This chapter starts out by laying
out a fear that is growing more common within our society: the fear that
machines will develop a will of their own. The biggest threat of this is a
learning machine. The most common application for this type of machine is in
the military, being used to develop strategies based on the opponent’s movement. However,
the main learning machines that are talked about in this chapter are those that
play simple games. There are three main advantages a computer has
over humans. The first is that it never makes a careless mistake. The second is
that a computer can look more moves that an opponent can make in the future
much faster than a person. Lastly, there is no limit to a computer’s learning
capabilities. The main points of this chapter are the descriptions of two types
of Matchbox learning machines, and how their learning compares to a human’s
ability to play their game.
The chapter
tells of a reader-friendly way to experiment with “game-learning machines” as
Gardner calls them. Donald Michie developed a tic-tac-toe learning machine he
called MENACE, which stands for Matchbox Educable Naughts And Crosses Engine.
As the name suggests it is comprised of three
hundred matchboxes with wooden V’s attached to the bottoms, and colored beads.
The “board” is designed “so that when one shakes the box and tilts it, the
beads roll into the V. Chance determines the color of the bead that rolls into
the V’s apex.” The robot’s move is figured out by skinning a box, opening it
up, and looking at what color bead is in the apex. The machine continues this,
leaving the box open until it either wins or loses the game. There is a system
of reward and punishment for the game that mimics the way humans are taught
things. Michie competed with the MENACE for two hundred and twenty games. By
the end of this time, MENACE had learned all possible variations a person could
make, and had become a master player.
Gardner
designed a smaller, simpler version of the game, called Hexapawn. In hexapawn
is played on a 3x3 board with three pawns on each side. You win by achieving a position where the
opponent can’t move, getting a pawn to the third row, or capturing all enemy
pieces. There are 24 of the little board set up, and the reward system for the
learning robot playing is made of Skittles.
On each of the 24 boards, the possible moves the robot can make are
shown. The person always makes the first move. You move the pawn, find a box
matching the way the 3x3 board looks, and with your eyes closed, pick a Skittle
from the box. You continue until the game ends. Gardner says that if you chart
the first 50 games, you can see how quickly the robot learned to anticipate
your moves.
Matchbox learning machines can be built to
play almost any game, with the exception of mini chess. Mini chess is
considered to be too complex. This made me wonder how much more advanced one would have to make a robot in order to play chess, and why it is so much more complex.
This chapter was very interesting. I think
that it would be fun to try and construct a matchbox learning robot, if only to
eat the Skittles it contained. I think that it would be cool to see if there were
any other ways to build a simple learning robot other than matchboxes.
I found this chapter extremely interesting because I have always wondered how computer/learning games were capable of doing this. The chapter does not totally explain it but it lays out the foundations. I was surprised that you could make your own Hexapawn with not too much work. To make this learning game you just need some matchboxes and some other materials. I say that this hexapawn/learning game is not too complex because I thought this kind of thing would deal with a ton computer programs and data. The only hard thing I found to understand about the match boxes is the V's apex. I just wish they could include some pictures or something but I understand the the match box as a whole. I wish the chapter would include some more current learning computers and how they work now because I want to know what they consist of. I also was wondering why mini chess was too complex for a matchbox learning robot, but I would assume that it deals with all the different kind of moves for different pieces.
ReplyDeleteLearning machines have always fascinated me ever since I was a little kid so this chapter hit close to home for me. In Chapter 35 A Matchbox Game-Learning Machine, they touch on the topic of machines becoming to smart and developing a will of their own. The area with the biggest threat of this happening in would have to be the military like they discuss in the book. The whole situation of machines becoming smarter than humans and developing a will of their own reminds me a lot of the movie The Terminator, where they develop a machine called Skynet, its jobs was to be a Global Defense Network/Information Grid. Skynet became self aware and sought out to exterminate the human race. Although it is a fictional movie many people fear that this may actually happen one day, and at the rate technology is advancing I can see why we would be afraid of this. I agree with Jo and I think it would be a lot of fun to build and test out our own Matchbox Learning Machine. The tic-tac-toe learning machine MENACE would be a relatively easy machine for us to make and run our tests on.
