1. One option would be to test the entity in the other room with an infinite loop. I could ask a question such as, “What is the sum of all the natural numbers?” If I never received an answer, I would assume that the entity in the other room was a computer, and that it had gotten stuck in an infinite loop. Unfortunately, this would only work with relatively unsophisticated computers.1
I might try to ask a question that reveals some quirk of human nature. For example, I could ask the entity in the other room to play the “ultimatum bargaining game” with me. (The ultimatum bargaining game is described by Ridley in assignment nineteen’s reading.) In this game, there are two players. Player One is given $100, and he must offer to share a portion of this money with Player Two. If Player Two refuses the offer made by Player One, neither player gets any money. Logically, Player One should offer Player Two a tiny amount of money, for example $1. Player Two should accept, because $1 is better than nothing at all. However, people care so much about reputation and fairness that Player One seldom actually offers so little money. When Player One does offer a tiny amount, Player Two hardly ever accepts it.2
Knowing this, I would offer one dollar to the entity in the other room. If it accepted, this might indicate that it was a computer. (One problem: I’m not sure if this approach is valid in the context of the Turing Test, because it does not fit the format of a simple question and answer. Also, it is probably possible to program a computer so that it will imitate human tendencies such as reputation maintenance.)
I could try spelling some of the words in my question wrong. This might throw off a computer, but it probably wouldn’t confuse a human. I could also try taking advantage of some of the linguistic peculiarities of English, which a human would be more likely to identify than a computer. For example: “Which word ends this sentence? ‘They pray not to become prey, he bawled about being bald, she ate eight cookies, and he filled in the whole _____.” (With the correct answer being ‘hole.’)3
If I was allowed to receive sound from the entity in the other room (rather than just written answers), I would ask it to sing me a song or play a piece on the piano. This would be a useful indicator because computer-generated music usually lacks the subtleties of music played by humans.4
2. Concepts are composed of ideas and their interrelations. They are useful and accessible only to the extent that they are connected to each other. Some concepts are universals, such as “tea” or “warmth.” Some are specific, for example my concept of the warm tea I am currently drinking. The power of thought lies in our ability to connect these two types of concepts. More generally, it is useful to be able to relate concepts to each other, even when the connections are abstract or tenuous.
3. Sphexishness is the quality of acting without an awareness of patterns. Entities acting sphexishly may get stuck in infinite loops and carry out the same action a senseless number of times. Antisphexishness is the opposite; it is characterized by an ability to identify patterns, including in one’s own actions, and to escape from infinite loops. Antisphexishness is associated with conscious purpose and requires the ability to apply one’s knowledge to a situation in order to decide which behavior is most appropriate. Sphexishness is associated with hard-coded behaviors. Every time a sphexish entity receives a certain stimulus, it will complete a given action, whether or not that action actually makes sense in the context of the situation.
Extremely sphexish: Simple computer programs completing tasks such as adding series of numbers.
Moderately sphexish: People with obsessive-compulsive disorder washing their hands again and again and again and again even when there is no reason to do so.
Moderately antisphexish: Teenagers continually changing the way they dress depending on what is in fashion.
Extremely antisphexish: Artists creating many pieces of art using the same medium or general technique.
4. A self-watching computer would have the ability to recognize patterns in its own behavior, not at the level of individual machine-language commands, but at the level of data structures. These observed patterns would be stored in other data structures. There would then be a second level of self-watching data structures, to look for patterns in the first level, and a third level, to look for patterns in the second level. This could keep going to infinity, but after a while, additional levels are meaningless. If a change in the second level creates a very similar change in the third level, which creates a very similar change in the fourth, fifth, and sixth levels, there is no point in having any watching levels beyond the second. To identify a meaningful number of watching levels, there would be another layer of self-watching, called the ω level. However, this does not completely solve the problem, because the computer would need a layer of self-watching to watch the ω level, and so on to infinity all over again.
Unfortunately, with any number of watching levels less than infinity, there is always a possibility that some patterns and meta-patterns will go unnoticed. The only way to achieve perfect self-watching is to make a computer with an infinite number of levels of self-awareness. However, this is impossible.5 (Even if it was possible, it is counterproductive to protect against infinite loops with an infinite loop.) At some point, the levels of self-watching must cut off at a finite number. An imperfectly self-watching computer is still better than one with no self-watching capabilities at all.
Self-watching is important because it prevents computers from being entirely sphexish. A self-watching computer would notice if it was caught in an infinite loop, and it would be able to escape from the loop. Aside from this, self-watching is useful because it gives computers meta-thinking abilities and makes them aware of the connections between their concepts. As a result, self-watching computers are able to form new connections between concepts, resulting in original ideas and behaviors.
5. Sphexishness is characterized by the inability to look at a situation and decide to act in a novel way. Entities acting sphexishly can only behave in “hard-coded” ways, regardless of whether these hard-coded behaviors make sense in the context of the situation. This is the opposite of creativity. Self-watching is an essential prerequisite to antisphexishness, because it allows organism to avoid infinite loops of behavior and to meta-think. Antisphexishness in turn allows entities to act creatively.
1. It is an interesting question of computer science: how difficult is it to program the capacity for the general pattern recognition which keeps a system from falling into infinite loops?
2. This is a specific example of the more general strategy of confronting the Turing test with emotional questions: What is it like to be in love? Describe the last time you were very angry. Why is hypocrisy so upsetting?
3. Thank you for the answer key.
4. Here’s a test: https://www.youtube.com/watch?v=QEjdiE0AoCU
5. One can view this from a metaphysical perspective. For instance, the entirety of all phenomena in the universe can be reduced down to (mostly) the standard model and general relativity. So that’s a case where everything can be simplified through a series of layers that is non-infinite.