The Decision-Making Puzzle
If you’ve ever played the classic puzzle-like computer game Tetris, you know that it starts out slowly. As the seven different pieces (called “zoids” by the initiated) descend from the top of the screen, a player has to shift the pieces horizontally and rotate them so that they fit into a gap in the stack of pieces at the bottom of the screen, or “well.” In early levels, the pieces might take 10-15 seconds to fall. The speed increases at each level.
In world champion Tetris matches, players often start play at Level 18—in which pieces are on the screen for about a second. Wayne Gray, a professor of cognitive science at Rensselaer Polytechnic Institute, calls it a problem of “predictive processing and predictive action.” Champion-level expert players, he says, are able to take in the state of the gameboard and react almost immediately, without going through the mental steps of figuring out how to move the piece and rotate it that a novice player requires. “They can see the problem and reach a decision at the same time,” he said.
Gray and colleagues have attended the Classic World Tetris Championship tournament for three years, collecting data from expert players using a modified version of the game that collects keystrokes and eye-tracking data. He joins Ira to discuss what the researchers are learning about expert decision-making, and what he hopes to study at this year’s upcoming Tetris tournament.
Wayne Gray is a Professor of Cognitive Science at Rensselaer Polytechnic Institute in Troy, New York.
IRA FLATOW: This is “Science Friday.” I’m Ira Flatow. A bit later in the hour, an update on current Alzheimer’s research. But first, if you’ve ever played the computer puzzle game Tetris, you know how it is. It starts off slowly, the pieces drift down, and you have maybe, what, 10, 15 seconds to think, position, spin, drop the pieces into position.
Players have to take in the board. They have to assess the piece and play. You have to make a snap decision. But what happens? It gets faster. And at the top levels of play, pieces fall down the screen in a second or less. Can that teach us something about how the brain works in learning, cognition, or decision making?
Wayne Gray is a professor of cognitive science at Rensselaer Polytech in Troy, New York. And for the past three years, he’s been bringing a team of cognitive science researchers to the Classic Tetris World Championships. And this year’s tournament starts next weekend. He’ll be there attending for the fourth year. Welcome back to the program.
WAYNE GRAY: Why hello. Thank you.
IRA FLATOW: Now so you’re going to the championship for the fourth year. Tell us what it’s like there.
WAYNE GRAY: Wow. Well, especially our first year, it was just chaos and confusion from our perspectives. The first day of the tournament– first two days of the tournament, people were trying to qualify to get a slot on the championship playoff schedule, which they had 40 openings for. And people could play all day as many times as they’d like. And their highest score would count on whether or not they made it. So there was a lot of sweat and nervousness going on.
IRA FLATOW: I’ll bet.
WAYNE GRAY: We had our booth set up to collect some different sort of data from them.
IRA FLATOW: OK. So let’s talk about that. Let’s first talk about how much better these people are than the regular person at Tetris.
WAYNE GRAY: Wow. Well, we thought we knew Tetris before we went there, because we had about 400 undergraduates play an hour of Tetris in a laboratory, because everybody knows Tetris, right? That’s why we picked this task to start with. None of those people got past the level– made it to the end of level 16.
In fact, only about to eight got to– actually, only about three got to level 15. They all died. And we said, well, of course not, because at level 15, it takes 1 and 1/3 second for this piece to fall all 28 lines. And at level 16, it takes one second. So, of course, nobody can do that. And then we go to the Classic Tetris World Championship and they start playing at level 18 where it takes one second to fall from top to bottom.
IRA FLATOW: That’s amazing. So let’s talk about the steps that are involved. You have to what? You have to see the board. You see the piece coming. You have to rotate it. Then you have to put it into position. And the best players are better than anybody at doing these things.
WAYNE GRAY: Yep. The best players are really, really good at that.
IRA FLATOW: Do they play it differently from us? Do they look at the board and just know or decide they know where that piece is going quickly?
WAYNE GRAY: Yes. Yes. And this is all about what makes Tetris such an interesting thing for studying extreme expertise and these perceptual motor skills. All such skills are different, of course. Racing car drivers or laparascopic surgeons are very different than Tetris players, but not different in the way they have to adapt, the way they have to learn to see things in advance before things happen.
These people know when they see a zoid, they know exactly how far to rotate it, how fast to move it, and all that stuff, whereas people like me will be spending a lot time maybe moving the Tetris piece around the board looking for a good place with it as it’s falling. But these guys and gals they just see the zoid, know where to put it and put it there by the shortest path, the minimum rotations so that when it fits it fits most tightly into the stack that they’re building.
IRA FLATOW: That’s amazing. And so how can you learn from them what you’d like to know about brain function?
WAYNE GRAY: Ah. Very good question. Well, mechanically, we’ve been– this coming year, we’re going back with different tasks than we had before. So one thing which always a little bit annoyed us is that when they play Tetris at our booth, and whenever we start at level zero, they’d be turning around in the chairs talking to us as these pieces are falling luxuriously and having a conversation up until the thing got to level 15 and then they started taking it seriously.
This year we’re going to play– have them play their version of speed Tetris where the goal is not to get points but to go from level 0 to 19 as fast as possible. And the winner is the one who goes– who not gets the most points but, again, gets there as fast as they can. And that’s important, because then we have good data at the slower levels from the experts that we can use compare to our college students.
IRA FLATOW: When you say good data, what kind of data are you collecting?
WAYNE GRAY: Oh. We collect everything. We’ve written our own version of Tetris– not for sale, not for distribution, we hasten to add.
IRA FLATOW: Nuts.
