···
Dick Robertson (2.28.2015)
DR: First, Happy Birthday, Rick.
RM: Thanks so much. Great to hear from you Dick!
DR: I liked your discussion of where Turing made a mistaken assumption about the way behavior works, but could you give an example of what you would think would be a test the Turing machine would fail after passing the typical human simulation task?
RM: As I understand it, the Turing test involves asking questions of a person or computer in another room and, based on the answers, declaring which is the computer and which is the person. If you are right only 50% of the time then the computer is considered to be a good imitation of a person (Hence the “Imitation Game” of the movie title).
RM: I think Weizenbaum showed, with his ELIZA program, which responds to questions with canned answers selected based on key words in the question, that people find it very tough to tell whether the answers are being given by a human or Eliza. But there are probably ways to do the questioning so that you can distinguish a person from a pseudo-person like ELIZA.
RM: I think ultimately the major flaw of the Turing Test is that it can be easily faked so that a person who wants to can make him/herself indistinguishable from a computer. In other words, people can have the purpose of fooling the questioner. The fact that people can have purposes like this – any purposes at all, really – is what is missing from the Turing Test. The Powers Test, on the other hand, cannot be faked. You cannot fake having a purpose in the sense that you cannot make it seem that you have a particular purpose when your purpose is actually different. People can do this in real like because it is hard to do Powers Test (for the controlled variable) in many rel like situation. So people know how to fake having purposes (usually by simply saying that their purpose is X) which they don’t actually have; it’s called lying. But you can’t fake having a purpose (can’t fake controlling a variable that you are not really controlling) when you use the Powers rather than the Turing Test.
RM: This is demonstrated in my “Detection of Intention” Demo (http://www.mindreadings.com/ControlDemo/FindMind.html) where 4 cars are seen following a red sports car. But only one of the three following cars has the purpose of following the red car; the rest are just moving in the same path by coincidence; they could be seen as non-purposeful impostors. The one that actually has the purpose of following is readily exposed when you move the mouse, to disturbance the path of all cars; the purposefully following car resists this disturbance to keep following the red car.
RM: Anyway, that would be my critique of the Turing Test. It can be faked. Faking is a purposeful activity. Purpose is not taken into account by the Turing Test. Hence, the Turing Test is not a good way to distinguish purposeful people from purposeless machines. Or even distinguishing purposeful machines from purposeless ones.
RM: Whaddaya think?
Best
Rick
Best,
Dick R
–
Richard S. Marken, Ph.D.
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble
On Sat, Feb 28, 2015 at 1:31 PM, Richard Marken csgnet@lists.illinois.edu wrote:
[From Rick Marken (2015.02.28.1130)]
philip 2/26. 17:30
PY: I also wanted to make this connection between Alan Turing’s work and PCT…
PY: In case you didn’t hear, Turing committed suicide in 1954, thus preventing him from realizing how relevant PCT was to the theory of computation and vice versa.
RM: I was very interested in Turing’s work back in grad school. I even wrote a story called “Testament to Turing” for one of my classes (a fun class where the final assignment was to write a science fiction type story illustrating some psychological concept) . It was about a psychiatrist who developed a simulation of human behavior that passed the Turing Test (it’s behavior was indistinguishable from that of a human) and so he used the program to simulate a patient in order to test the effects of different therapies on the program before trying them on the patient. You can guess how that went;-)
RM: But I wrote the story well before I knew anything about PCT (it was written in about 1970 or so; wish I’d saved a copy). So while I suspected that there was something wrong with the Turing Test I didn’t know what was really the problem (I was a fan of Joseph Weizenbaum’s critique of Turing type AI; Weizenbaum was the creator of the ELIZA therapy simulation program, which is probably where I got the idea for the short story). Now I know that the problem with the Turing Test is based on the tacit assumption (made by all psychologists, then and now) that behavior is generated output, like the printed output from a computer.
RM The behavioral output of greatest interest to Turing was natural language. So the Turing Test was based on asking questions of what might be a person or a computer and noting the responses (behavior). If the responses (output) of a computer could not be discriminated from those of a person then the computer passed the Turing Test. After learning PCT I realized that all the shortcomings of this test (such as those noted by Weizenbaum and many others) system from the fact that behavior is not generated output (generated by a Turing Machine) but, rather, controlled input (controlled by a control system).
RM: Moreover, PCT shows that behavior is the control of a hierarchy of controlled perceptions and language is just one of those perceptions; the lower levels are the perceptions controlled as the means of producing sentences (the perceptions of such things as phonemes, morphemes and grammatical structure); the higher levels are the levels that use language as the means of achieving their goals (like explaining how to do things or or what kind of help wanted to accomplish their goals).
RM: PCT also provides an objective criterion for determining whether the behavior of an artifact (such as a computer simulation) is equivalent to that of a person. The behavior of the artifact is equivalent to that of a person if it controls the same perception(s) using the same means.
RM: Given the current interest in Turing (a brilliant and tragic man who I admire greatly) thanks to the movie (“The Imitation Game” ) about him I think it would be great publicity for PCT if someone would write a paper comparing the Turing Test to the Powers Test (the test for the controlled variable) as approaches to determining whether the behavior of a computer is like that of a human. Any volunteers?
Best
Rick
Richard S. Marken, Ph.D.
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble
–