Slashdot Mirror
Go For It On Fourth Down? Ask Coach Watson
jbrodkin writes "If humans can't beat a computer at 'Jeopardy!' why should we trust them to make the right call on fourth down in the Super Bowl? That was the fundamental question asked by some researchers at the recent MIT Sloan Sports Analytics Conference. With thousands of variables to consider on the basketball court or other fields of play, it only makes sense to let computers handle questions of strategy, says Tarek Kamil, whose company built a chip-containing basketball which takes 6,000 measurements per second. 'Fifty years from now, we're going to laugh about how we used to give coaches this much responsibility,' he says."
Go shotgun, if they bunch up on the line to stop the run you can fling it out to the sides for a run or pass.
I never punt in video game football.
But the Super Bowl and Fourth down are football things. Not basketball things.
As for sports at the upper levels, there is more involved than merely picking the correct play. You need not only the play, but the execution of it. Coaches do far more than just come up with strategy, they also, as the name implies, COACH.
If brevity is the soul of wit, then how does one explain Twitter?
Why should we even let humans obey the orders? Machines can do it more efficienty.
And then why do we need to do it in the physical world? It might be more interesting if there's no gravity, or higher gravity or something.
So the entertainment of the future will involve us seeing computers play video games in front of us.
Jeopardy is not Strategy. Jeopardy is taxing the computers ability to understand reasonably complex language, and find the solution to a question. A computer has a database of answers, and there is only 1 correct one to select. Football strategy does not imply only 1 right answer. It relies on so many more things than that. All in all, a pretty ridiculous statement.
Sports are rife with things that seriously improve performance, but are illegal since they don't enhance the competition of man v man, how will using a computer to do the job of the coach be any different?
It actually worries me a bit that we're letting computers make decisions for us. I can see doing labor and computations, but when you place the machine in the decision making process and the human follows, then the machine has become the master. I know right now it's all well and good and very acedemic, but I really REALLY don't like the idea of say, a machine manager. People feel that we're already cogs in a great machine, this is just a baby step towards a very scary future.
Also, when we let machines make decisions for us, we officially stop thinking for ourselves and we let those who created the machines do the thinking for us. The machine is becoming a crutch rather than a tool, and if that happens we as a people will cease to intellectually grow.
They consider knowing the answer to everything ever with no limits one variable while numbers of gameplays more?
If anything this is exactly what a computer is designed to do, not fish through millions of facts and try to use them to answer a question which it may or may not even understand.
Ridiculous comparison.
Game theory is based on rules and statistics which is what a computer loves to crunch, random facts? Not so much.
Sounds to me like a salesman selling his wares...
Replace the coaches and players with robots. Or just simulate the whole thing in cyberspace.
Games are exercise. Pro sports forget that. A big lot.
If both teams relied solely on computer models to make the decision, both teams would likely know whether an attempt on 4th down would be attempted. There would almost never be an unexpected attempt, and the only unaccounted variable would be the actual play to be run on the attempt, which could also be predicted relatively accurately by considering coach play calling tendencies.
Make everything as simple as possible, but not simpler. - Albert Einstein
What are these things? (Blatant US-biased post...)
"The greatest lesson in life is to know that even fools are right sometimes" - Winston Churchill
trying to outfight a jet? almost like believing one could take down a giant steel monolith with a flying taxi-cab.
there are two risk reward ratios
the ghetto risk where you risk a lot for little potential reward. choose any stupid scam you read about where the idiot criminal gets caught for stealing very little money in the big picture. say a few thousand $$$.
the good risk where you risk a little or even a lot but for a good reward. like say finding a new job at google.
fourth down is a bad risk unless you're losing and there is very little time left since most plays in sports result in very little reward or none at all. like in baseball where hitting the ball 1/3 times means you're a superstar.
Seriously, there was a Simpson's episode about this very topic.
One of our competitors trademarked the term "hypothesis". From now on, we will call them "boneheaded ideas".
My football has only three downs you insensitive clod!
Actually, the blog 55 Yard Line has an excellent article on whether or not more coaches should go for it on third down based on yards to go for a first down and the line of scrimmage.
I call it 'The Aristocrats'
How do we quantify a human characteristic like courage? Knowing a player is competing despite having a minor injury is one thing; knowing how hard he will compete and how well he will play is something else. Until we can quantify both qualitative human characteristics like courage, fear, ambition, and stubbornness, as well as the "gut call" a coach makes based on his impressions of the individuals on the field and how they function as a group, computers aren't going to be better at this. They will probably be wrong more then humans are. How do you teach a computer that when dealing with people, the whole is often greater than the sum of its parts? We are basically talking about creating computer programs that are not completely logical, but that are both logical and intuitive--which implies a certain amount of irrationality.
