Introduction to AI 2

by Albert Major

1. Introduction

• The lecture is about how humans see (visual perception) and how computers “see” (computer vision), and the similarities and differences between the two.
• Focus is on the biology of seeing, history and progress in AI vision, and deep learning methods (like neural networks and transformers).

This slide shows a classic visual illusion, where two squares look different in brightness, but are actually identical—demonstrated by the cartoon elephant and mice. The lecture used this to illustrate unconscious inferences: our brain interprets images using context and prior knowledge, so we “see” differences that aren’t physically there. This illusion highlights how visual perception is under-constrained and the brain automatically fills in missing information, making vision much more complex than just receiving raw data

2. Visual Perception is Hard

Key Idea: Unconscious Inferences

• Unconscious inference: Your brain makes automatic, fast guesses about what things are, based on limited information and what you already know from experience.
• Visual perception is “under-constrained”:
  • Definition: The raw information from our eyes isn’t enough to figure out exactly what things are. The brain “fills in the blanks.”
  • Example: Ambiguous images (like an unclear drawing) can look meaningless until you recognize a familiar shape, at which point it “pops out.”

Key Idea: Prior Knowledge

• Prior knowledge: Everything you’ve seen and learned before, which helps your brain interpret new images.
• Application: Babies learn to see by building up this “library” of visual knowledge as they grow up.

Key Idea: Visual Illusions

• Visual illusion: When our brain interprets the same image differently based on context.

• Example: Two squares with the same color can look lighter or darker because of shadows or surrounding images. The brain tries to “discount” the effects of lighting and makes inferences about 3D scenes.

  

  This slide shows the eye and optic nerve connecting to the brain, introducing the biological basis of visual perception. It highlights how visual information travels from the eye to the brain for processing—key for understanding how seeing starts at the retina and is processed in the brain

  

  This slide diagrams the retina’s structure, showing how light hits photoreceptors (rods and cones: S, M, L) and signals pass through layers of retinal cells before reaching the optic nerve. It illustrates the “inside-out” design (wires in front of sensors) and emphasizes that the retina itself does complex pre-processing before visual information ever reaches the brain

  

  This slide shows how the optic nerves from each eye meet and cross at the optic chiasma, sending information from each visual field to the opposite side of the brain. It illustrates the neural pathway that organizes visual input for further processing, which is key for depth perception and integrating information from both eyes

  

  This slide shows how visual signals from each eye are split and cross over, so the left visual field from both eyes goes to the right visual cortex (V1), and the right field goes to the left V1. It explains how the brain integrates input from both eyes to build a unified visual world, a key concept for understanding visual processing and depth perception

3. How the Visual System Works (Biology)

3.1 The Eye

• Retina: A layer at the back of the eye that contains light-sensitive cells (photoreceptors).
  • Rods: See brightness (work well in low light).
  • Cones: See color (three types: S for short wavelengths/blue, M for medium/green, L for long/red).
• Wiring: The nerves that carry information from the retina are actually in front of the photoreceptors in the eye (“inside-out design”). Evolution solved the problems this creates.