ReplyDeleteLearning machines is a very intriguing topic to look at because of the infinite possibilities for them. As Vito mentioned in his comment, this chapter reminded me instantly of a movie as well. The movie I, Robot exploits the idea of a computer becoming smarter than its human creators. The computer, who runs its system through robots, develops feelings and a will to "protect" the human race by terminating it and starting over. I believe people fear this possibility of the future much like it is mentioned in this chapter. As the Jo mentioned and the author describes, the computer has limitless possibilities. Its ability to continuously learn and track data is what make computers so incredible, but also for some people scary. Computers have contributed so much to mankind and have allowed us to grow as a species immensely by allowing our ideas to be put into reality. In terms of gameplay, these Matchbox game-learning machines have taught us different ways to look at and approach the way we play. By, the computer being able to retrieve data of every game it has played and use it to strategize the best way to win it eventually becomes a master and unbeatable. Although this is true, we are able to look at that data and learn from it and how it "thinks" and apply its gameplay to our own strategies. There is a study that says it takes 10 000 hours of practice at one task to become a master at it (for humans). As the text describes, these matchbox game-learning machines take a lot faster to become a "master".
ReplyDeleteI think it would be very interesting to play against a matchbox game-learning machine and see how long it takes the machine to develop a "master strategy" to win every time and then look at the data to see how it developed that strategy.
I found this chapter really fun to read! I think machines are really interesting because they can undoubtedly become smarter than humans. They already have an advantage of not making mistakes, as Gardner and Jo both stated, but they also learn much faster and can remember more than most humans. This chapter doesn't elaborate on it, but the crazy potential machines have become smarter than humans is rather terrifying to me; having machines control everything including us humans, who created them, is frightening. The machines Gardner talked about here like HER and MENACE aren't types of machines that are worrisome, however, because they're programmed to play games. I thought how the machine was taught all the scenarios in the game was rather interesting; the “trainer” of the machine has to “punish” it when it screws up, and theoretically, Gardner says it should only screw up 11 times before it learns the perfect game. In fact, it only takes 36 games for HER to learn the perfect game, which is crazy!
ReplyDeleteI think it would be really fun to play against a matchbox game-learning machine and see if I could win against it. The game would be especially awesome if candy was involved.
This comment has been removed by the author.
ReplyDeleteChapter 35: A Matchbox Game Learning Machine really captured my attention. I find the idea of having these "smart" machines or robots to be both exciting and terrifying. The writer did a great job of explaining how non-advanced these robots are, but as time goes on technology only becomes more successful in discovering new ways to improve what already exists. As I previously stated I think it would be pretty cool and scary if computers became advanced enough to think on the same level as or beyond the human brain. It would be pretty nerve-wracking to imagine a robot finding human beings to be illogical, but it would be very exciting to interact with artificial brain activity. A robot that can calculate what your next move is going to be in a simulation is very interesting and would be rather remarkable to create. Though it is still at a lower level as far as technology is concerned it is a step towards creating robots that can do more tremendous things and contribute to society. Though the robot can only play something as simple as Hexapawn, it shows how much potential there is for these pieces of work. Lastly, whoever the genius is that incorporated skittles is my hero.
ReplyDeleteThis Chapter caught my eye just by reading the title and I was a little envious that I would not be writing the original post myself. Luckily though I was still able to read the chapter and make my own comment. Having made my own tic-tac-toe game through programming I understand how tedious the task is of making an AI to play against. At best I was only able to make a program that would randomly pick an empty space and play there, a far more inferior opponent than a human. I can't even imagine the process of making a computer be able to think for itself and not only learn from its mistakes but also become so "smart" that it becomes an unbeatable opponent. This chapter has also compelled me to write a program for the Hexapawn, just to play of course not to make a learning machine since that is far beyond my level of Computer Science knowledge. On the topic of minichess, I believe this is too complex solely because of the different types of movements the computer would have to take into account. It would be extremely difficult to not only make the computer learn the movements but also to be able to anticipate the opponents moves as well. This chapter was one of my favorites that I have read this far and I believe you did a great job of summarizing the chapter
ReplyDeleteThis chapter was very interesting to read. Just thinking about the numerous possibilities of robots and how fast they can master a simple game like tick tack toe, for instance. Even if machines are smarter than humans and are able to think way faster than us, but machines are only as good as we make them. Yes they may be able to do little games insanely fast, but it was because we gave them the capabilities to do so. Everyday technology is increasing and getting better and faster. Its scary to just look back ten years ago and smart phones weren't even a thing or twenty years when computers were just starting to get popular. It is an interesting topic to think about how fast technology is evolving and robots that can be made by this. So my question is why was chess the exception for these games. Indeed it takes a lot of thinking and there are numerous moves that could be made, but there has to be other games that are like chess. I would like to see the machine that would be made to play chess look like and how different it has to be than the other, more simpler ones.
ReplyDelete