Minds and machines

Table of Contents

Minds and machines

Prev: reducibility-and-gödel Next: complexity

Completness vs. Incompleteness

Godel published an incompleteness theorem as well as a completeness theorem.

There are three things to distinguish them:

  1. True: The statement is true in the realm of positive integers
  2. Entailed by the axioms: the statement is true in any situation where the axioms are true
  3. Provable from the axioms: the statement is derivable by starting from the axioms

The completeness theory equates provability with entailment. Thus, if something is logically entailed by the set of axioms, then it is also provable from said axioms. The incompleteness theorem differentiates entailement from truth. There is no set of axioms that captures all and only the true statements about the integers. Any set of axioms that does, will also describe other universes, and if it is true for the integers but not other universes, it won’t be provable.

Implications

There are some actual questions where it is required to prove a statement like G(F) without it being proved in F itself.

Continuum Hypothesis

Cantor also had a question if there was any infinity in size between the real numbers and that of integers. He formulated the Continuum hypothesis stating that there is no set strictly between that of the integers and the real numbers.

Godel and Cohen’s results

Godel showed that the Continuum Hypothesis can be assumed consistently. Cohen in 1963 proved that you can assume the Continuum Hypothesis is false without introducing inconsistency.

Thinking Machines

The dream of computing would be to realize “Artificial Intelligence”, or a machine that can think like humans can.

Turing Test

Turing proposed a criterion called the Turing Test to distinguish between humans and machines. If a human interacting with a machine cannot reliably distinguish it from a human, then the machine ought to be regarded as intelligent, just like a human.

This is probably a sufficient but not necessary condition for intelligence, since many people believe AIs are actually human.

Searle’s Chinese Room

Searle’s experiment involves locking Searle into a room where a person inserts slips of Chinese characters, and Searle has a textbook inside the box that can translate the chinese into chinese, and respond. Thus, Searle could fake speaking Chinese without understanding Chinese.

One rebuke is that Searle may not understand Chinese, but the system (the textbook, and its operator) do understand it. As well, with enough computational power, it may be possible for the brain to do something like this yet get a lot more mileage out of it.

Competing Analogies

Basically, the question boils down to this. Is intelligence something that can be emulated by calculation (which computers can do) or is it something more?

The Practical Question

Is it possible for a machine to pass the Turing Test? Definitely. We have ChatGPT in 2022, but also, CAPTCHAs, which are supposed to be Turing tests for web bots, have also been broken for a long time.

Godel and Thinking Machines

Penrose used Godel’s Incompleteness Theorem to prove the impossibility of thinking machines. Since \(G(F)\) is not provable in \(F\) by any machine in \(F\), but provable by us, looking at the universe as a whole, Penrose believes that we must have some extra capability that machines do not have.

Views of Consciousness

  1. Simulation produces consciousness (Turing)
  2. Consciousness can be simulated, but mere simulation does not produce consciousness (Searle)
  3. Consciousness cannot even be simulated by computer, but has a scientific explanation (Penrose)
  4. No scientific explanation (Most people)

Prev: reducibility-and-gödel Next: complexity