i am making this post so someone can explain it to me
April 29, 2016 10:26 AM   Subscribe

Menace 2 is an artificial intelligence which learns how to beat a human player at Tic-Tac-Toe and also it is made entirely of wooden drawers and colored beads
posted by beerperson (36 comments total) 22 users marked this as a favorite
 
I would love to see a similar project for chess or go. How big would that cabinet be?
posted by any portmanteau in a storm at 10:41 AM on April 29, 2016


As far as an explanation goes, is it needed for how the algorithm works or how you can have an artificial intelligence made up of drawers and beads?
posted by any portmanteau in a storm at 10:43 AM on April 29, 2016


This is really a very cool thing.

Of course, it is also really just a program that uses a simple database implemented in drawers and beads. I have qualms about calling this kind of exercise an "artificial intelligence". Then again, maybe those qualms are themselves kind of instructive.

I would love to see a similar project for chess or go. How big would that cabinet be?

I think you'd run out of atoms trying to build it.
posted by brennen at 10:44 AM on April 29, 2016 [11 favorites]




It's been a long time since I read it, but wasn't this described pretty much exactly in Fred Saberhagen's book Berserker?
posted by adamrice at 10:48 AM on April 29, 2016 [3 favorites]


It's pretty cool, and I'd love to play a game against it. That said, it strikes more as an elaborate parlor trick based on random effects and human intervention, rather than a machine "learning" anything. But, it certainly makes for a clever illustration of how a machine can learn. That MENACE must always go first takes a little of the shine off. But, once again, I'd jump at the chance to play a few rounds with it.
posted by Thorzdad at 10:48 AM on April 29, 2016


it strikes more as an elaborate parlor trick based on random effects and human intervention

Honey, all of life is a an elaborate parlor trick based on random effects and human intervention.
posted by wemayfreeze at 10:52 AM on April 29, 2016 [35 favorites]


Metafilter: made entirely of wooden drawers and colored beads
posted by fairmettle at 10:58 AM on April 29, 2016 [6 favorites]


See, I would have gone with:

Your mom is an elaborate parlor trick based on random effects and human intervention!
posted by leotrotsky at 10:59 AM on April 29, 2016 [12 favorites]


Heh, this is really cool. It's also really quite accurate to how a simple AI* would work! Each board state (drawer) is a node on a graph. Edges on the graph (moves) are represented by the colored beads, and those edges are weighted (the number of beads corresponding to a given move / total number of beads in the drawer gives you the weight). As you play, you adjust the weights of the edge by adding or removing beads based on the game outcome.

"Adjusting graph weights based on outcome" is pretty much the definition of how reinforcement learning works.

* It's probably worth mentally substituting "agent" for "AI" in 99% of cases.
posted by ReadEvalPost at 11:01 AM on April 29, 2016 [6 favorites]


Has it figured out the only winning move yet?
posted by Dr Dracator at 11:02 AM on April 29, 2016 [4 favorites]


The only winning move is regurgitation of tired movie quotes
posted by beerperson at 11:03 AM on April 29, 2016 [8 favorites]


>Has it figured out the only winning move yet?

It has, and it does it automatically by default. It's probably doing it right now.
posted by Two unicycles and some duct tape at 11:04 AM on April 29, 2016 [8 favorites]


Calling this a trick implies that what appears to be happening is not what's actually happening. But the rules are transparent and all moves the system ends up making are based on random effects and human intervention memorization of past outcomes.

Nobody is actively and consciously controlling the moves the system ends up making. That would indeed be a parlor trick. It is confronted with a randomized environment, makes moves randomized within increasingly tightening boundaries and improves over time based on past occurrences.

it just requires some human assistance for execution because it lacks a suitable powered sensory and motor apparatus.
posted by Hairy Lobster at 11:04 AM on April 29, 2016 [6 favorites]


So let's put it on tank treads and give it cameras and a bunch of laser turrets. Then we'll see if it's a trick.
posted by Hairy Lobster at 11:06 AM on April 29, 2016 [3 favorites]


But, once again, I'd jump at the chance to play a few rounds with it.

