R. Port, Q500, Intro to Cognitive Science
Universal Turing Machines, 1937 - his main work on computing. `Turing machines' - very simple device, unlimited tape, finite set of states, read/write head, move L/R - can do ANY calculation for which an algorithm exists
Algorithm - an explicit series of instructions that are guaranteed to reach a final state.
Unification of disparate areas of math. Post Production Systems Church's Lambda-Calculus
Church-Turing Thesis - informal thesis, not a formal proof
`Any function that can be computed algorithmically can be computed by a Turing machine'
Limits. In addition to unification of math, it was also discovered that there are inherent LIMITS on what formal systems can do. Eg, Godel's Theorem: Any formal system is incomplete. There are truths that it cannot express.
Concl: There's reason for both joy and despair.
But Turing went further. He wanted to show that algorithmic solutions can simulate even a Mind well enough. Just make the test fair: allow verbal IO only. People will probably be fooled.
Turing's Test `Can machines think?' How to test? 1950 - essay on artificial intelligence
The Imitation Game two teletypes, trading questions
The Objections (from Turing)
1. Theological. Why couldnt machines have a soul?
2. Head in the sand.
3. Mathematical objection - based on Godel
4. Consciousness. Does one have to BE a person?
5. Disabilities: But it cant do X where X is: love, be friendly, etc - Just a failure of imagination
6. Lady Lovelace: `machines only do what we tell them'
7. Machines are discrete unlike us. But we can approximate as close as you please.
8. Telepathy.
9. Learning. But machines can learn too.
Problems
1. Verbal-only constraint; tests only stuff you can TALK about
2. Input-Output format; no time constraints (except to be fast enough)
3. Is this a deep, metaphysical, UNIVERSAL notion of COMPUTING? No. Its just a behaviorist-influenced notion, an I/O-based notion, a static one (fast enough only). Much that is potentially essential is left untested - defined as irrelevant.
4. Can we discover the TRUE nature of human COGNITION from from this definition? I doubt it. Much that is essential and central is defined as irrelevant.
Can the test be passed? Apparently yes. It's all too easy to fool the naive. Cf, Eliza, Doctor, etc.
Weak AI - models intelligence as well as possible
Strong AI - claims AI programs are genuinely, really intelligent
The Setup. Searle (or whoever) is passed notes in Chinese and executes the rules, but never understands Chinese. Clearly execution of formal rules is insufficient to exhibit intelligence.
Conclusion. Thus, executing formal rules is not the essence of intelligence.
Searle's Replies. 1. System 2. Robot 3. Brain Simulator 4. Combination 5. Other Minds 6. Many Mansions
Discussion
1. Wrong kind of `stuff' for intentionality p. 200
2. Simulation: IO concept ``confuse simulation with duplication'' p. 201
3. computer has syntax but not semantics p. 202
4. confusion about ``information processing'' p, 201
5. residual behaviorism and operationism, p. 202
6. residual dualism, pll 202-3