The Computational Universe

The discussion over artificial intelligence is one that I just can’t seem to make up my mind about.  Like many of my classmates, initially after reading the Searle article I was pretty convinced by his arguments.  I found the Chinese room hypothetical compelling and in general his arguments seemed to validate my previous inclinations against AI.  Yet as I have continued to ponder the possibility of AI my views have continued to fluctuate and I am not sure whether this is the result of Brooks, class discussion, or just my own changing temperaments.  While I do find Brooks’ discussion of the human mind as one big machine with many subparts to be an effective analogy, my indecision surrounding my views on AI seems a prime example of the type of “conscious” processes a machine would lack, whether it is able to pass the Turing Test or not.  Essential to human thinking and cognition is an ability to think critically, to assess and re-assess our views and to attempt to combine our emotions with logic and rational thought to reach our final views and opinions.  Again, my ever changing views on AI is one example of the quest to balance my fear that computers or machines could ever attain the same consciousness as humans and more reasoned and substantiated arguments on the issue as presented by Searle and Brooks.  Yet this incorporation of emotion and irrationality is something that I would think machines would lack, but that I find essential to human consciousness.  I guess ultimately, while my views remain in limbo and after reading Brooks I am more open to the possibility of AI, it’s this absence of irrationality and spontaneity triggered my emotion, mood, interaction etc. that really trips me up when considering the possibility of AI.  I don’t doubt that a computer can pass the Turing test, but I still question the possibility for a machine to achieve what we consider human consciousness.  Nevertheless, considering the possibility has been somewhat mind boggling and certainly put me in touch with my more erratic side.

Searle’s argument frames an interesting question, but it does not eliminate any doubt or prove anything. It does not eliminate the specter that perhaps we can’t conceive of a complicated enough program that could mimic a brain. He poses a strange hypothesis that the brain has a special goo for giving meaning to ideas. Also, perhaps the program for having a strong AI would be conceivably much different than anything I can conceive with my knowledge of limited computer programs.
Also, you can take similar Chinese room experiments and change the parameters so that it does not seem as reasonable. If the brain was blown up to large proportions and a human was inserted into the brain to be in charge of a neuron, then that person would not understand the neuron inputs and outputs, but surely there is still consciousness in the brain. Therefore, it seems unreasonable to say that just because the guy in the Chinese room is just manipulating signals that there is no consciousness occurring.
My major problem with Searle’s article was that he was disguising his arguments regarding Philosophy of Mind in such a way as to make them relevant to computing, but also to take cheap shots at other Philosophers without necessarily making a thorough argument.
Also, though he does make some convincing arguments, his allusion to Occham’s razor rings somewhat hollow. It seems that the nature of the machine that most effectively simulates consciousness will probably be very complex and a degree of understanding removed from what we already think.

My understanding of Artificial Intelligence has evolved considerably over the course of this semester.  My initial understanding was guided by Brook’s logic and many of his points still hold true for me.  Brook’s logic remained intact in my opinion until we got to the Scribbler lab in which we were to create some form of artwork with our Scribbler.

Our class discussion the other day highlighted Searle’s main point, that while computers may pass the Turing Test, it is still not a valid test of consciousness.  Searle’s argument leaves me wondering about the features that would be necessary for a computer to have limitless creative capacity.  When I think of our Scribbler lab I remember that the closest my Scribbler could get to creativity was randomness followed by a programmable action—which in my opinion is still very far from real and conscious creativity.  Placing creativity on par with consciousness, I find Searle’s argument most compelling in that the Turing Test cannot determine the consciousness of an entity, despite the fact that it may be possible for a computer to pass the test.

It is difficult to side with either Brooks or Searle about whether or not it will ever be possible to duplicate the mechanical properties the brain as well as the rules that govern its actions.  Unlike Searle, I’d like to think that AI is possible at some level yet I will take from him that the Turing Test is not the proper benchmark of determining whether or not a machine is intelligent.

When I was reading Searle’s article, especially his discussion about the distinction between syntax and semantics, I found myself thinking of a famous thought experiment from philosophy of mind. In debating qualia or the qualitative aspects of consciousness (akin to what Searle calls semantics) philosophers have proposed the following scenario:

Imagine a woman named Mary is in a room and she has had an operation where she can only see in black and white. She has access to all the cutting edge medical technologies and research and becomes the world’s foremost expert on color vision. In particular, she knows everything there is to know about seeing the color red (wavelengths, rods, cones, etc). No matter how much she knows, however, we have an intuition at least that if the procedure is reversed and she is taken outside the room and sees something red, then she will learn something new namely what it MEANS to see the color red or what red looks like.

Some argue that Mary will not learn anything new when she first sees red in which case artificial consciousness seems plausible. However, if she does learn something new then there seems to be something about qualia intimately bound up in consciousness which does not bode well for AI. I am not sure if this connection I am drawing is exactly correct, but it is what I found myself thinking when I was reading Searle’s article.

