Computer Game Predicts Player Moves

willatnewscientist writes "A couple of Hungarian researchers have developed a computer game that knows when you're going to press the 'jump' button ... 2 seconds before you do it. The researchers use neural networks to analyse several type of biofeedback signal — heart rate, EEG and skin conductance — and discovered that skin conductance alone is enough to predict a jump up to 2 seconds beforehand. They say the technique could ultimately be used to make aircraft controls that respond more quickly to a pilot's actions. But it could also be used to create so-called 'frustration games' that respond to a player's actions before they occur."

Reminds me of that story Predictor

[Pojut]

http://www.nature.com/news/2005/050704/pf/436150a_ pf.html [nature.com]

Re:

[Maljin Jolt]

In certain advance and surprisingly effective magical systems, like a hindu saiva siddha yoga for one example, causality is treated as symmetric relation in time dimension. Determinism is an illusion in such paradigm, letting free will remain superior above everything. Pitty you anonymous coward.

Re:

[try_anything]

surprisingly effective magical systems

We're going to need some definitions here, buddy.

Re:frist pist

[A beautiful mind]

I've predicted this post five minutes before you've posted it!

Big deal?

[Eudial]

Big deal? You can train chimpanzees to realize that I'm going to jump the hell out of the way when I see a RPG headed my way.

Re:

[Lehk228]

the point of this is to know which way you jump so the other bot can land a morter shell on your head after you jump.

Re:

[antic]

Me too. Much prefer an FPS.

You haven't played against some players

[Moraelin]

You haven't played against some players. Half the newbies will stand watching the pretty missile coming their way, then half of those start screaming "cheater!", "aimbot!", "wall-hack!"

My favourite example from way back when I still played CS goes like this:

So I just pick the first server that's not full or empty, from a list sorted by ping. It has only 2 players, but everyone has to start somewhere. More will probably come.

The first thing I notice is that suspiciously I'm assigned to counter-terrorist, whi

Guessing Game

[DaveMMR]

Yeah. Call me crazy but I'm not sure I want to fly in a plane where the computer is attempting to take a stab at guessing what the pilot is going to do. Just a bit unnerving.

Re:Guessing Game

[GrievousMistake]

"In two seconds, the pilot is going to panic and accidently push the stick sharply forward, as he reacts to my sudden preemtive forward tilt."

Re:

[Katmando911]

I think it's a great thing, it should greatly improve responsiveness. If the pilot iniates some action that takes x seconds to set up and then y seconds to execute then you can use these techniques to cancel out the x seconds set up time. It will be great in the computer field where you can have multiple pipelines predicting what the user wants and then just discard the invalid predictions.

Re:

[DaveMMR]

But is that elimination of the "x seconds" going to create a problem if there were a computer problem and the pilot has grown to expect the assistance and has to adjust his reaction time in an emergency? I would assume pilots know their way around an airplane without computer assistance though, so I can see an advantage to this technology, but only if they can also do without it.

Re:

[Katmando911]

It shouldn't be a problem, if the prediction was wrong, the system discards any setup it may have done and you are back to the original x+y seconds. If the prediction was right, you save the x seconds and get an improvement.

Misleading

[SwordsmanLuke]

They're not really predicting the player's actions here, are they. In their game, you're just timing a jump. Given that the swinging vines are only near enough to jump between some part of the time, they're just predicting which opportunity the player will try for the jump on. Since they're measuring skin conductivity and the player has usually decided to jump some time before the timing is perfect, these results don't really surprise me. All they've done is train a neural network to measure stress leve

Re:

[jythie]

Which even this has some interesting potential.

It reminds me of people who practice martial arts and learn to read when/how someone is going to attack by looking at which muscles are tensing and such.

And while I can see why people are mocking the entire 'aircraft' potential, any system that could potentially reduce reaction time by reading what someone will do faster then their body can do it could have application.

