Mirror neurons fire while both performing and observing an action and enable us to understand and predict what others are doing. This function arises because a) the visual-motor matching of mirror neurons are a consequence of stimulus-response mapping mechanisms that transform sensory input of observing someone else’s action into a matching motor response, or b) we understand what we have done ourselves and what others are doing simply because action and action observation are coded in the same representational format, and mirror neurons are an instantiation of such common coding.

Prediction in the motor domain, but perhaps also in the cognitive domain, is a universal function of the human cerebellum. The cerebellum contains and maintains two internal models of the world to coordinate and control behavior: an inverse model to generate motor commands and a forward prediction model; as well as an error detection mechanism and a learning process that corrects the prediction errors.

The philosophy of science suggests that, on a fundamental level, a scientific theory is only a good theory to the extent that it fulfils a set of basic criteria of adequacy. The study of the predictive mind thus should benefit from an examination and evaluation of the extent to which theories of prediction adhere to these ground rules. There are six reasonable criteria further elucidated in this chapter that are useful to assess the merit of a theory. These criteria are far from perfect benchmarks but, considered as a whole, provide a useful guideline to evaluate theories of prediction. Six criteria are applied to theories of prediction in the remainder of the book. These are: parsimony and simplicity, theoretical precision and mechanistic specificity, testability and predictive power, falsifiability, test of time, and utility. The credibility of a scientific theory is also intrinsically connected to the credibility of the experimental evidence supporting it. This book uses three criteria that provide good benchmarks: the reliability, generalizability, and the validity of the experimental evidence that has been collected.

Connectionist networks consisting of large numbers of simple connected processing units implicitly or explicitly model aspects of human predictive behavior. Prediction in connectionist models can occur in different ways and with quite different connectionist architectures. Connectionist neural networks offer a useful playground and ‘hands-on way’ to explore prediction and to figure out what may be special about how the human mind predicts.

The social environment and our social experiences provide a rich source for social theories about the behavior of others. Important prerequisites are social knowledge about who people are on an individual and group level and the ability to attribute the behavior of others to their mental states, often called mind reading, mentalizing, or theory of mind.

The brain does not passively register sensory input but actively predicts it. The activity of the sensory input is ‘explained away’ and only activity that was not predicted remains. This remaining activity is treated as an error signal that is used to update the predictive coding system. Learning is predictive coding.

Deterministic and probabilistic mathematical theories have in common that they construct mathematical representations of real-world phenomena. On a basic level this can be regarded as a type of explicit problem-solving. This involves presenting the problem in ‘abstract form’ in symbols (often numbers, letters, or geometrical elements). These symbols are then manipulated in accordance with precise rules: Strings of symbols in sets of equations come to represent ideas. The construction of mathematical theories involves testing whether experimental observations fit the postulated ‘mathematical rules.’ If they do not fit then the ‘mathematical rules’ may be refined, extended, or new ones may be formulated. Newly mathematically formalized ideas are validated by testing whether they align with observations but also by examining whether they are consistent with other, previously established, mathematical rules.

Prediction science is likely to push on toward distinct reconceptualizations or the dismantling of the cornerstones of traditional cognitive science, away from rule-based symbol manipulation and toward a comprehensive systems prediction science, toward theoretical unification and simplicity, toward figuring out the pros and cons of the representation-light and representation-heavy, toward incorporating analog representations and common codes, toward proactive, probabilistic, mechanistic, and formalized theories, and computationally specified models of the predictive mind. The paradigm shift of the predictive revolution is no longer only emerging: it is continuing at an ever-increasing pace.

Larger regularities in the environment are coded into mental models, schemata, scene or event knowledge and are the crucial level at which the predictive mind operates. Such event models are used to understand the past and the future. They are structurally close to the situations they describe or predict but are more small-scale and impoverished representations of how the world works.

Prediction makes use of neural rhythms, oscillating at different frequency bands in the brain, that can be detected and exploited for information processing. Temporal prediction may involve entrainment, that is, alignment to the timing and frequency of external rhythmic stimuli, and oscillatory phase resets that are triggered internally by high-level brain processes.

Perception and action are continuously running cycles of sensing and perceiving, predicting, acting, and adjusting. Sensation and perception are assumed to be intrinsically functional and forward-looking in the service of action. This is because relevant information from the environment is needed to guide our actions.

Information processing is a process of uncertainty resolution. Information-theoretic constructs such as surprisal and entropy reflect the fine-grained probabilistic knowledge which people have accumulated over time. The information-theoretic constructs explain the extent of processing difficulty that people encounter, for example when comprehending language. Processing difficulty and cognitive effort in turn are a direct reflection of predictability.

People use fast and simple mental shortcuts for their predictions rather than making weighed assessments and rational decisions based on huge amounts of data. Heuristics are a cognitively cheap and efficient way to solve complex or novel problems because large amounts of the information available in the environment can be ignored.

Mental imagery can be used to simulate imminent, distant possible, or even impossible futures. Such mental simulation enables people to explore the consequences of different actions they want to perform or the consequences of being in different kinds of situations. Predictive simulation retrieves embodied knowledge but also creates new knowledge because people can compare different simulated scenarios and draw conclusions from that.