samples are often ignored. Strong inferences are frequently drawn from small and unreliable samples. Regression – the fact that extreme events will, on average, be less extreme when observed again – is poorly understood; instead, extreme events are frequently used to predict future extreme events.
How flawed is the social judgment process? Three perspectives on this issue have been voiced. The first position suggests that, at least under certain circumstances, judgmental errors and biases may produce severe distortions, and therefore it is advisable to find ways to correct the inference process. Decision-making methods that can correct for common biases or errors include the use of statistics and computers for aiding judgments. In addition, developing reasoning skills through training improves people’s inferential capabilities.
A second perspective suggests that the experimental literature makes people look worse than they really are, and that intuitive inferential strategies are actually quite effective in the real world. According to this viewpoint, intuitive inferential strategies often serve us well because they take into account efficiency pressures, the specific content and context of an inferential problem, and the possibility that the environment is changing. In addition, some errors may be inconsequential for behavior, others will cancel each other out, and others may be detected through communication.
A relatively new position regarding human inference is that, at least under some circumstances, heuristically based judgments are actually better than more thoroughly considered ones. For example, experts may be able to make rapid use of vast stores of nonconscious knowledge to produce judgments that are more accurate than judgments made via conscious deliberation. For many mundane tasks as well, our minds are able to integrate or extract information from a broad array of stimuli to reach complex judgments in a matter of seconds, even milliseconds. Moreover, under at least some circumstances, conscious reflection on the beliefs that go into our inferences may actually harm the inference process.
Where is the study of social inference headed? Increasingly, insights from social cognition are being integrated with insights from economics and neuroscience, and one outcome of these interactions is the field of neuroeconomics. Neuroeconomics draws on the expected utility normative model, descriptive research from social cognition, and insights and methodologies from neuroscience to identify exactly what neurotransmitters and brain regions are implicated in particular types of inference task. The assumptions guiding this endeavor include the fact that, although normative models are not descriptive of human inference, inferential shortcuts may approximate normative models in ways that are useful not only for identifying what neural mechanisms may be involved in normative calculations but also in the automatic, theory-driven, and affectively based shortcuts that people often use. The potential promise of this integrative effort is that research will be able to identify the interaction of multiple subsystems governed by different parameters and different principles by testing hypotheses derived from normative and descriptive research using the methodologies of neuroscience.