Re:

[ComaVN]

It reminds me of people who practice martial arts and learn to read when/how someone is going to attack by looking at which muscles are tensing and such.

I'll remember that, next time I'm fighting naked ninja's.

Re:

[zippthorne]

To hide the fact that they're really pirates moonlighting as ninjas?

Re:

[SwordsmanLuke]

Actually, I teach European Broadsword fighting and watching the major muscle groups does help a great deal, so I can see where you're coming from with this. But it only goes so far. Certainly, you can learn to predict what your opponent is about to do by watching muscle groups - but it's not a guarantee. You *can* deliberately mess with such a predictive system quite easily, once you know which muscle groups tense properly, of course. The more common scenario, however, is when the aggressor simply chang

Unfortunately....

[aicrules]

Unfortunately the research was found to be flawed when they discovered that 99% of the test subjects were playing Night elf characters in World of Warcraft. This was pinpointed just prior to a potentially embarassing follow-up news release of a study showing the connection between frequency of spacebar usage and use of the phrase "come on flip already dammit!"

Re:

[Pojut]

"Let's, let's go roll Alliance on like a PvE server...would that be fun?" - teh_pwnerer

Hands-free "mental mouse"

[UbuntuDupe]

This sounds like it has applications for having a mouse you control with your mind. It seems like the way their system works is that it notices certain physiological changes in you when you are mentally preparing to make a jump. Now, if you can learn to recognize what you are thinking that activates those physiological changes, then you could intentionally make them, and then those changes could be used as the input.

Re:

[booyagrandma]

that exists already: http://en.wikipedia.org/wiki/BrainGate [wikipedia.org]

shameless plug for my alma mater: go brown!

Re:

[UbuntuDupe]

Very interesting. However, what I suggested is different in that it doesn't require an (extremely invasive and risky) brain implant, since it would monitor your brain signals indirectly through changes in your skin.

So is any kind of mental mouse already commercially available?

Re:

[booyagrandma]

i'm not aware of one. i'm not sure its really possible from skin conductance.. my extremely limited understanding of that it that it mostly fluctuates with emotion, i suppose in this case the anticipation of being about to jump. that sounds ok for a binary input, how would you do directions though?

Re:

[UbuntuDupe]

The device in the story seems to be similar to a polygraph [wikipedia.org], which typically measure three things (one of which is skin conductance) that supposedly correlate with lying. Now, regardless of whether they actually so correlate, as long as you can learn to control your mental states that alter those properties with enough precision, and enough relative independence, you should be able to control more than one binary variable.

I'm hopefully not the first person to think of this...

Re:

[FleaPlus]

Very interesting. However, what I suggested is different in that it doesn't require an (extremely invasive and risky) brain implant, since it would monitor your brain signals indirectly through changes in your skin.

Here's info on the research into non-invasive versions:

http://en.wikipedia.org/wiki/Brain-computer_interf ace#Non-invasive_BCIs [wikipedia.org]

Must be an easy game..

[Chris_Jefferson]

I play plenty of games where I'm positive I have no way of knowing if I'm going to jump 2 seconds before, unless I can somehow tell where enemies are going to move in advance.. which would be very interesting. I'd be interested to know how long in advance they can detect button presses on more 'twitch' games.

Re:Must be an easy game..

[dreddnott]

My thoughts exactly. Games of Unreal Tournament and the like are won by decisions made in 100 milliseconds. Waste 50 extra milliseconds not pulling the trigger and your brains get splattered all over by the guy sitting across from you (or the Godlike bot). If I start "flowing" enough my conscious mind doesn't even get in the way (hence the 'twitch' appellation), that's how I come out on top. Economy of thought. Doesn't work as well in CTF with team damage on...

Re:

[ucblockhead]

You might (and probably do) make the decision in 100 milliseconds, but it takes longer than 100 milliseconds for the result of that decision to travel from you brain to your fingers.

Human reaction time is about 1/4 of a second

[Colin Smith]

It takes much longer than that to think. Bright young things are maybe 1/5 of a second.