One of Professor Arora’s lecture slides also brought up an interesting point: surely we would all concede that human consciousness is a strong form of consciousness. There are other life forms that we would admit to be conscious, but in a much weaker sense. Could then computers become conscious in some weaker sense? I think this has to seem much more plausible than a computer passing the Turing Test. However, the difficulty becomes how would we test if a computer has become conscious in this weaker sense? Could the Turing Test be extended to say that if humans and some examples of whatever given animal cannot tell the difference between that animal and a computer, then a computer has attained that level of consciousness? On an intuitive level this seems wrong (it is no stretch to imagine a robot that might fool a dog into thinking it is a dog, and because humans are not well acquainted with the consciousness of dogs, also fool a human as well), yet it is hard to think of a better analogous test to the Turing Test.

Anyways, sorry to ramble, just thought I would share some of those reactions to Searle’s article and our larger AI discussion which I have really enjoyed.

After re-reading Searle’s article and scanning over a few of the other students posts, I am probably as confused as I started when it comes to this AI business. I think the best way to go about understanding AI and making either an argument for or against its potential is to first just speak openly and not worry about potential counterarguments. I found that when I would try to argue that the Turing Test really wasn’t a suitable test of AI, that I could easily tear my own argument down. So I’m going to stick to one side of the debate in this one and simply say that I find a major difference in thinking and performing human-like functions.

To me, there is a very real difference between the abilities of AI to convince people that it is “human” and actually proving its ability to think. To be human is very much contingent on thinking and reasoning, two factors that make up general speech and conversation. Sure, you could probably hold a respectable conversation with a strong AI machine, and maybe that machine could even pass the Turing Test…but what does this mean for society at large? In my opinion, it does not mean much. I think that the Turing Test is an oversimplified and convenient way of seeing if AI is “smart”. As I sat in lecture the other day, I couldn’t help but think about the number of incredibly smart people in the world who would not pass a Turing Test. Does this not make them real or human or intelligent? If anything, it makes them incredibly more human than the oversimplified machines that exhibit AI. It is a confusing and messy thing to try to determine, but I will certainly maintain that a machine will never reach the level of human interaction and intelligence that occurs in real beings.

Despite our class discussion of Searle’s article, I still think there is a clear conclusion in the article that computers can never think. I don’t really agree with the fact that Searle’s only objective is to show that the the Turing Test is not a good and reliable measure of “thinking”. The example with the Chinese room has a really convincing and deep meaning - it does not only ridicule the test consisting of (in our case) speaking with a computer on a few topics but it also says that “thinking” is not the ability to pick from a list or “a rule book” and combining theoretical principles. Thus, anything that has the ability to think doesn’t consist of rules only - what is then a computer - a program written with fictional symbols - 0s and 1s. If the human brain worked like this, there would be no sciences like psychology, biology, etc. Therefore, “thinking” is not at all an integral part of anything we have been able to construct so far. I don’t really like the way, however, Searle relates the terms “semantics” with the rest of his paper - it is not even necessary to use it in order to achieve the aims of the paper.

Now, I would like to address a few issues we talked about in class. The argument that the Turing test is a fair (no obvious way to cheat ) one is not really convincing. Although there are a lot of different combinations of topics, it is relatively simple to produce relevant databases which comprise of the variety of topics a “normal” person would talk about, and thus simulate a Turing test passing computer. It is in fact even possible to create a program which learns from the people it talks about and imitates ways of writing and stylistic patterns. There is, however, a “slight” difference in the word “thinking” - in my opinion, it should relate more to the class of feelings than to the class of actions.

In conclusion, although the matter is somewhat philosophical, it is, at the current level of science, impossible to program a computer to “think”. This is, however, not at all discouraging as computers can be programmed to do almost any task (including decision making, using algorithms or approximations) any human can do and, further, tasks which will require immense physical or intellectual work (computations and a lot more). Thus, we don’t really need the computers to be able to “think”.

I really enjoyed reading your responses. Clearly this topic touches something deep down in all of us. A couple of final thoughts.

First, are humans machines as Brooks suggests? All of us in CS116 think we are not. One property of a machine is that it is manufactured —you can buy as many identical copies as you can afford. But as we discussed in class, if human cloning ever happens, this picture starts getting really muddled. That is one of the reasons most of us find cloning abhorrent. On the other hand, people have cloned sheep and dogs already, so how far off is human cloning?
Second, I want to raise the problem of the hypothetical bunny rabbits once again. What if there had been two intelligent species on earth? Dolphin lovers, or chimpanzee researchers will argue that there already are. But assume this other specie is so advanced it actually can carry on a conversation with us. We would obviously not need to consider them human or want to date them (at least I can’t imagine dating bunny rabbits). But how would we react to them? Speaking for myself, I can’t be sure.
I fully grant you that the bunny rabbit situation may be too hypothetical. We may be the one and only example of intelligence in the universe for all time.

To sum up, I hope it was useful for you to confront all these possibilities for a few hours or days. Unlike Brooks or Searle, I am not vested or fixed in my views on AI. If I start thinking about it, things get very confusing very fast. (Cardero, Mike, Lily etc., you are not alone.)