I don't care about strategy, but the NHL is lousy at telling if the puck is above the height of the crossbar when a goal is deflected out of the air. My team has been obviously burned on this twice this season, one game-winning goal for the other guys allowed and another disallowed for us that should have counted. I would love for stats so the ref could instantly tell if a puck crossed the goal line under a goalie, too.
Peter predicted that you would "deliberately forget" creation 2000 years ago...
If humans can't beat a computer at 'Jeopardy!' why should we trust them to make the right call on fourth down in the Super Bowl?
Good point. I will replace the head coach of my NFL team with a computer as soon as I'm done having dinner with the queen of England.
You're an editor at some network-central industry rag. One of your staff writers heads up to Boston to attend the Sports Analytics Conference. He rubs elbows with big names in professional sports, attends a few break-out sessions that discuss gathering data to reduce player injuries and data acquisition inside basketballs, and then writes up a rambling article to justify his expense account. What's are you to do? Still hung over from the "Watson vs. the humans" Jeopardy! party, you write the headline: "Go for it on fourth down? Ask Coach Watson." This creates such a media firestorm on /. that all coaches in all major sports retire simultaneously. Talk about march madness!
Perhaps Watson could just replace editors?
I need trepanation like I need a hole in the head.
The example given was a tactical, not strategic decision.
"For the honor of the regiment" /obscure
Why not just use robot players too.
'Fifty years from now, we're going to laugh about how we used to use humans as players'
This is what inordinate faith in computers gets you
http://www.youtube.com/watch?v=yoF0a32DhLw
Really. Following that logic, just build a series of air cannons on wheels that can pass the ball back and forth and make baskets with perfect accuracy from anywhere on court. After all, then computers will "play basketball better than humans". Oops, I forgot...no one would give a damn. We go to basketball games to watch players play ball. That includes the mistakes they make. It's entertainment, not an engineering problem.
Just simulate it and publish the score.
Warning: this article may contain humor, sarcasm, parody, and perhaps even irony. Read at your own risk.
The robots will be laughing at the robot faction that claims there were once biological beings on the planet.
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
I usually related the story like this: In the 1700s, Isaac Newton and Leibniz invented calculus (you know that really hard stuff we still have a difficult time learning today.) By the late 1800s, math guys knew how to do almost anything using just a pen and paper (calculate orbits, really advanced mostly graduate level math stuff). They felt brilliant. And they said, "shoot, 30 years from now, we're going to essentially be God. From any starting point, we'll be able to predict any outcome."
Then the undecidability stuff with Goedel happened, and then we had intutionism with Brouewer, etc, and they realized it wasn't to be.
This sounds a lot like that.
"Fifty years from now, we're going to laugh about how we used to give coaches this much responsibility"
WTF? Talk about completely missing the point.
Deleted
That comment is far more insightful than you probably meant for it to be. The current generation of coaches learned to coach without computers. In a few years, the next generation will include people like you that have also played video game versions. With the accuracy of video game simulations improving all the time, more coaches will trust the instincts that they've learned are reliable in the video games, and the typical coaching strategy will change.
Take it a step further (which some NFL team will), and get a good video game (i.e., simulation) that includes the stats on all the players on both the home and opposing teams, and run lots of plays in the week before the game to see which ones tend to work better than usual with the expected lineups.
Baseball has long been a numbers game, primarily because so much of the game is a matter of batter vs. pitcher, so it's relatively easy to quantify. It's just a matter of time before other sports follow.
Just read again your Captain Tsubasa mangas or watch the 1994 animation series.
Fifty years from now, we're going to laugh about how we used to give coaches this much responsibility
I hope not; coaches are essential to the nature of a team; these are human competitions, and if we're not considering robotic players, we have no business looking into robotic coaches. Each coach has his or her own take on what should happen, including intuition, foresight, insight, and motivational talking. No computer is going to be able to give an exhausted player a second wind by talking about probabilities and describing an unrehearsed play.
Consider American football, which is often dubbed a chess match between the two coaches. To remove the human element of a game makes it boring, and it's too often forgotten that the coaching staff is what makes players tick. Off the court/field is another issue altogether; we already have lots of extra advisers and computing power to crunch the numbers, and that's where they should stay.
That said, I'm all for a human-v-machine all-star matchup, where the champion team plays an all-star team conscripted and managed by a computer. If the computer is that much better, perhaps the champs could play the season's most average team, or an all-star lineup of the minor leagues or NCAA.
I couldn't help but also tag this article Skynet for the path it seems to steer us towards.
Use my userscript to add story images to Slashdot. There's no going back.
Because humans are fallible and can't comprehend as many variables as a computer. It's that lack of perfection that makes games great.
