Readings
Diana Applegate
There isn't much to respond to this week...we continued with propositional calc on Tuesday and then had our exam on Thursday. I haven't had any discrete math or logic classes before, but the ideas behind propositional calc seem pretty straightforward. However, since I am new to it, I'm having a bit of a problem applying propositional calc (i.e. doing problems like the ones on the test). The only way that I can grasp anything remotely mathematical is to do lots of step-by-step written-out examples. I'm not saying that our class should do such a thing during lecture or for homework just because of my mathematical shortcomings...but I freeze up on tests when I have to do problems that I haven't explicitly done before. I assume that there's gotta be at least one of two other people who feel this way...
As you might have guessed, I'm also struggling with the Konane project. Although I find AI to be very interesting, I regret taking the course with only CS110 under my belt. I did extremely well in 110 but since then it seems that I've lost all of my programming abilities. I took 110 last semester with visiting professor Clare Congdon, and our assignments were, in my opinion, a lot different than this semester'sCS110 programs. We did a lot of work with graphics and stuff and not a whole lot with designing programs to solve real problems. I am especially frustrated since I was hoping to be a CS major. Now that seems like an impossibility to me.
I realize that it's much too late for this, but are tutors available for CS as they are for math? Also, what is the role of the TA (Sarah) in this course? Are we allowed to approach her with minor questions that we may have about programming, etc? Does she hold any lab hours for this course?
Jocelyn Arcari
This week's class focused primarily on propositional logic. We were given the "formula" for converting wffs to a clause form and that is used in resolution but I do not quite understand why/when we use resolution. I think we were going to discuss it again in class on Tuesday, but I wanted to be sure to mention it. Also, is there a longer example of "reductio ad absurdum" we could look at? I have done proofs like this in math time and again, but I'd like to see a longer one using just wffs, connectives, etc. In working on my paper, I am finding that my understanding of the different search mechanisms we discussed in class, in particular minimax and the alpha-beta algorithm is very helpful when reading about the different chess programs for computers. I have not yet found literature that pinpoints the style of search that Deep Blue uses during its games but will continue to look for that. All in all, I'm learning a lot and am enjoying the opportunity to find out more about this area of AI. People may find it interesting that IBM's web site for Deep Blue states in its FAQ's that Deep Blue does not use artificial intelligence. IBM says it is neither trying to mimic human thinking or intuition nor would that be possible since we do not understand it very well to begin with. Deep blue is able to organize a Huge database of "knowledge" from previous games. IBM states, "Kasparov isn't playing a computer, he's playing the ghosts of [chess] grandmasters past."
David Costello
Chapter 14 from Nilsson seemed pretty clear to me. At first I did not understand how to convert arbitrary wffs into conjunction of clauses but after studying section 14.2 for the test, I can now comprehend the concept. One question I do have concerns the last exam. I though I knew how to use the alpha-beta algorithm correctly but I still had trouble with the tree on the test. I was wondering does the evaluation process start at the leaf nodes (where the static evaluations are) or does it begin at the start of the tree? Besides that question I am ok with the rest of the test.
Benjamin Flynn
Hello. One thing I've found with this course is that it relates to almost everything else I've learned at Haverford. I suppose this shouldn't be surprising as the subject of the course is intelligence, be it artificial or not. This came up recently when I was meeting with my thesis advisor and talking about how the arangement of data seemed to have a lot in common with some of the philosophical concepts I have been learning about in other courses this year, and concpets I'd learned previously in english courses. He was telling me that many of the algorithms used in compiler design and memory allocation developed from artificial intelligence studies. The more abstract concept of what thought is has perviated through most of the humanities courses I've taken, as it is a question with widespread social and philosophical significance. At the same time, the nuts and bolts application of algorithms, search and logical, that we have been learning about recently, I have found in almost all the math and science courses I've taken. When I first read the course description, I thought the tie between it and philosophy would be weak, at best, but I've come to see that, more and more, it is very much the child and partner of many philosophical insights, as well as a composition of logical and mathematical technique.
