The Computational Universe

Before this course, I had very minimal study or knowledge of specifics concerning artificial intelligence.  Like most in my generation, I grew up watching science fiction movies and watching the maid robot on “The Jetsons” but it never really crossed my mind that it was and could be a reality.   My thoughts were always that robots could be taught to do many things, but they were never truly “intelligent” because they couldn’t truly learn and use judgement like a human.

Taking this course has shown me, however, that if we are able to map out and come to as close as possible to predicting human actions, then we can have a robot do it as well.  The ever increasing nature of technology, as well as our greater knowledge regarding the workings of the human brain have convinced me that artificial intelligence is not only a possibility, but the future. Once I realized this, I began to get frightened, because it then occurred that perhaps we can get to far ahead of ourselves and create intelligence that is TOO intelligent, which then makes me think of the Terminator movies with ARNOLD(Scary indeed!).

Up until a few months ago, when I heard the phrase “artificial intelligence,” I immediately thought of the Haley Joel Osment movie, “A.I.”  As funny as that now seems, I think it shows how our conceptions of complicated issues in technology/computer science can be shaped by pop culture/media references.  Because intelligence is such a fundamentally human concept, it follows that we (unless we know better) should associate it with artificially created beings that walk, talk, act, and look like people.  But now I understand that our best approximations of intelligence have little to do with robots in the shape of little boys, and more to do with cognitive processes that equate to thinking.

Despite some flaws in John Searle’s argument, I buy his overall contention that just because a machine can follow commands (a rulebook), that does not mean it has genuinely learned anything.  There is a reason why a computer can never pass the Turing Test: it cannot come up with original questions or responses, but instead reproduces formulated phrases based on key words that is has been programmed to recognize.  In the case of the machine performing as well on the SAT analogy test as the average high school senior, we have to point to cases of seniors who did significantly better than average–or even got all of them correct–and conclude that while there is a lot of variability in human intelligence, it certainly exists, while the machine will perform equally poorly (56% is still failing in my book) every time.  One might argue that these tests are overly anthropomorphic and ignore the kind of intelligence that dogs and dolphins possess, but even these creatures demonstrate originality in the way they communicate and react in novel situations.  Until we can artificially create neurons, we will not be able to reproduce this kind of fresh thought and the behavior it engenders.

Thus, based on what I have gathered from this course, I still do not believe that we will achieve actual artificial intelligence, though even more advanced machine learning seems to me a viable goal.  We can continue to refine our computers’ current abilities to recognize images and speech, and to perform intricate procedures (like surgery), but these are all programmable tasks.  I do not think there is enough technology in the world to produce a computer that can generate new ideas, that can think up the kind of technology it would take to build other computers like itself.  To me, cognition, not computation, is still what defines intelligence, and that is not something you can fake.

Before this course, artificial intelligence sounded to me like something straight out of a science fiction movie. To me, it seemed like robots straight out of Star Wars or the Steven Spielberg movie with the robot that looked just like a real boy. How could a computer or robot possibly “think”? Computers and people are so incredibly different, I thought the two can never be compared in this manner. In that sense, I thought of AI as something absurd or wildly fantastical - not so far off from Searle’s argument.

However, this course has really opened my eyes to what computers can do and are capable of. Slowly I realized that AI is something very real. With machine learning, a form of AI, I realize computers are capable of “learning” a great deal. While they may not have a brain, they still learn how to function and perform certain computations, even sometimes as well as human can, as evinced by the SAT analogies. The simulation argument provides strong reasoning for the existence of AI. Even with a robot like the Scribbler, I see the possibility of AI becoming more of a reality. On that note, I don’t quite agree with Searle’s argument on why strong AI is impossible. Searle seems to make a point that his argument is quite obvious.

He argues that thinking is inextricably tied to the brain. On one level, yes, computers will never be humans, nor will they have a brain. But they are capable of computing and “learning” operations almost as well as we do. In that sense, then, there is some notion of AI. And unlike Searle, I don’t think we can see it as simply “weak AI”. I have realized now that there is much more to it than that. His argument is not quite as “obvious” as he states.

               When I first started taking this class, I had a very skewed view of Artificial Intelligence. The first was colored by science fiction. I imagined something like Hailey Joe Osmond in Stephen Spielberg’s AI: a humanlike robot that could not only think but feel. Or along similar lines, maybe something like the robot from I, Robot that again was primarily characterized not by its exceptional intelligence but by its capacity to feel human emotion. Needless to say, when I entered the class I didn’t really think we had achieved anything that could be called Artificial Intelligence because I was quite certain that science had not reached the point of endowing an autonomous machine with the capacity to feel.

