Final Flashcards
Intelligence
Aggregate ability to achieve goals in the world, adapt to environment, mental self-governance
Minsky cognitive revolution/AI
Importance of knowledge representation in machine so it can think;
Program would need terms for relations & ways of specifying hierarchical level;
Challenges of complex problem-solving: knowledge rep, perception, language & communication
Debate nature of mental representations
Propositional (Pylshyn) vs Kosslyn & Schwartz visual imagery supported by map navigation & spatial transformation
Turing’s role + goal
Universal Turing Machine + Turing test - can machines think?
Grounding
Relationship exists b/t rep & referent eg sensory or H20=water;
Mental representations and decision making
Intentionality
Link between representation and referent
Symbol grounding problem
Symbols must be connected to environment to gain semantic quality. Use embodiment (sensors & effectors)
Role of computation in cognitive science & AI
Mind performs computations on representations
Marr’s Tri-level Hypothesis: computational, algorithmic, implementational level
Classical vs connectionist computing approaches
Formal systems view rule-based symbol manipulation independent of meaning
Connectionist knowledge as distributed patterns of activation, parallel processing
Dynamical view representations constantly adapt to new info
Production rule
Rules used to perform operations required to get to solution;
If A then B
Information processing approach
Automatic parallel processing + serial controlled thought
Churchland’s neurocomputational theory
Most human knowledge should be understood in terms of activation of prototypes within connectionist frameworks
Dennet’s multiple drafts model of consciousness
Sight/sound/touch streams processed in parallel, editing & awareness anywhere in stream
ANN
Information processing, focus functionality, computer simulating operation of neurons
Localist representations
Each node represents single concept/
Activation in single node (compare distributed representation among group of nodes)
Semantic networks
Knowledge representation, focus abstract structure/relationships
Spreading activation
Activity of one node spreads to related nodes, weakens as it moves through links
Help recall through multiple association
Decentralized coordination
Information flow distributed among multiple nodes operating independently, agents rely on local data & interactions, greater flexibility, scalability, resilience
Universal grammar
All human languages share common underlying structure, general principles + language-specific parameters
Visual word form area and dyslexia
Activated more by pronounceable pseudo-words;
Dyslexia linked to reduced activation, lack automatic word recognition;
Application to educational practices: learning to read reshapes cortical networks
Linguistic competence
Formal system overlooking language use/context
Linguistic performance
Deviation from ideal former behaviour attributed to memory, processing capacity etc. How language used in real contexts
Cannon-Bard theory of emotion
Subjective emotion and bodily arousal occur simultaneously
Cognitive theory of emotion
Experience of emotion (physical + subjective) triggered by cognitive evaluation of event
Social brain hypothesis
Large primate brains evolved to manage complex social groups, 2.5yo human child higher social intelligence score than chimpanzees & orangutans
Tomasello & Carpenter 2005 chimpanzee RJA & IJA
Chimps show responding to joint attention but lack initiating JA
Lieberman 2001 cognitive dissonance reduction
Preciously thought required conscious reasoning, but they used amnesiacs to show that behaviour-induced attitude change is fairly automatic, does not require explicit memory or processing
Dual-attitude theory
Evaluative (amygdala threat assessment) vs nonevaluative (slow, semantic & associative processing)
Choice blindness & biases
My side bias, confirmation bias
Bottleneck theories
Explain selective attention/information filtering
Treisman’s attenuator
Threshold effect: minimum activation required to produce conscious awareness, threshold determined by word’s meaning
Multimode model of attention
Moveable filter depending on task demands, early sensory feature selection less demanding
Wicken semantic memory interference
Words can interfere with recall of other words in same category eg apple/orange
Heuristics
Rules of action most often leading to problem solution:
Sub-goals, means-end analysis, diagram, analogy.
Evaluation function to identify promising areas
SOAR approach
State, Order, and Result: represents all tasks and problem space, follows decision cycle. Accumulate evidence/analyze problem, apply operator, evaluate outcome, generate subgoals.
General problem solving applicable to many domains, dynamic learning adjusts to obstacles, mirrors real-word human decision-making. However, assumes unified memory structure
Brute-force search
Trying all possible paths of solution tree;
Problem: combinatorial explosion.
Breadth-first search guaranteed to find shortest path, depth-first search may take many steps if deep solution
EEG strengths
Orthogonally oriented neurons (cortical surface);
High temporal low spatial resolution
MEG strengths
Tangentially oriented (within folds);
Good temporal and spatial resolution, OPM MEG room temp
Contribution of 19th C phrenology
Brain organ with localized specific functions eg Broca & Wernicke’s area
Samuel’s checkers program
Used pruning strategy, learned from experience instead of rules, play against self to improve
Reason AI field has grown according to Stanford
Because lacks single, rigid definition
Stuart Russell’s view on AI
Powerful but risky, can make high-quality decisions but may not share human values, advocating human-compatible AI instead of specific objectives
Singularity
Lines between man and super intelligent machine blurred
Pruning
Cut off unpromising paths (Samuel checkers)
