CSCE 420 Lecture 2
Jump to navigation
Jump to search
« previous | Thursday, January 17, 2013 | next »
Intelligence
What is Artificial Intelligence?
We ask ourselves a more fundamental question: What is intelligence?
- Inter... (Interpretation?)
- Learning/Adapting
- Problem-solving
- Goal-oriented behavior
- Responding: perception of environment
- complexity/difficulty [1] — expertise
We have tests to measure intelligence: IQ, SAT, GRE
- sometimes we do poorly on a test, but it does not describe our true ability/potential
- Intelligence is multidimensional
Are humans the only measure of intelligence?
- dolphins
- non-human primates (great apes)
- use of language (even fuzzy among humans, plausible in dolphins, but probably not among other animals)
(Leave it open-ended)
| Thinking like a human | Thinking ideally |
|---|---|
| psychology behaviorism = stimulus/response cognitive = mental models of representing concepts, imagery,and productive systems |
philosophy Sometimes humans don't think ideally… |
| Acting like a human | Acting ideally |
| operational[2] implementation doesn't matter Turing test: evaluate whether computer is intelligent (indistinguishable from human behavior) by interacting with it (winner gets the Loebner prize |
rationality: doing the right thing optimization (heuristic/approx. solutions to NP-Hard problems) control theory objective performance |
Philosophy
- Aristotle: syllogism (A → B ∧ B → C. ∴ A → C; e.g. Socrates is a man, all men are mortal, therefore socrates is mortal)
- Renaissance: formal logic and mathematics (Boole, Frege, Tarski, Leibnitz, Wittgenstein); incompleteness (Gödel's theorems)
- Can intelligence/knowledge be represented as logic?
- Can intelligence be represented as a computable function? f(state, goal) -> action
- Does an intelligent system require embodiment/grounding? (any intelligent being/system has to be connected to the real world: sensors)
Psychology
Strengths
- Interpretation: we are good at visual/auditory perception and parsing natural language
- Disambiguation: we automatically choose the most likely interpretation
- Expertise and Judgement
- Creativity
- Analogy: we can solve a brand new problem by relating it to a previously solved problem
- Adaptiveness
- Rationality
Weaknesses
- Irrational: influenced by emotions, morals, ethics (in good ways)
- Biased
- Calculating (with the exception of Arthur Benjamin)
- Memory limitations [3]
Architectures
- Symbolic Systems Hypothesis (Simon, Newell; CMU; 1960)
- Using associations/pointers/links/references to represent concepts
- Manipulation of symbols in algebra (CAS), etc.
- Knowledge-based Systems (grew into Expert systems)
- Connectionism: neural networks, distributed encoding (abstract representation of neurons as nodes; each neuron has certain threshholds)
- Uncertainty, Bayesian probability