Part 2 in the series is Computing Machinery and Intelligence by Alan Turing.
Published in 1950, this landmark paper introduces many topics in Artificial Intelligence (A.I.), including the Turing test and machine learning.
The key topic explored in the paper is the question, “Can machines think?” However, since the words “machine” and “think” have hard to define meanings, Turing defines the “machine” in question as a finite-state-machine (FSM) and introduces The Imitation Game.
The Imitation Game
Instead of attempting to define what “think” means, Turing replaces the question with whether a FSM can pass the imitation game. The game is as following:
There are three people in the game: a man (A), a woman (B), and an interrogator (C). All of them are in separate rooms, communicating only through writing/typing. The objective of the game is for C to find the true genders of A and B by asking each of them questions. However, A tries to misguide C by giving false answers while B does the opposite. Thus the original question “Can machine think” is replaced with “Can a machine imitate the behavior of either A or B?”
This game is more widely known as the Turing test, which is generalized as whether a human can distinguish a computer through conversation. Thus “think” is implied to mean the ability to generate human cognitive ability.
Turing defines the machine in question as a digital finite state machine consisting of three parts:
- Store: The memory of the computer. Assuming unlimited store size helped Turing to refute many objections to AI.
- Executive Unit: “The part which carries out the various individual operations involved in a calculation”. The CPU of a computer…
- Control: “Table of instructions”, think programs.
Turing then mentions the equivalence of all digital computers, based on that any FSM can “mimic any discrete-state machine”, provided that the store is unlimited.
Turing then proceeds to give counter-arguments to 9 common objections to why even passing the imitation game does not mean a machine can “think”:
1. The Theological Objection
The objection is summarized as
Thinking is a function of man’s immortal soul. God has given an immortal soul to every man and woman, but not to any other animal or to machines. Hence no animal or machine can think.
Turing poses a brilliant counter argument (as well as his honest opinion “I am unable to accept any of this) in theological terms. Turing observes the above objection puts a restriction on God’s omnipotence…why can’t He give an elephant a soul if He sees fit?
Turing then suggests the theological objection is nullified by saying
In attempting to construct such [thinking] machines we should be irreverently usurping His power of creating souls, any more than we are in the procreation of children: […] we are instruments of His will providing mansions for the souls that He creates.
2. The “Heads in the Sand” Objection
The objection is summarized as
The consequences of machines thinking would be too dreadful. Let us hope and believe that they cannot do so.
Turing points out this occurs especially frequently among intellectual people, because “they value the power of thinking more highly than others”. This objection is dismissed due to argumentum ad consequentiam.
3. The Mathematical Objection
This objection is a little bit confusing, as I have to read a bit more on Godel’s Incompleteness Theorem. Here’s the short version: opponents of AI say the machine can make mistakes in the imitation game, Turing says that’s ok because a human player can make mistakes too…and us humans really need to get off our high horses.
The theorem basically states that all consistent formulations about number theory (or the relations between all natural numbers) contain unprovable propositions.
By applying this theorem onto the machine, the opponents of AI says there are limits to what a machine based purely on logic can answer, and thus in the imitation game, the machine could either give a wrong answer, or never reach an answer (infinite loop). Turing counters that us humans are very often wrong as well and uses this argument to be pleased at machine’s fallibility.
Reminds me just a little bit of the Halting Problem
4. The Argument from Consciousness
In the words of Professor Jefferson’s Lister Oration for 1949, the objection can be summarized as “Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain”.
This is similar to the famous Chinese Room argument about whether a machine can have a mind. However, Turing points out, reminiscent of the Matrix movies, that we have no way of knowing if anyone other than ourselves feel emotions. Therefore we should accept the test as a measure of thinking.
5. Arguments from Various Disabilities
The arguments are summarized as
I grant you that you can make machines do all the things you have mentioned but you will never be able to make one to do X.
where X ranges from “have a sense of humor” to “make mistakes” to “make someone fall in love with it”. Turing sure has a subtle sense of humor!
He then points out many of these arguments rarely have any legit support other than naive application of scientific induction: someone never sees any machine that enjoys the strawberries and cream, and therefore decides that no machine can ever enjoy the strawberries and cream. He compares this to “otherwise we may (as most English children do) decide that everybody speaks English, and that it is silly to learn French”.
Turing also makes specific counterarguments to some X’s:
- Machine cannot make mistakes: one can program machines to look like they are making mistakes.
- A machine cannot be self-aware: A program that monitors a machine’s own internal states can be written.
- A machine cannot have a diversity of behavior: With unlimited storage, unlimited sets of behaviors can be instituted.
6. Lady Lovelace’s Objection
The argument is from Ada is:
The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform. It can follow analysis; but it has no power of anticipating any analytical relations or truths.
Turing points out that Ada was disadvantaged by a lack of knowledge at the time of her writing. Further, the objection can be nullified by a the concept of machine learning later introduced in the paper.
7. Argument from Continuity in the Nervous System
This objection says that because the nervous system has inherently analog components and mechanisms, a digital computer cannot imitate thinking behavior.
However, Turing observes that any analog system can be modeled by a digital system to a certain degree of accuracy such that the human interrogator in the Imitation Game cannot distinguish.
8. The Argument from Informality of Behavior
This argument states that “it is not possible to produce a set of rules purporting to describe what a man should do [or think] in every conceivable set of circumstances”, and since we can produce such a set of rule for a machine (its program), then machines cannot imitate thinking behaviors.
Turing counters this by saying that just because we can’t find a set of rules of conduct for humans right now, doesn’t mean that such a set does not exist. “We certainly know of no circumstances under which we could say, ‘we have searched enough. There are no such laws'”.
Indeed, this is what heuristics is all about.
9. The Argument from Extrasensory Perception
I actually laughed out loud upon reading this. It seems Turing believed strongly in ESP and “this argument is to my mind quite a strong one”. He suggests a specific argument based on ESP:
Let us play the imitation game, using as witnesses a man who is good as a telepathic receiver, and a digital computer. The interrogator can ask […questions about cards he’s holding]. The man by telepathy gives the right answer 130 times out of 400 cards. The machine can only guess at random, and perhaps gets 104 right, so the interrogator makes the right identification
Turing can find no counter-argument other than suggestion perhaps the machine’s exposure to telekinesis will cause it to guess right more often…our they can just ESP-proof all the rooms.
Finally, Turing discusses how one would go about building a machine that imitates an adult human mind. He breaks down an adult mind into three components:
- The initial state of the mind, say at birth.
- The education to which it has been subjected.
- Other experience, not to be described as education, to which it has been subjected.
He then suggests instead of simulating an adult mind, construct a “child” mind and takes it through the same education and life process. Concepts such as conditioning and (sort of) genetic algorithms are also explored.
This paper was a great read, more than the ahead of his time AI concepts, I enjoyed Turing’s wits in parrying the objections and the subtle humor. Highly recommended!