Readings
I'm very happy overall that I took CS372. As I look back, there were parts of it that were incredibly frustrating, but also parts that were a lot of fun. Since we covered so many different topics, I guess all of us had our favorites. Interestingly enough, I think I've enjoyed the very last section, on computational linguistics, the most. The drawback to covering such a broad amount of material was that we couldn't spend too much time on each area...and sometimes I felt like I simply needed more time to absorb the material. However, I'm glad that we took the "overview" approach because now I really feel like I have a good idea of what's going on in AI. In addition, looking into the various realms of AI helped me in terms of considering a career/area of concentration. If I could take the course again, however, I'd like to do so having discrete math and data structures under my belt. If that had been the case, I feel like I would have gotten more out of the course. Nevertheless, I still feel like I've learned a lot and that the experience has been a beneficial one. I was fortunate enough to have taken Philosophy 244, or Philosophy of Cognitive Science concurrently with CS372. There were a number of cases when these classes overlapped, which was quite interesting since I got to see "both sides" (philosophical vs. scientific) of the story at hand. In addition, CS372 has given me a lot of ideas about research projects that I may want to undertake in the future. And I've really enjoyed writing (and reading) these responses...it was great to hear what all of you had to say. Not only did I learn a great deal from Deepak and the lectures,etc., I also was able to learn a whole lot from the other students. Such interaction doesn't usually occur in most of my other classes, and the fact that it did occur here was wonderful. Good luck to you all!
P.S. - I saw a three-second clip of the Penn State Robot Competition on News10 this weekend...it looked like fun! Did anyone end up competing or going to watch?
http://mainline.brynmawr.edu/~dapplega/marsAI.html
David Costello
It is very hard to come up with questions concerning last week's lecture. The transition networks you described seemed pretty straight forward. One question I do have concerns the limits of this method. What are the factors which stop an extremely sophisticated version of what you presented in class from being implemented? In other words, since the method does succeed in teaching the computer how to reason in a limited environment, what stops us from just expanding the environment to fit our own comprehension?
Sonia Dubielzig
as the course ends, I'm still wondering, "what is intelligence?" Can it be that these computers can truly learn, think, and act? What are they thinking, and can it truly be called "thinking"? Do we mean a specific kind of thinking? Are computers conscious of themselves?
we started the course discussing the Turing Test, which AI programmers solved by manipulating common language structures and using rules so that computers could have a realistic conversation; we're ending it with a discussion of how computers might understand complex language. Most folks don't accept the Turing test as a test of intelligence anymore, after the rapid writing of programs that seemed to solve it. However, would programs that can parse and analyze language and answer questions given to it in the same way we learned about last week be "more" intelligent? Intuitively, they seem to be. What is the line to draw?
Ben Flynn
This is more of a response to the project I have been working on, than to class or CFG's. My final project was to implement a system that would teach itself how to play tic-tac-toe using neural nets. Many of my friends who had watched me battling with Konane wondered how what I was trying to do with Tic-Tac-Toe could be any harder than what I did with the checkers game. I think there are several noteworthy points being made here.
The first is that radically different technologies can end up looking about the same when completed. It is necessary to look below the surface of the technology, in such simplified cases, to see that what is going on is radically different than what we expected.
The second is that a proper medium must be found in which to judge intelligence. A human playing tic-tac-toe could easily be seen as less intelligent than a computer playing checkers if all we look at is the results of their games. Intelligence cannot be located until it is asked to adapt, and either succeeds or fails.
Ayishih Hakim
I am very interested in the natural language parsing that we are studying now. I never thought that it was so simple to decode the english language. I understand that it is not a simple thing to produce a lexigon with all possible words and compound words. However I never really thought that English grammer was so systematic (i.e. if the current word is a noun then the next word must be a verb) . I know people are wondering if I skipped grammer school, infact Im wondering the same thing. I just never sat down and thought about it. I know that parsing is just a small part of natural language that determines whether or not the sentence makes sence but it is still intriguing.
My general comments on the class are:
I really enjoyed the experience that you have brought to the class. I really feel as if I am being taught by someone who cares and knows alot about the subject. The extra blurbs about what is going on in the current world of technology really helped me feel the relevancy of the concepts that we were learning.
I also very much enjoyed the robot labs and the hands on experience with robots. If you remember one of my first responses, I had alot of anxiety concerning the building and programming of robots. Now I am glad that you allowed up to jump right in and begin our search for immergent behavior and intelligence in our robots.
The video aids were also very good, especially the one from "back in the day". I liked the look in the past and the challenge of examining past predictions and concepts of AI.
Overall I have gained a new understanding of the capability of computers. I am not completely convinced that computers can demonstrate intelligence. However I am begining to doubt my conception of intelligence and I am beginning to question how exactly humans learn information.
I think that this class tackled many important issues and I have surely taken many things from this class.
Ada AC Hogan
In parsing, when is it more useful to use top down or bottom up methods? In the augmented transition networks, how long are values stored in the registers? I was also wondering what happens exactly when, for example,the verb is going from S1 to S2, and it doesn't agree in number- if the process stops there, how is it communicated that the number is not in agreement; how does it know not to "pop"? Is it through the register?