Greenfest Emily
I'm afraid I have not much to say about the past few weeks. I have found the lecture material to be interesting: I've always been intrigued by formalized logics/mathematics, although I had always studied it safe in the assumption that the old logic I learned in Geometry was LOGIC. So I have a little question: are the logics we are studying now spinoffs from AI or are they older mathematical derivatives? What are the main differences between that logic and the Propositional one, and if they are not the same, why was the "mathematical" logic not used for AI purposes?
Hakim Ayishih
Alpha Beta Pruning seems like a very ingenious algorithm however after taking last weeks test Im convinced that I do not possess a full understanding of its ability. I understand the concept and the general underlying rules that govern the pruning process but fro some reason it was very difficult for me to demonstrate this understanding. Maybe we can talk about this sometime during office hours.
Another thing that perplexed me during the exam was the problem where you asked us to draw a search tree from information that you gave us. I hope that you can go over this one in class. I know it will give way to an easy explanation. What confused me the most about this problem is the fact that you gave us the nodes in a particular order. But I thought the order that nodes appear is always determined by a search procedure. However the last part of the question you asked us to write the order the nodes would be searched if we used various search procedures.
WhenI am confused about something it is terribly hard for me to articulate it in a understandable way but I hope you have grasped the jest of my concerns.
Ada AC Hogan
I had a couple questions from chap. 14:
Subsumes? "a clause, y1, subsumes a clause y2, if the literals in y1 are a subset of those in y2". p or r subsumes both p -or- r -or- q and p -or -r -or- s (233). How can it be that one OR the other subsumes the clause, and what does that mean?
How is it that the empty clause is a contradiction? (I'm thinking about resolution refutation)
About the block lifting example: why isn't 'liftable' also included in step one with 'bat_ok'? If step three is the combination of steps one and two, shouldn't it (liftable) be in step one?
I'm also not clear on the "linear input strategy"; when is one of the clauses being resloved not one of the members of the original set of clauses?
I've found that through my research of the use of AI in the Mars Pathfinder, I've come across a new project that seems to have even more uses of AI- NASA's new millennium remote agent; I'd like to redirect the focus of my paper to autonomous systems- specifically to the Deep Space One project. On the internet I found the NASA site, which includes "Model-based Autonomous Systems Research Area", projects autonomous remote agent, autonomous navigation, facts on the new millennium project. Since I've already researched the Pathfinder project, it would be interesting to see how DS-1 has evolved from this earlier project, while keeping some of the same goals while approaching problems that the Pathfinder encountered.
I think I have enough information, but there are a couple papers I'd like to see that I couldn't get off net- they are
A Remote Agent Prototype for Spacecraft Autonomy (appears in Proceedings of the SPIE Conference on Optical Science, Engineering, and Instrumentation);
Design of the Remote Agent Experiment for Spacecraft Autonomy (appears in the Proceedings of the IEEE Conference on Aerospace 1998);
Immobile Robots: AI in the New Millennium ( In AI Magazine, Fall 1996)
would you happen to have back issues of the AI magazine on hand?
Peter Ingebretson
I would have been interested to learn about Horn clauses in class; it seemed odd that a clause with an empty implication could make any sense. Otherwise, the chapter seemed pretty interesting, but straightforward. I like the idea of resolution though, it seems to generalize the other rules of inference in a very clean manner.
Project one is interesting too, although I have a somewhat difficult time gaguing how well my program actually is playing. It can beat me, and it can beat itself when searching to a lower level, but this doesn't tell me much, since I don't really know how to play well myself. I suppose we must have faith in the algorithm to some degree, and in the tests we put the algorithm through.
