Can A Computer Think? The Turing Test For Artificial Intelligence
How do we detect if an A.I is actually intelligent? We should just ask it.
This is an essential idea in the Turing Test. A game where a human judge attempts to detect which entity in a conversation is human and which one is an imitation of human intelligence. An imitation game.
The Turing Test
The father of Artificial Intelligence, Alan Turing, provided a test for detecting human intelligence in a machine. The test is comprised of an interrogator, a human and a machine. The human and the machine engage in a conversation via a text presenting interfaces, such as a teleprinter or a messaging platform. The identity of both entities typing is hidden from the interrogator. The goal for the interrogator is to detect which is the machine. The claim is that if the machine passes the test (the interrogator cannot detect which is the machine), human intelligence can be represented in a machine well. If the machine fails, that does not entail that the machine does not have human-level intelligence.
Can a computer think?
Turing viewed that the question of whether or not a computer can think was not rather interesting. He was more so interested in whether a computer, a symbolic manipulation machine, can implement such a system or algorithm that has no detectable difference in intelligence from a person.
Why would Turing by uninterested in such a question?
It would require an entirely new conception of ‘thinking’ and ‘machines’. People may commonly object to such an idea that a computer can think since they hold so strongly that their thoughts, their mental states are entirely independent substances then their physical body. (To those I say how would two independent substances interact with each other?) With such a conception, a common objection would be to simply deny that a physical computer could possibly come up with thoughts! Thoughts are independent of physical objects!
But, of course, this survey of people’s conception of thoughts and machines would ultimately by unsatisfying, thus what is more fruitful is whether a computer could be programmed to imitate human thought.
What the Turing Test is not.
The Turing Test is not a “litmus test” — failure doesn’t
tell us anything. It could be the case that a human judge thinks that both systems are machines, does that make the human unintelligent? No. If the machine fails at being intelligent, it does not tell us anything at all about its own intelligence.
Intelligence exists in many forms, it is not just through speech. Animals are said to have some intelligence. An animal would surely fail such a test. A human could also fail the test their answers are suspiciously computer-like. Thus, we could not say anything about how intelligent a computer is if it fails.
The Turing probably cannot tell us much about human intelligence either. It may the case that an entirely different system could produce the same consequent intelligence, like Einsteins Relativity and Newtonian mechanics producing similar results on earth.
The Turing test is simply a sufficient test for human-level intelligence in a computer as long as the system is indiscernible from a human in a relevant way.
Problems with the Turing Test.
There are a few objections that Turing presents in his paper, “ COMPUTING MACHINERY AND INTELLIGENCE ” to his test.
- Theology
This objection says that thinking is a function of the human’s soul. Since machines have no soul they have no ability to think.
Turing finds this very problematic since God is all powerful and it can produce anything. This is similar to how, before the Copernican turn, religious believers thought that the earth was a privileged planet at the center of the system. However, we reject that argument in light of empirical evidence. Thus we should do the same for the case the only humans can think. We should show that it cannot or can.
2. Heads in the Sand
This argument against Turing is that the consequences of a machine thinking are so bad that we should hope that one is not possible.
This really doesn’t require an objection. It does not tell us if it is not possible or not, it just tells us not to think about it.
3. Godel's Incompleteness Theorem
One of the stronger objection to the test extends Godel’s incomplete theorem Godel’s incompleteness theorem shows that in any logical system, statements can be formulated which can neither be proved or disproved in the system unless the system itself violates logical rules. An example of this is ‘This machine is always lying’. If the machine can produce such a statement then it gave a definite answer (true) to something that is false.
Turing counters by questioning whether or not humans can make similar mistakes. The Turing test is a test that could expose these mistakes. It is also the case that a machine could just make its own way of proving things and thus capable of doing every proof or making mistakes if it is a limited system.
4. Consciousness
Perhaps consciousness is the most exciting and controversial objection. Isn’t it strange that there is this elusive feeling to experience? That there is something that it is like to be you? Is this not concerning machine intelligence?.
Generally, the argument is since humans require emotions to be humanely intelligent and the machine cannot feel, therefore machines cannot be intelligent in the same way as a human beings.
Turing says that this does not tell us if the test is valid or not, it just tells us a concern for understanding intelligence itself.
This response is a bit unsatisfying from Turing since it what we would require is an understanding of intelligence to see whether a system has the capacity to implement such a system and we could logically conceive of an empty human that is functionally equivalent to a human but in fact does not have the necessary components to be intelligent.
5. John Searle’ Chinese Room — Thought Experiment
The main claim by Searle is that syntactical machines (like digital computers) cannot produce the semantics (meanings) as humans do.
Searle asks us to imagine ourselves for a moment as strictly English speakers “locked in a room, and given a large batch of Chinese writing” and “the second batch of Chinese script” and “a set of rules” in English “for correlating the second batch with the first batch.” The rules “correlate one set of formal symbols with another set of formal symbols”; meaning you “can identify the symbols entirely by their shapes.” Another batch of Chinese symbols and instructions in English allows you to“to correlate elements of this third batch with elements of the first two batches” and tell you to “to give back certain sorts of Chinese symbols with certain sorts of shapes in response.” The symbols are the script, the second batch is the story and the last are questions. You are tasked to give them back answers. However, you do know none of this. You “get so good at following the instructions” that “from the point of view of someone outside the room” your responses are “absolutely indistinguishable from those of Chinese speakers.” Nobody can tell you “don’t speak a word of Chinese.”, as far as the Chinese are concerned, you behave like a computer.
But in imagining yourself to be the person in the room, Searle thinks it’s “quite obvious . . . you do not understand a word of the Chinese stories. “For the same reasons,” Searle concludes that the computer understands nothing of any story since the computer only understands Chinese us much as you do, nothing.
Unfortunately, Turing had passed away years earlier. I am sure Turing’s response would have been brilliant and it would have opened a great debate between the two insightful thinkers.
Conclusion.
Turing proposes a detection test that aims to show whether or not a machine can be good enough to fool a person into believing it has human-level intelligence. Turing does a good job of presenting and refuting objections to a machine having human-level intelligence. Perhaps the Turing test is a good detection test and machines are able to behave with human-level intelligence. What do you think? Is the test good enough, can computers think or are computers inherently limited in some way??