I thought that the sample lexicon and grammar that we looked at in class were really helpful for understanding the transition network.
Last response.... when I came into the class in september I knew NOTHING about AI, and relatively little about computers. It's not often possible to take a class that is completely out of your field, and I feel that this was a opportunity to do so. Now I'll understand what is really up with those new pet robot dogs... Even as I was doing the Cyc project, I began to appreciate some of the difficulties of large knowledge systems, that is how the system could come up with such strange answers for my queries. It has given me an idea of just how prevalent AI systems are.
Sarah Klaum
In my response last week I questioned whether there was work that had been done involving AI with how we learn a first language. Thinking over this again I consided how a language grows and evolves, and i now wonder what could be done with a context free grammar and something like genetic programming. Has anything remotely along these lines been done, is there any insight or knowledge that could be learned from such a tactic? Now I realize that perhaps it would not reflect much on how a language trully evolves since there are so many variables in the evolution, and GP is used to derive an optimal and efficient program, not to model anything similar to the evolution I'm interested in (as far as i know). I am also interested in something that was mentioned in class this past week about intention, or implied meaning in a sentense such as "it's cold in here (turn on the heat)." I find it engaging to think about something like a grammar of intentionality/implication, or how such a concept could be implemented in a natural language understanding system and exactly what functionality it would add.
As for my overall feelings about the course, I'm afraid what I have to offer comes across as a rather general statment. I do feel, however, that the material we have covered has strengthened my initial interest in the broad field of Artificial Intelligence, and I am happy to have a better understanding of the many facets of the dicipline. I am still excited by the prospect of what the field has to offer and what remains to be studied and applied. I think that despite what many see as a history of failure it is best to approach the study of AI with optimism, because there is no doubt that advances will continue to be made as we gain greater insight into human and machine intelligence.
Edina Sarajlic
I have found this weeks reading on natural language parsers very interesting, but somewhat hard to follow towards the end. I think that the amount of detail Allen used exceeds the limit set by the time we have for actually covering it. Even though our exam is an open book one, I hope it will not require of us to reproduce all of the described parsing methods.
Also, I am looking forward to learning about other peoples final projects in lab tomorrow. I must say I am not as excited about presenting my own. I have learned a lot, but I feel that the project did not turn out quite the way I planned it. (Having better programming skills would have helped , but all things considered, it still worked out relatively well.)
Looking back at the beginning of the course, I feel that I have learned a lot this semester. In the pursuit of the answers to some fundamental AI questions (for example, defining intelligence or thinking ), we have covered many different AI topics. The beginning was somewhat frustrating, because I expected the class to provide clear answers to my questions However, along with giving answers, the class provoked more and more new questions. Since the field is so broad and diverse, it was not possible to cover all the topics in a great detail . Even though I still so not know all the answers, I feel that I have gained a deeper appreciation and understanding for the subject of AI.
The lab was a very important part of the overall learning experience. Implementing in practice the AI concepts learned in class clarified many unknowns. Lab offered a glimpse of the issues present in the practical side of AI, from the engineering questions in the lego robot design to the design of software which models some aspect of intelligent behavior. In any event, the occasional headaches caused by the moodiness of our HandyBoards were more than rewarded by the satisfaction from seeing Leia find her way out of some tricky corner. Without some intrinsic intelligence of its own, Leia was a great help in gaining a better understanding our own intelligence.
Jim Speer
I have enjoyed the discussions about logic, wwffs and natural language. It is areas like these, when a computer attempts to decipher and handle human tools, in which the contrast between human thinking and artificial thinking is the sharpest. The computer must have its verbal input formatted strictly and orderly if it is to be able to process it for understanding. On the other hand, once processed, the computer has the advantage of being able to make numerous logical derivations at great speeds. For humans, ambiguously phrased input can often succeed in being understood precisely, but logical derivations and implications may not be as quick to follow.
For me, this pretty much sums up my impression of AI. The mechanical intelligence must be spoon fed everything, and all of it's subsequent successes are achieved through speed. Even if someday computer input is allowed to be entered in a more haphazard and ambiguous fashion, the computer must still crunch every number to reach a conclusion. A human intelligence often reaches its conclusion without having had to calculate every option.
This difference was quite apparent during the konane tournament. Human players cannot be expected to calculate every possible move to any depth beyond 1. The computer, which would foresee and evaluate thousands of outcomes per move, must consider even the ridiculous in order to come up with its preferred action.
Emily Sweeney-Samuelson
I enjoyed the level of detail in the handouts this week. They provide a good mixture of theory and application of concepts. I wish we had more time to talk about them in class and go over the concepts, because I want to learn them thoroughly, but it feels like we just touched on them. And I had trouble keeping track of some of the more involved examples in writing, but I know they are much easier to follow in class. It helped to go over resolution in predicate calculus so much, and I know if we had that much time to devote to this material, we would learn it well. As it is, I'm very glad we did as much as we did with it, because it's fascinating, but I find myself hoping it isn't on the exam, because of the shortage of classtime spent on it. It hasn't sunk in, we haven't had time for many questions after reading both chapters, etc.