WAYNE GRAY: Yeah. We could get in trouble for that.
IRA FLATOW: Curses.
WAYNE GRAY: But every single keystroke is timestamped to the nearest millisecond. Every single zoid that’s on the board, every Tetris piece on the board, every configuration. We know where the holes and gaps are. We know everything and have timed everything. So we actually can play back our log files if we want to and watch a game be played.
IRA FLATOW: Do you create this special game?
WAYNE GRAY: Yes, we did. One of my former doctoral students did, John Litsead. And actually now have a new version, because it turned out that the old version didn’t incorporate the right bugs. Believe it or not– well, I guess it’s easy to believe that the people who programmed Tetris the first time round actually hadn’t played Tetris themselves, the ones in the ’80s who basically created the original software. So they left behind bugs because they weren’t good enough for the bugs to bother them.
And it turns out that some of these bugs apparently, Tetris lore has it, are what the extreme experts are exploiting to get to the higher level places where the pieces falling from top to bottom in one second. How can you do that? That’s too fast for most people to actually move a piece from the center to the side. But these people have all sorts of tricks for doing that.
IRA FLATOW: Well, let me ask you about that. Do you find like in all sports there are naturals at the sport, or can you be trained to be great at Tetris?
WAYNE GRAY: You know, until last year, I thought there was an easy answer to this question, because we get it a lot. But last year, the person who won the World Endurance Championship was a fellow named Joseph, who was about 17 years old at the time. And apparently he’d been playing Tetris for about the last six months. And he basically trashed the European champion, the Japanese champion, and the former world champion. So it was quite an exciting Tetris tournament last year.
IRA FLATOW: So you think it’s natural talent then?
WAYNE GRAY: You know, at least not for most people. Joseph may have had something else going for him. And so we’re very eager to get him back. We collected some data from him last year. And we’d like to get more data from this year to see if we can pin it down. He played a lot of music and several musical instruments. And he thinks that may have something to do his uncanny ability to play Tetris so well.
IRA FLATOW: That is interesting. I understand that your research is funded by the Office of Naval Research.
WAYNE GRAY: That’s right.
IRA FLATOW: Why would the department– why would the Pentagon be interested in this?
WAYNE GRAY: Believe it or not, I get asked this a lot.
IRA FLATOW: And your answer is?
WAYNE GRAY: Well, my answer is that there’s a lot of types of expertise out there which do involve perception and action. And we are not very good at training those things. We don’t really understand what do you have to do to train such things. And the first step on that is, well, how are these– what really are these skills? What makes people so fast? How are people able to do that?
So we believe that learning how people get good at Tetris and what they do when they are good at Tetris might lead us, for example, to design better laparoscopic trainers for Navy surgeons, or perhaps better displays and controls for Navy pilots, people who Navy has a heavy investment in these real-time dynamic decision making tasks where you just can’t sit around doing nothing. You have to make a decision, you know.
IRA FLATOW: You know, they have to do that everywhere in the military. Maybe the training is being applied to operating drones and making quick decisions like that or remote battlefield.
WAYNE GRAY: Exactly. And we do have some colleagues at other parts of the country who are studying drone pilots, probably not from the level of expertise that we are. But perhaps after we get to talk to them more about our Tetris findings, they will want to bring in more expert-level drone pilots.
IRA FLATOW: So you don’t have any idea how there is a transfer of skills here– if you’re good at Tetris, you’re good at something else– can be carried over?
WAYNE GRAY: Well, think about it this way though. You know, if you’re good at playing chess or you’re good at playing Go, if you’re good at typing on a QWERTY keyboard or you’re also good at writing script, the nature of expertise is that you become very, very good at something which most people are just sort of OK at.
If you’re good at four Olympic sports, are you also the best person for the broad jump? Probably not. Specialization seems to be the key here to getting really, really good. I mean, most people could get good at any one or two things they put their minds to. But we just don’t have the time. We don’t care enough about being the world’s fastest touch typist or the world’s best Tetris player.
IRA FLATOW: Well, can you teach artificial intelligence to learn from what these people are doing?
WAYNE GRAY: Ah. Ah. Well, as matter of fact, my doctoral student Catherine Seibert has models of Tetris playing which she has been building to try to figure out what do these models tell us about decision making in Tetris. Obviously, the models can move pieces at the blink of an eye so we can’t look at that component of it, but she’s trying to isolate the decision making component itself and how much this is influenced by things like time.
It turns out that when we take one of her models, which can play over a million lines of Tetris, and we say, OK, let’s see how well you’re doing after 500 lines with Tetris, well, 500 is a number we picked because that’s sort of about the highest number that we ever had a human being play in our laboratory. So for our Rensselaer students, surviving 500 lines would be very good. Well, it turns out that the models that are good in human skills start looking like humans.
So models that play forever, they do this by playing one piece at a time only clearing one line at a time. The ones who stop at 500 lines, they’re more likely to get four line clears at once, which is what’s called a Tetris. And you get 7.5 times as many points as you do clearing one line four times. So it’s very interesting. The constraints of the task environment straight to the situation seem to force both the models and perhaps the humans to adopt similar strategies if they’re only going to be playing for 500 lines.
IRA FLATOW: Well, Dr. Gray, I want to thank you for taking the time to be with us. We’re all more knowledgeable about Tetris. Wayne Gray, professor of cognitive science at Rensselaer Polytech in Troy, New York. Thanks again for taking time to be with us today.
WAYNE GRAY: Well, thank you very much for having us.
IRA FLATOW: You’re welcome.