               After having taken this class, I feel that my conception of Artificial Intelligence may not be giving quite enough credit to how far we have come technologically in the mimicking of human thought. I also believe that I may have initially defined intelligence too broadly by encompassing emotion, which some people would argue is a feature of the human mind that extends well beyond what the layman would classify as intelligence. Or by another token, perhaps it is emotion that is the less complex mental process. In any case, I now feel that I was mistaken in discounting the existence of Artificial Intelligence based on the fact that robots can’t feel, when machines are already capable of so many complex computational processes. At the very least, it is a question that requires some thought and reflection about the very nature of intelligence and its uniqueness to living creatures or lack thereof.

               Now that I have read the Searle reading, I realize that there are a lot more issues to consider when I think about whether or not we have achieved artificial intelligence, and whether or not we will. The idea of intelligence is a very complex concept. While I don’t disagree that intelligence is more that the ability to perform computational processes, I also remember a time when I had taken it for granted what intelligence was and that some things didn’t have it. For instance, I implicitly assumed that humans and that dogs were intelligence, but I didn’t think much of the intelligence of a fish. Now, my gut instinct is to call a fish intelligent in comparison to a computer, but not only do I not know much about the thoughts of fish, I also don’t know what subconscious distinction I am drawing. I also know that in general, when I watch science fiction films that involve talking supercomputers, except in the cases where it was explicitly stated that technology had been able to create artificial intelligence, I never regarded even the most sophisticated of thinking technology as true Artificial Intelligence. All of this is to say that, as a sociologist, this class has made me realize that my conceptualization of Artificial Intelligence is socially constructed.

Before this course, artificial intelligence was a very sci-fi concept for me. Artificial intelligence was something that I thought was rightfully placed in Star Trek and Star Wars - with Data and C3PO, respectively. Artificial intelligence always seems mimick human form in these sci-fi popular television and movie representations, and we can see that both Data and C3PO are machines fashioned in human form. Likewise, the Terminator series was another very interesting portrayal of different forms of artificial intelligence - from the original terminator to the liquid-metal replacement. But this sci-fi articial intelligence always had little to do with any actual manifestation of artificial intelligence, which I always imagined to be much less elaborate.

In this course, we have seen that artificial intelligence is not a simple concept. Having read John Searle’s article Is the Brain’s Mind a Computer Program, we can see that he argues that there is more than one type of artificial intelligence: strong Artificial Intelligence and weak Artificial Intelligence - the existence of the former is in question, while the latter most certainly exist. A Strong AI is one which can equal or surpass human intelligence, while a weak one simply has the capacity to act intelligently.

Searle was very convincing with his Chinese Room demonstration: a computer could learn the language using a rulebook that guides it without actually understanding chinese. In this way, the AI is weak - it is acting intelligently. However, it would be hard to imagine a computer actually learning Chinese, which would be necessary for a strong AI.

Nevertheless, I think it is still possible to create a Data or Terminator like being - perhaps not within the current available technological framework - but it seems that we have made great leaps in artificial intelligence in the last 50 years. It is not unreasonable to assume that we will make further leaps in the future.

Before taking this course, I thought of artificial intelligence as a hypothetical concept mostly reserved to sci-fi movies and books and very few scientists making theoretical claims about the concept. I thought that the possibility of artificial intelligence was as remote as the possibility of humans traveling at the speed of light. Therefore, I thought that the problem of artificial intelligence was not one that is seriously studied in computer science or other disciplines. I thought that AI consisted simply in physically imitating the human brain, which, given the current state of technology, seems like a task that lies beyond the realm of reality.

However, after taking COS 116 and especially after the second to last lecture reading Searle’s article, I began to think of AI in a completely different way. I realized that a key aspect of AI was defining what “intelligence” actually is, and whether any other creatures besides humans actually have intelligence. Therefore, I learned that a big constraint on creating AI is one of definition of terms and discovery of what actually constitutes intelligence and that the problem is not simply one that is subject to physical limitations. I also learned that the Turing test, which I had always thought of as the definitive method of testing whether artificial intelligence had been attained, is an imperfect test. The problems with the test are that it focuses only on human intelligence and that it tests the presence of intelligence from a behaviorist framework. Searle’s article also got me thinking about the physical possibility of actually building a computer that can give all the “right” responses to pass the test. In class, I was surprised to find out, that the complexity of human language actually makes hard-coding all the right responses physically impossible given the constraints of our universe.

In the end, the ultimate question that the class left me with but did not give me an answer to was: Is there something about human behavior and human physical characteristics that lends human characteristic inimitable by computers? In other words, what are the constraints on and limitations of computers? Whether or not those questions can ever even be answered, remains to be seen.