Jim Speer
Not much concrete to report this week. I have been thinking again about the question of machine intelligence (surprise, surprise) but this time particularly with regards to creativity. If we have trouble defining intelligence then we'd surely have touble defining creativity. I was in a discussion with someone recently who told me that artificial creativity is theoretically the ability of a larger algorithm to produce a smaller algorithm. My question is: what standard is a machine capable of applying to to its own product? If a machine comes up with an algorithm, then does it not have to evaluate its work to truly be called creative? A game playing program seeks solutions through search and by means of its capacity for quickly seeing ahead and quantifying future situations. If we were to ask a machine to write a play, I imagine it might start with whatever it knows about drama, and assemble tiny bits of it in different combinations,at super speeds, until something acceptable passed its static evaluation. I do not think that is creativity. An infinite number of monkeys at typewriters have the same advantage as a creative or problem solving machine. Sure, the monkeys will eventually produce the script for Hamlet, but they'll also produce an incredible amount of really really bad stuff. It is not remarkable that something worthwhile is occasionally produced by a system that works tirelessly and at super speeds. What would be remarkable is if the monkeys were able to figure out which of their works were any good, and which of their awful plays should be discarded. Similarly, Deep Blue's victory over Gary Kasparov would be a lot more impressive to me if the computer had devised its own assesment of the game it is playing, instead of being told by its programmers what the algorithm for the static evaluation of the game board is. I would say of course , a machine will never be able to develop a talent in an area completely outside of its original programming -- but on the other hand, our light seeking robots' wiggling-forward behavior was an emergent behavior which was not specifically programmed. But finally again, I must conclude that the robot has no internal standard for evaluating its emergent behavior -- The forward wiggling is merely a side effect.
Ben Sprecher
Since there was no reading assigned for today, I wanted to talk about the AI talk Tim and I went to on Friday. We went into Drexel to hear a robotics professor from Carnegie Mellon talk about her soccer-playing robots. She explaine dwhy she had become interested in the problem - she felt that soccer playing involved the fundamental issues associated with intelligence - the ability to work both autonomously and together as a team, to deal with a dynamic and uncertain environment, to think and act in real-time and carry out actions which are non-deterministic. Uncertainty about the world, learning, communicating, and problem solving, she held, are characteristic of truly intelligent behavior.
After giving us some background on why she did it, she told us how she designed the robotic programs. She described for us the algorithms used to determine trajectory of the ball and the optimal intercept course, to avoid other players and differentiate them from the ball, to decide whether (and where) to pass the ball or whether to shoot it, and to determine optimal field position for all of the robots.
She also showed us some videos from the event and explained how her robots were able to beat all the other teams. It was very interesting, and I especially enjoyed seeing the little guys in action in her videos.
Emily Sweeney-Samuelson
Resolution/resolution refutation seems fairly straightforward, and very handy. I hope we will do a lot with propositional calculus; it's interesting. I read in someone's response from last week that this comes up in linguistics. I'm glad, because that's a subject I plan to explore during college.
The search strategies section was a little more complicated. It reminds me of the missionaries and cannibals problem, which I'm not too fond of, in that a computer program to solve either type of problem would work best with a very specific order of possibilities to try (generated by a specific search algorithm, which performs the same way every time), whereas humans would work out these problems a little more randomly. It is the same way with many programs, of course, but it struck me that with both the missionaries and cannibals problem and with resolution refutation, the best ordering strategy is not at all obvious (to me, anyway) on the first examination of the problem. The conventional search methods used in those programs are not intuitive to humans. But I wonder if there is a certain way that humans tend to "search" for possible solutions, or steps toward solutions, of any problem? That would be interesting to know, and try to apply to AI.
Leslie Zavisca
As we had only one hour of lecture and just a few pages of reading, I don't have anything profound to say about last week. I like studying propositional calculus because it is extremely straightforward (although I can't say the same about Nilsson) and the rules and definitions are pretty clear cut. I also like the fact that I'm already seeing some payoff for having to suffer through discrete math even though I'm still in the middle of the course. I am looking forward to continuing our discussion on resolution refutation as it seems to be the most direct application that I've seen so far of the kinds of proofs we are doing in discrete math. One thing that I'm still not clear on is the breakdown of exactly why resolution refutation is supposed to be much better for proving things than truth tables. Also, after our last lecture, I finally understand the definitions of soundness and completeness. And although I think I have a firm grasp on propositional logic, I would like to see some examples of programs with specific applications of wffs, rules of inference, resolution, etc. In other words, what exactly is their purpose?