Overall, this semester has introduced a new discipline, or at least a new aspect of computer science, to me. I'm glad I took the class; I only wish it was a year long! There just wasn't enough time to look at everything in enough depth, in my opinion. I feel like the things I was most interested in, we didn't spend a great deal of time on, but I got a great, broad introduction to AI, which is what I was hoping for. Konane just consumed me for a while there, which wasn't the most positive experience. Game playing is not one of the things I wanted to work on implementing, but I think it was the best choice for a project to assign to the whole class.
I really enjoyed the part of Mind Children that we read. I wish we had finished it -- the responses to that would have made for good discussions, I think. But again, there really wasn't time, and now we have the book, so we can finish it on our own if we want, so I'm content. I hope the rest of it is as thought-provoking as the first few chapters were. I liked Nilsson's text, and I thought the course was structured well to incorporate it.
The robot lab gave the course added personality, and made it fun and memorable. I'm glad we got the opportunity to try out some of the concepts we were discussing in class! It was great to see the different robots in action, get to know other people in the class, and have Fun With Legos. I wish that, too, could have extended throughout the whole semester, but the concepts became too complicated to implement on the little guys. I would have liked to continue the robot stuff, and culminate to a robot soccer tournament instead of a Konane tournament, but maybe that's too much of the same concepts. Not broad enough, or ambitious enough, I guess. I think it's the childish side of me, wanting to play with the cute robots. Oh, well.
This class was a great opportunity to learn about an exciting, developing field. I wonder how much progress will have been made in four years, for instance if I take an AI class in graduate school? There are so many things happening in the field -- but I feel like I have learned a little about many of those things, some fundamental concepts about each, that (hopefully) won't change, and some history, which helped clarify some things for me, besides being very interesting.
Sarah Waziruddin
This weeks lectures were a bit confusing but the reading has put things in perspective. To know how to compute the syntactic structure of a sentence, grammar and parsing technique need to be considered. Parsing technique is the method of analyzing a sentence to determine its structure according to the grammar (p. 41, Chp. 3 of handout)
There are two basic ways to represent grammars. A context free method and using recursive transition networks. To use context-free grammars, you must know what types of structures that are legitimate in English. The two common forms of parsing associated with this method are top down and bottom up parsing. Recursive transition networks are another way to represent grammar. The method is more visual than context free grammars. Again, both top down and bottom up parsing can be used with this method of representing a grammar.
To produce an analysis of a sentence, we need to extend recursive transition networks and context free grammars.
I am unsure about the structure and content of the next exam. I know the exam covers topics in the last third of class but what will be the structure of the questions? Will they be similar to the examples that we did in class?
AI was an interesting experience for me. Like Jocelyn, I came into the class with a firm belief that computers could not think for themselves. What we learnt at the very beginning of class is the most relevant when discussing this topic-- the question of the definition of intelligence. The answer to the question of whether or not a machine can think is entirely dependent on your definition of machine and think. I still do not think that machines can think for themselves but intelligent behavior is a result of programming-- both a direct and indirect result.
I really liked the inter-disciplinary aspects of this course. Even though I will probably never take a linguistics course, I have had a brief introduction to that field.
Anyway, this class has been an interesting experience for me and I have learnt many things that are now at the forefront of computer science.
Leslie Zavisca
I am going to treat this week's as a general response to the overall course. I was just looking through my class notes to refresh myself on all of the material that we have been introduced to throughout the semester. I can't believe that yesterday we were discussing English sentence structure and only a few months ago we were just introducing the question, "Can machines think?" Since that is the central AI question, I have pondered again and again over the semester, and I have to say that, right now at the conclusion of this course, I'm not convinced that machines will ever be able to truly think for themselves. I think that, to a certain extent, we are underestimating our own minds (or perhaps some outside force). If we scarcely understand our brains, how can we say that we are approximating the same processes in robots? I am, of course, open to and interested in opposing arguments.
As for the specific topics we covered, I have to agree that a two semester course allowing more depth in study would have been preferable. My favorite topics were neural nets and searches. I definitely felt that we covered some material rather quickly though--not that I didn't understand it, it's just that I would have liked more time for exploration. I often found myself asking questions that there just wasn't time to answer. However, I'm not sure what subjects I would have sacrificed so that others could have been studied in more depth because I am satisfied with the knowledge base that I gained from this class. Something thing that I found very valuable was the lab component of the course. Although some weeks' assignments took an insane amount of time to complete, I generally enjoyed the subject matter and being able to present the final product. I did have the advantage of getting along pretty well with my group though so I know that some other students might have hated the lab if they were stuck with incompatible group members. Also, everyone in my group had some programming experience and I'm not sure if things would have run as smoothly if someone hadn't had at least 110. I think that the programming aspect would have been pretty hard to catch onto as well. Oh yeah, Nilsson wasn't so bad either except his writing did tend to get a bit dry sometimes, but it was bearable.