A generative Model
aims to capture the statistical structure of some set of observed inputs by tracking […] the causal matrix responsible for that very structure. A good generative model for vision would thus seek to capture the ways in which observed lower-level visual responses are generated by an interacting web of causes – for example, the various aspects of a visually presented scene.Clark (2013), 2
In other words, a generative model predicts the causes for what we perceive.
This can happen on multiple levels,See Ramstead et al. (2019) for “a multiscale ontology
of cognitive systems“, i.e. a framework for integrating accounts of
systems on these different levels.
from individual perception to scientific explanation: our
brain tries to figure out (predicts) the objects and movements that
cause our visual perception; a scientific model tries to figure out
(predicts) the systems and processes, entities and laws that cause
experimental observations.
Thus a generative model can be an Implicit Model or an Explicit Model.
References
- Clark (2013): “Whatever next? Predictive brains, situated agents, and the future of cognitive science”
- Ramstead et al. (2019): “Multiscale integration: beyond internalism and externalism”