If you really want to, it seems like it wouldn't be hard to build your own. All you need is 304 matchboxes and a large number of small beads in nine colors.
posted by baf at 11:06 AM on April 29, 2016 [1 favorite]


I don't believe their website name.
posted by clawsoon at 11:08 AM on April 29, 2016


it is also really just a program that uses a simple database implemented in drawers and beads. I have qualms about calling this kind of exercise an "artificial intelligence". Then again, maybe those qualms are themselves kind of instructive.

Have you looked at current trends in "deep learning"? Because I have some bad news for you. Would that it were so simple as an orderly database of little drawers; instead it's a giant hairball of logistic function weightings.

Although as you point out, those qualms are instructive and part of the point of the exercise, I think. See also the Chinese Room thought experiment.
posted by Nelson at 11:10 AM on April 29, 2016 [5 favorites]


The only winning move is regurgitation of tired movie quotes

This is my new go-to example for "Pyrrhic victory".
posted by howfar at 11:15 AM on April 29, 2016 [5 favorites]


We made one of these when I was a kid (I'm in my 50s now). I'm pretty sure there was a Martin Gardner column describing this. I remember that we used candy jujubes for some of the beads.
posted by Numenius at 11:28 AM on April 29, 2016 [3 favorites]


So if I wanted to start some kind of club for people who appreciate and enjoy the Menace 2, what would I call it?

Menace 2 Guild?
Menace 2 Fellowship?
Menace 2 Lodge?
Menace 2 Order?

I know there's a more apt name, but it's just not coming to me...
posted by Strange Interlude at 11:36 AM on April 29, 2016 [19 favorites]


I love this low-tech computing stuff. Same reason I like mechanical calculators. Makes me feel like I'm in a steampunk novel (or would it be clockpunk?)!
posted by Mogur at 11:40 AM on April 29, 2016 [2 favorites]


Obligatory link to Danny Hillis' tic tac toe playing computer, built out of tinkertoys.
posted by mr vino at 11:46 AM on April 29, 2016


I built a version of this for a science fair, based on a Martin Garnder column, around 1991 or 92.

We used tiny beads and matchboxes to keep the size down.

The project had three machines. One left to run the whole day, it became unbeatable a couple of hours in. One that we would reset to an intermediate state every dozen games or so, to give players a chance of winning about half the time, and one that we reset to zero regularly, to show how stupid it was at the start when it knew nothing.

We lost, got a C, and were accused of cheating by at least one teacher (how the fuck would we cheat, the whole thing is there in front of you!!!). The first prize went to someone pouring liquid nitrogen into milk and juice to make ice-cream, which was given to the judges to eat. The second prize went to the son of a big cattle farmer, who just put some bull semen under a microscope for people to watch. I can't remember the third prize, but they used large format full color printed posters in their presentation, which at the time was quite expensive.

My team was so disappointed, I was as angry as I have ever been at authority figures who just don't get it. It took a long time for the lesson to sink in. Sex sells, bribery works, presentation trumps content, and everyone likes free stuff. If you want to sell science, make it shiny and sexy, and give free food.

Am I still bitter? Just a bit.
posted by Doroteo Arango II at 1:16 PM on April 29, 2016 [87 favorites]


If you include rotations and mirror sets, the number of drawers doesn't turn out to be that big. Years ago, there was a class taught to kids at OMSI (Oregon Museum of Science and Industry) that did this with match boxes and M&M's. Each match box had a picture on top with a color on each unused square, and there were M&M's of all those colors inside the match box.

The kid would play against the "machine", of course. Each move by the machine he would find the appropriate box and take a random M&M out of it to see what the "machine" played. The M&M would be left on top of the box.

When the game was over, if the kid won he'd eat all the M&M's. If the "machine" won then all the M&M's got put back into their respective boxes.

