Perceptual distortions and deceptions: what computers can teach us.

Area of Science:

• Neuroscience
• Psychology
• Philosophy of Mind

Background:

• Perception's nature is a long-standing philosophical inquiry.
• Perceptual abnormalities are hallmarks of psychiatric disorders, including hallucinations and illusions.
• Differentiating normal from abnormal perception presents a significant challenge.

Purpose of the Study:

• To propose a quantitative distinction between normal and abnormal perception.
• To explore the role of prior expectations in perceptual inference.
• To connect theoretical models of perception with practical applications in computer vision.

Main Methods:

• Review of philosophical and scientific literature on perception.
• Theoretical argument framing perception as Bayesian inference.
• Illustrative example from Google's computer vision research.

Main Results:

• The distinction between normal and abnormal perception is proposed as quantitative.
• The degree of influence from prior expectations on perceptual inference is the key differentiator.
• Computer vision models demonstrate how prior expectations shape inference.

Conclusions:

• Abnormal perception arises from excessive influence of prior expectations on perceptual inference.
• This quantitative framework offers a new perspective on understanding perceptual disturbances.
• The findings have implications for both clinical psychology and artificial intelligence research.