If I start "flowing" enough my conscious mind doesn't even get in the way (hence the 'twitch' appellation), that's how I come out on top. Economy of thought.

That's how the martial arts masters work. No thought involved. And it's why even if you know and understand the techniques intellectually, you'll still get your arse handed to you by a street thug if you haven't practised several times a week, for years.

Interestingly the flinch reaction is actually faster than the regular reaction time, about twice as fast, 0.15s but the response is hardwired into the brain, you can't do

Re:

[MikeFM]

I used to have a text-based RPG (ie a MUD) that could predict major events for players or groups of players. It was sort of cool. We had an Oracle that you could visit and it'd predict battles, deaths, etc. We'd record when things happened and then let our program find patterns in it. Of course it wasn't 100% accurate (and couldn't say who'd do what in the next few seconds) but it worked well enough to give you an idea of things likely to happen soon. The fuzziness played along well with the theme of a fort

Re:

[fractoid]

I'd be interested to know how long in advance they can detect button presses on more 'twitch' games.

And also how accurate their prediction of time is. "You're going to jump sometime in the next 20 seconds" isn't very helpful when split-second timing is required.

Re:Must be an easy game..

[Wog]

I like to call that a Wide Prophet Margin.

Re:

[TriezGamer]

This is one of those times where I really wish I had mod points, so I'm just going to go with the "I LOLed" response. Literally.

Re:

[Blimey85]

You bring up an interesting point but it's actually rather simple. What you think you are seeing already happened approximately 2 seconds before you think it happened. It's like this: light normally travels pretty fast but for video games it's slowed down to approximately the speed of sound. That is why what you see and hear seem to go together. Instead of the visuals traveling at the typical speed of light they are traveling at only 60 frames per second, for the average user. For some it's a bit faster and

Re:

[Aladrin]

That was my thought exactly... Unless I can predict what's happening on-screen for more than a few seconds, there's no way -I- could know I'm going to jump in 2 seconds, let alone the computer detect that I know it.

At any rate, this is just 1 more study that says our bodies give off unconscious signals... -yawn-

Matrix

[jadin]

Oracle: I'd ask you to sit down, but, you're not going to anyway. And don't worry about the vase.
Neo: What vase?
Oracle: That vase.
Neo: I'm sorry--
Oracle: I said don't worry about it. I'll get one of my kids to fix it.
Neo: How did you know?
Oracle: Ohh, what's really going to bake your noodle later on is, would you still have broken it if I hadn't said anything?

If you know that the game knows you're going to jump, will you still do it?

Re:

[Applekid]

What if the game knows that you know that the game knows you're going to jump, and might not do it?

Ow my head.

Re:

[kontos]

Inconceivable!

Re:Matrix

[UbuntuDupe]

From what I understand, it correlates the changes in physiological metrics of you, with the times you jump. So, if you intentionally prepare to jump, like you normally do, but then deliberately hold back at the last second, opposite from how you acted while it was being trained, you can fool it. You'll show the signs of jumping, but then "change your pattern" so its guess is wrong.

Oh, and it probably would have been more relevant to make a cutesy reference to Minority Report. ;-)

Re:

[Dorceon]

But you would have to prepare to jump without preparing to deliberately hold back.

Re:

[master_p]

And 2 seconds is enough to make a conscious decision...

That mask..!

[Anonymous Coward]

It's ok; just switch controller ports.

Re:

[DesireCampbell]

Oh come on! That was gold! Where are my mod points when I need them?

Bring on the Frustration Game

[Nymz]

I'm certain many of us could out-frustrate it's edge. Choose Game Difficulty:

• Chewtoy
• Normal
• Hard
• Frustration!!!

Re:

[scribblej]

There was a game back in the day - Suspended, by Infocom. It remains one of the scariest games I have ever played.

Anyhow, it had difficulty options, and the hardest difficulty setting was "Impossible."

