Is prediction nothing more than multi-scale pattern completion of the future?

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• The brain as a prediction machine is a dominant theory in cognitive science.
• The Bayesian brain framework posits explicit prediction generation and error-driven adaptation.
• Alternative mechanisms for predictive processing are being explored.

Purpose of the Study:

• To investigate if predictive processing can emerge from simpler mechanisms than explicit Bayesian inference.
• To demonstrate that pattern completion and dynamical systems can exhibit prediction-like behavior.
• To challenge the necessity of formal probabilistic representations for brain-like prediction.

Main Methods:

• Analysis of pattern completion in visual perception.
• Modeling predictive processing using entrained dynamical systems.
• Evaluation of the TRACE connectionist model for language processing.
• Development of a novel unsupervised neural network model inspired by reservoir computing.

Main Results:

• Pattern completion in perception demonstrates temporal prediction.
• Dynamical systems exhibit predictive processing without explicit prediction mechanisms.
• The TRACE model shows predictive processing hallmarks without explicit prediction or error representations.
• A novel unsupervised neural network displays prediction-like behavior for homeostasis.

Conclusions:

• Brain-like systems can achieve predictive processing through mechanisms like pattern completion and dynamical systems.
• Explicit Bayesian inference and formal representations may not be necessary for prediction.
• Prediction can emerge 'for free' from the dynamics of neural systems.