AI summary by TwinMind
5. Attending: Limiting the Information (audio only)
MIT OpenCourseWare
1h 21m 46s
Introduction and Course Context
- Professor forgot lecture notes but chose attention as today's topic - his research specialty
- Attention research lab pays subjects $10/hour for visual attention experiments
- Attention lecture placed between sensation and perception because humans cannot process all incoming sensory information simultaneously
- Core problem: vast sensory input vs. limited perceptual recognition capabilities
Basic Attention Theory - The Bottleneck Model
- Visual system has severe bottleneck/constriction limiting processing
- Only one thing at a time can pass through recognition system
- Bottleneck governed by selective attention mechanisms
- Analogy: drinking from fire hose requires restricting flow
- Attention allows some information through while leaving other information "on the floor"
Demonstration: Auditory Selective Attention
- Two volunteers (Nina and Zena) read different texts simultaneously
- Audience task: listen to one reader, detect third instance of word "her"
- Results showed successful selective attention to target voice
- Unattended voice: could detect gender, tone, that someone was speaking
- Could not extract content/meaning from unattended stream
- Demonstrates cocktail party effect - can shift attention between conversations
Visual Search Experiments and Reaction Time Methods
- Laboratory method: computer screen displays, yes/no responses, measure reaction times
- Basic features (color, orientation, size) show flat reaction time functions regardless of set size
- Complex searches (T among Ls, 2 among 5s) show linear increases: 20-30ms per item for "yes", ~40ms for "no"
- Suggests searching 20-40 items per second
- Eye movements only 4 per second - attention moves faster than eyes
Basic Features That Escape the Bottleneck
- 12-18 basic visual features can be processed across entire visual field simultaneously
- Include: color, size, orientation, motion, 3D orientation, curvature, line terminations
- Motion detection: easy to find moving item among stationary, hard to find stationary among moving
- Upright vs. inverted faces require serial search despite being salient in real world
Guided Search and Feature Integration
- Real-world searches use basic features to guide attention intelligently
- Conjunction searches (red horizontal line): combine basic feature maps
- Two types of guidance:
- Bottom-up: stimulus-driven (salient oddball automatically grabs attention)
- Top-down: user-driven (knowing target is red allows ignoring black items)
- Top-down guidance can halve search times by eliminating irrelevant items
Feature Binding and Object Recognition
- Before attention: features exist as loose bundles
- Attention binds features together into recognizable objects
- Plus sign experiment: red verticals harder to find when part of integrated objects vs. separated
- Attention directed to objects, not individual features
- Making objects appear separate (depth cues) makes search easier
Change Blindness Phenomena
- Objects collapse back to feature bundles when attention moves away
- Dancing chickens demo: participants miss when chicken falls apart
- Red/green dots: poor performance detecting color changes in uncued locations
- Sistine Chapel: multiple changes go unnoticed except fig leaf (area of interest)
- Dan Simons experiments:
- Gorilla suit walking through basketball game
- Person substitution during door interruption (50% miss the change)
Visual Short-Term Memory Capacity
- Can hold ~4 objects in visual memory
- Color array experiments show capacity limit
- Can track multiple features per object but only ~4 objects total
- Performance drops to chance after 4+ items in memory array
Two-Pathway Model of Vision
- Selective pathway: bottlenecked, one object at time, detailed recognition
- Non-selective pathway: parallel processing of basic features, scene statistics
- Non-selective provides "wallpaper of world" impression
- Selective provides specific object identification
- Together create impression of seeing everything while only recognizing current focus
Neurological Evidence - Agnosias and Neglect
- What pathway (temporal lobe) damage → agnosia: can describe but not recognize objects
- Prosopagnosia: specific face recognition deficit
- Category-specific agnosias (fruits/vegetables)
- Where pathway (parietal lobe) damage → spatial attention deficits
- Balint syndrome: simultanagnosia (only one object visible)
- Neglect: ignore contralateral space/body
- Neglect patients eat only right side of plate, deny ownership of left body parts