Of course, eventually the "machine" "learned" to play a perfect game.
posted by Chocolate Pickle at 1:23 PM on April 29, 2016


Thorzdad:

We simulated a version that could go second, in Basic or Logo, I can't remember.

The naive approach results in 9 complete machines, one for each starting position.

One of my friends realized that if you rotate the board, you can reduce it to 3 starting positions: Corner, center or edge.

That is why we ended up with three copies of the machine, but we decided it was more interesting to have three different 'levels' than to be able to play the second move.
posted by Doroteo Arango II at 1:24 PM on April 29, 2016 [3 favorites]


Am I still bitter? Just a bit.

If it makes you feel any better about the experience, this comment has brightened my day considerably.
posted by brennen at 1:40 PM on April 29, 2016 [7 favorites]


Neat! I made a matchbox version of this too as a kid, also after reading Martin Gardner. I used England's Glory matchboxes. It was just for fun, and it was a lot of fun! My parents were not too sure about the large container of unused matches I generated though.
posted by carter at 2:10 PM on April 29, 2016 [1 favorite]


Huh, apparently a lot of people have built this matchbox learning machine. Here's James Bridle's (yes, that James Bridle).

I think I found a scan of the Martin Gardner article.
posted by Nelson at 3:20 PM on April 29, 2016 [3 favorites]


I would have gone with;
God is an elaborate parlor trick based on random effects and human intervention!

And;
Menace 2 Guild?
Menace 2 Fellowship?
Menace 2 Lodge?
Menace 2 Order?


Menace 2 society?
posted by boilermonster at 4:01 PM on April 29, 2016 [1 favorite]


(or would it be clockpunk?)

Clunkpock.
posted by sneebler at 4:10 PM on April 29, 2016


I would love to see a similar project for chess or go. How big would that cabinet be?

You could at least implement an opening book for chess, probably best as actual books. Although since there are 10 million positions after 7 moves you're going to need a lot of volunteers.
posted by RobotVoodooPower at 4:26 PM on April 29, 2016


Menace 2 society?

Congratulations, you get a MeFi no-prize...
posted by Strange Interlude at 9:34 PM on April 29, 2016 [2 favorites]


I remember this back in the sixties, where it used empty matchboxes (when everyone smoked and there were no lighters) with different coloured buttons. I was fascinated by it and realised that it was a learning computer. I desperately wanted to build one but I didn't manage to collect enough matchboxes.

And so began my long journey into technology...
posted by arzakh at 6:32 AM on April 30, 2016


And now in modern times Amazon Web Services will rent you as many matchboxes as you need for $0.01 / hour. What a world!
posted by Nelson at 7:25 AM on April 30, 2016 [1 favorite]


My team was so disappointed, I was as angry as I have ever been at authority figures who just don't get it. It took a long time for the lesson to sink in. Sex sells, bribery works, presentation trumps content, and everyone likes free stuff. If you want to sell science, make it shiny and sexy, and give free food.

I took an AP Physics class my junior year of high school. For our final project, we were given free reign to build something using the principles we learned over the year. I somehow lucked into an absolutely brilliant team of people. We (okay, they) came up with a working scale model of something very like the hyperloop train using PVC pipe, a large ball bearing (for the train) and linear induction motors made by coiling wire around magnets (or something like that; I don't remember the details well at all, unfortunately). This took hours and hours of our time, but was probably one of the coolest things I could remember doing in school. At the demonstration in class, we got the ball to roll forward on the track, showing the possibility of such a train working. Our teacher sneered at it, told us it wasn't useful, and then awarding one of the highest grades to a person who had created a rotating plastic fork that made it easier to eat noodles.

I'd say that the long term lesson I picked up was to know the room, to empathize with the user (sorry, buzzwords) and to think critically about how you sell and present your work.
posted by rhythm and booze at 12:17 PM on May 3, 2016 [1 favorite]


« Older "Dad, I'm almost there..."   |   Is This A Bubble? Newer »


This thread has been archived and is closed to new comments