First thing - computers can't beat humans at strategy games where there are huge variables in risk versus reward, such as No Limit Texas Hold em. They can use strategies of course but a pro player can tear them to pieces. See the notes on Man versus machine where Polaris went up against poker geniuses like Bryce Paradis and then read his post analysis if possible. And that was a game of limit. No limit adds a layer of complexity that makes the mind boggle. And that's a simple game. Imagine applying strategy to a game where variables such as human emotions adds weight to a decision.
Two - Could the machine ever gather the same information in the time given a coach? A coach intakes an immense amount of information using all his senses and his staff of assistants. When a computer can intake the same amount of information on a field of play we will have created artificial life. Not something trivial.
Game strategy is ultimately based on information gathering inside information sparse environments. How can a computer compete in that realm?
Tiger Blooded Bi-Winning Machine
And I thought that the reason that statisticians weren't good as sports coaches or generals was because there were 1000's of variables that couldn't be quantified or computed.
oh... wait...
--------------------------------------------- "In the end, we're all just water and old stars."
It has always bothered me how college and NFL football fans think being a good coach is such a rare gift that only a few people in the world can do it. Teams will give contracts for millions of dollars on some coach hoping to bring their team back from a few bad seasons. Then the coach does poorly and gets the boot. And the cycle repeats itself. Truth is that high schools around the country are filled with coaches who have equal chances of succeeding at the NFL level as even the most sought-after coach. Yes there are good coaches and bad coaches, that's not my point. My point is that reputations for talent in coaching get greatly exaggerated. Good coaches are more common than people think. I'd prefer a world where the coaches salaries are back down to earth and ticket prices are more reasonable.
The fact that football coaches don't go for it on fourth down is actually really interesting. It's basically a question of risk aversion -- kicking on fourth down is "safe", while going for it is expected-value-positive but highly risky.
In real life, it makes some sense to be risk averse. Money has decreasing utility the more of it you have, so it's completely rational to refuse to bet your life savings even if there's a 51% chance you'll win. In a game like football, however, utility is necessarily proportional to the probability that you'll win the game. If a play increases that probability on average, it's the correct play no matter how risky it is. Football coaches are treating the probability of winning as if it had decreasing utility. They're being irrationally risk averse.
Once you start looking for irrational risk aversion in games, you start seeing it everywhere. It comes up just as often in Jeopardy as it does in football -- people tend to risk less than they should when they hit a daily double (early in the game, you should almost always bet everything). Funny enough, I don't think Watson got it right either -- in the second half of the first match, it made a relatively small bet when the rational strategy was probably to bet everything.
Don't computer games, Madden and others, basically already do this? At least on a simplistic level?
Any coaches not already using computers to help with their strategy are doing themselves a huge disservice. I'm sure they are already crunching all the statistics they have, like how often a 4th down conversion works and using that to help with strategy.
However, I'm not sure this microchip does anything that anybody is interested in. It probably costs a fortune too, and they're putting it in a ball that's going to get knocked around? I'm sure the NBA is not sold on paying a lot of money to find out, as the article mentions, whether Johnny is 14% more dominant with his right hand than his left. And useful stats like "Time of Possession" will still have to be done by a human.
In fact, I can't think of a single stat important enough for a microchip in a ball to transmit in real time, and even if it was, it transmits to BOTH coaches. It's in the ball, so it creates somewhat of an arms race and just creates more information and work for the coaches to consider.
Adidas To Bring Back Sneakernet
I research AI for strategy.
A computer is really good at finding optimal strategies if you can properly quantify relevant variables. In sport there is advantage in not taking an optimal strategy, because your opponent won't know which non optimal strategy you've chosen until it's too late. If you're going to use randomness to determine which strategy to use, then the computer is no better than a coach.
That assumes, probably wrongly, that you can quantify what's going on. Is that opposing quarterback's limp important, a fake, how serious is it (numerically)? Even if a computer is good at predicting one particular game, (say the superbowl) that would be based on the data from the whole of the rest of the season to assess how good the players are.
There's a lot of sport to be had in running AI's against each other, especially based on the same sets of data and see what they do. But that is a *very* different problem from actually simulating a real match, yes, the average of 10000 trials may be correct, but there are only a few real games, not thousands. That randomness, sportmanship, and people doing extraordinary, unexpected and great (or stupid) things is what separates a real match from a statistical model.
Wow 6000 measures per second!!
If only there were a machine being able to process massive visual information(tenths of millions of measures per second), and sound (tenths of thousands), along with 3d perception and emotional and physical information about your players and being able to valid extract info from that and structure that into abstractions like a language and logic and make decisions about it(strategy)... oh wait...