Before this course began, I understood next to nothing about artificial intelligence. I have now come to understand that artificial intelligence is bordering near-human intelligence, and it appears as if “the future” is actually happening. Through my experiences with Scribbler earlier on in this course, I have a newfound appreciation for those contributing to the field of AI, and the difficultly of this daunting science. To even program a robot to go through a simple maze (albeit, it was Scribbler) involved a lot of tedious work, so I can only imagine what it would take to program a robot to behave as a human would.
As the field of artificial intelligence is more and more gearing towards designing a robot that is nearly identical to a human (in regards to how it interacts with others and the its environment) I have no doubt that it will not be long before computers can routinely pass a Turing test and we will judge them as having human intelligence. I am curious as to what the effects on society would be if we ever design such a robot. Will our society’s social skills diminish if we all have “perfect” robot friends? Will we become too dependent on these machines to even the simplest of tasks, like getting a drink from the fridge, or setting the table? I am not advocating that we halt any sort of progress we are making in the field of artificial intelligence, but I do believe that in the future, utilizing these machines for their efficiency and speed should be the goal, and not depending on them for personal relationships and menial tasks which humans can easily do.

Artificial Intelligence is very hard to define specially strong AI. This is because strong AI implies that machines can have an understanding of what they are doing to the same level that humans have. In other words, it means that machines can not only use sintax but also semantics for computational processes. There are two problems with this view the first one is what  precisely does semantics mean.  If semantics means being counsious of what we are doing and understanding every step of a process, can we claim that semantics actually exit in our minds. For example, we we talk in our native language do we really understand what is going on and we are conscious of the whole process, or do we simple use the “chinease book” open our mouth and let the brain translate ideas into words automatically? I think that is more of the second option. Semantics is not a clear definition. It is more than an abstract concept given to something that is not even clear to us.

However,  the fact that is not clear does not mean that it does not exist because in fact Searle has a point in saying that even if we have a very powerful computer that could simulate every synapsis in the brain, could say that the simulation of the brain process is the brain itself? Searle says no. The reason he gives is that a simulations are just representations of a reality but not the reality itself. I agree with him which implies that there must be something more than simple sintax. The question is then what is this concept that makes a model differ from its real object? Strong AI needs to solve this before being able to say that stong AI is possible.

As a believer of innovation, one thinks that one day computer are going to talk, interact and pass the Turing  test as humans do. Yet, many quesitons need to be answer first. Is the Turing test enough for proving that strong AI exists? Does semantics exist?  Also, even if computers are like humans, there will be many people that will doubt that strong AI does exist. This is because as humans we tend to be very human center. I guess that  this debate is going to be  very difficult to settle.

I really didn’t know anything about AI before taking this course; for me, it was something that existed solely in the world of sci-fi and perhaps computer scientist’s wildest dreams.  I knew that it must exist to some degree, that there were robots and computers that could simulate human actions and possibly even “outsmart” humans.  The existence of such “weak AI” doesn’t seem to be contested.  Searle, however, thinks that strong AI, the kind I’ve seen in movies, does not exist.  His Chinese room example was certainly compelling; it seems reasonable to say that computers aren’t intelligent if they’re simply following a manual written by an intelligent human.  Before this class, I would have agreed that strong AI does not and will not exist for a while.  I have been surprised, however, by the fact that human intelligence is perhaps not as unique and incapable of being duplicated as previously thought.  Perhaps the AI that is portrayed in movies is nearer than most of us would believe.

However, one thing this course has shown me is that intelligence needs to be clearly defined before artificial intelligence’s strength/success can be gauged.  The word intelligence is used to describe such a multitude of exceptional skills, some of which I sincerely doubt computers could have.

I never thought much about AI before this class, so I definitely know more than I did 12 weeks ago.  I think I agree with Searle that strong AI doesn’t exist now, and I don’t think computers will ever be able to really act human, but this class has made me much more hesitant to say it will never exist.

Before I took this course, I didn’t know that much about AI, other than that lots of computer scientists were working on it. I probably assumed that one day, with enough work, they would be able to create a computer with a human-like mind — in other words, that “strong AI” was a feasible goal. But reading Searle’s article really changed my opinion. He raised the issue of “causal powers” of a mind (see Axiom 4: Brains cause minds, and Conclusion 2), in other words, any system which successfully recreates a mind will have to have equivalent causal powers to our brains. This made me want to read more about “consciousness” and how scientists think it arises.

The second thing that our course discussion made me think a lot about was the importance of precise definitions. (As we have seen since the beginning of the course, the English language is imprecise, and we often have to clarify exactly what we mean.) So we can’t just say “we want a computer to be like a human”; instead, we have to define in great detail which traits and skills we consider essential. We talked about things like a sense of humor or self-awareness, and whether these might be necessary or not. Also, the question of whether a computer must just “look” like it is carrying out all these functions (the Turing test) or actually be doing them (Searle’s argument) is a really interesting one. I think earlier I would have agreed with Turing, but after reading Searle’s article he has me pretty convinced; although I’m still not sure!