Ocean's cluster of ninety IBM Power 750 servers (plus additional I/O, network and cluster controller nodes in 10 racks) with a total of 2880 POWER7 processor cores and 16 Terabytes of RAM. Coming soon to a theater near you!!
Can the two break the banks at major Las Vegas casinos? The difficult part is getting Watson to do all those spectacular acrobatic tricks. I guess they'll have to use a stunt Watson. Or maybe Watson can generate his own stunt scenes, just like Jackie Chan does his own stunts?
Um, sorta.
Man, that huge cast is going to cost a fortune!
Schroedinger's Brexit: The UK is both in and out of the EU at the same time!
Not video games. Get out and see it, geeks, Jesus H.
Responsibility?
It's 22 idiots fighting over a piece of leather. It's the definition of irrelevant. Using a computer for it would be a waste.
WTF am I doing replying to an AC at 5 A.M on a Friday night?
That's elementary, my dear Watson! ...wait what?
even if he was 'just some guy', it would not be surprising if he shed a tear noting how far we've come, based on his request that we honor life (not just our own) above all else. even if he's not god, ALL of the manuals say the same thing, save one, in georgia. egg hunting, or altering?
While I absolutely believe that a computer with enough processing power could make better decisions than the best-informed coach, it starts the slow death-march away from being a sport when the coach is removed. Football (like most other team sports) is a contest of human strategy as much as human ability and, as such, is simply less worth watching if computers run the show.
-- Adam McCormick
So I'll watch less sports than I do now, which is basically none anyway.
The price is always right if someone else is paying.
Basketball and football aren't chess, and if both teams have the same software and stats, then it all comes back to a human making the call again.
I can see something like an arms race, different clubs trying to see who can come up with the best software and analysis, but there are limits to that, too.
If this were carried to its logical conclusion, with software calling all the shots, it would make sports very boring, very quickly. No thanks.
But it's so simple. All I have to do is divine from what I know of you. Are you the sort of man who would make an attempt on 4th down, or let his enemy? [pauses to study the computer] Now, a clever man would make an attempt on 4th down, because he would know that only a great fool would make an attempt on 4th down. I'm not a great fool, so I can clearly not make an attempt on 4th down. But you must have known I was not a great fool; you would have counted on it, so I can clearly not choose to make an attempt on 4th down.
computer: You've made your decision then?
Not remotely. Because making an attempt on 4th down comes from Australia, as everyone knows. And Australia is entirely peopled with criminals. And criminals are used to having people not trust them, as you are not trusted by me. So I can clearly not make an attempt on 4th down.
computer: Truly, you have a dizzying intellect.
Wait till I get going! Where was I?
I hate soccer
You better watch out, there may be dogs about . .
It's in today's New York Times, a short story in the Bits section about a Congressman who did it. Here's the link.
If you think about it, Jeopardy's a much easier game. It's a question and answer with a 1-to-1 mapping of question to answer. If IBM didn't waste all those millions, they got really close to having 100% of the answers in the database. The hard part might have been some of the word associations to find some answers and real world knowledge to avoid bad answers. In any sports, there's no way to predict what the opponent will do strategically, nor how to discern the moods of the players and the coaches, or deal with slips and errors. How is this coaching Watson figure out who was hung over or sore from too much sex? Or who was madder than hell that day? How is all the data about the way a basketball bounces going to help?
There's an interesting take on how this new item was sort of buried in the newspaper here.
You may have a quarterback that wants to call his own plays.
When a QB sees how a defense is falling into a preprogrammed arrangement he can call an audible.
Just look at sports that could or can already rely heavily on technology and computers. Car racing comes to mind. Think of all the things you have in your very non-high tech, very commercial car. Breaking and steering aids, automatic transmissions that switch gear at the optimal time (and automatic transmissions are no longer slow and sluggish, we're at the point where a gear shift takes a few milliseconds, faster than any human ever could, and more gentle with less wear on the gears too), traction control and all the little things that allow you to drive more safely. Most of these things are not allowed in racing cars, because with them, the pilot would essentially very soon become a passenger. Just think of it, how hard would it be to create a computer controlled car that races perfectly?
Hence I think that such "strategy aids" will soon make and appearance and soon after they'll be banned.
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
So here is what I wonder. Given a really complicated model, will it ever be possible to determine all of my actions in the future, and at what level of sophistication would that model have to contain? If it is possible to determine every one of my actions, what would that imply for the concept of free will? I don't think that the model would have to be at the molecular level, and that knowing external pressures would be enough to determine a large variety of actions. At what point does life become so predictable that it isn't really fun anymore?
take the fun out of everything department
Also, I'd love to see its speech synth tackle Japanese names. That could just be funny.