Turns out it /actually/ is impossible -- if you select it, then three moves into the game the sun goes supernova, destroying everything.

Doesn't sound all that impressive. I suspect ...

[shaobohou]

all they have achieved is predicting how far away the game has placed the next jump, based on indirect observation. You can probably achieved the same result with any statistical prediction model. If the player can jump freely without having to catch the branches then I doubt their system would be that accurate.

Re:

[ALeavitt]

It's one thing not to read the article, but you clearly didn't even read the summary. I have my doubts as to whether you will even read this reply to your post.

Re:Doesn't sound all that impressive. I suspect .

[shaobohou]

fine, so the game doesn't quite work the way i thought it did, but it is not that clearly described in the article (not at all in the summary). I still think that it is just a simple prediction model, anyone can do it if they had access to the right equipment. I stand by my original comment, what is your point?

So... it just plays itself?

[Anonymous Coward]

Obligatory penny-arcade reference. http://www.penny-arcade.com/comic/2006/09/07/ [penny-arcade.com]

Subsequent Releases

[Babbster]

One can only hope that version 2 will play a Kris Kross clip every time it makes it's prediction.

2 seconds?

[dosboot]

2 seconds seems like a really long time. The player isn't reacting here, they are timing. If one makes a very quick reaction, which happens in less than one second, then your fingers can't possibly know 2 seconds in advance.

Re:

[absorbr]

I don't get it either. A lot can happen in 2 seconds.

Re:

[Feanturi]

2 seconds seems like a really long time. The player isn't reacting here, they are timing.

Yeah that's my thought as well. Think of a platform scroller like Mario or something like that, where you are running along and can see there's a jump ahead you'll need to make. That's the sort of thing this system is predicting. If you have to jump because something is suddenly flying at you fast, I doubt it would pick up that blip in your skin in time for its "frustration" reaction to have relevance.

I'm also not sure

Re:

[hajus]

I read about this work a couple years ago. This isn't really about reaction time and won't be able to handle the twitch stuff. It's meant to handle situations in which you make a concious decision to do something rather than reacting to something. For example in an FPS game, if you see some guy jump in front of you and start shooting at it, this prediction method is useless for a reaction. But if you have been standing still for 15 seconds in front of a door and suddenly decide to move forward, that's w

That's smart.

[CODiNE]

They say the technique could ultimately be used to make aircraft controls that respond more quickly to a pilot's actions.

*Pilot cruising around Washington DC on anti-Terrorist patrols*
(Thinking to himself)
"Oh wow, I'm like... right over the White House now, it would like totally suck if I accidentally shot a"
*Missle launches at White House*
"Oh crap!! Do NOT think about launching missles!!"
*Missles launch at various historical monuments*
"Craaaaaap!! My Sarge is gonna KILL M"
*Plane explodes*

Re:

[revengebomber]

Uh oh. Er, uh, Allah made me do it.

Human consciousness

[Menoyoda]

This sounds like the experiments done by Benjamin Libet [wikipedia.org], in which he found that actions are initiated by the brain before one consciously decides to perform them.

Re:

[aicrules]

Which is why, if you're not careful, or if you're not fully in control of your faculties (drunk) and you think about bitchslapping someone, you will bitchslap them without consciously trying to.

Re:

[salec]

You mean, like, all our "selfness" does is after-the-act damage control, assembling a good story, making up excuses?

Danger in the cockpit

[Nanidin]

Wouldn't this be dangerous in an airplane? 2 seconds is a pretty big window of time between when a pilot might decide to do something and when he actually does it. During those two seconds, a lot could happen - unexpected turbulence, new weather data, a gust of wind on the runway, etc. If the computer acts on the command two seconds before the pilot intends for it to happen, bad things could happen.

Re:Danger in the cockpit

[mlts]

Another issue is if the AI makes a wrong guess, assuming the pilot wants to "zig", when the pilot wants to "zig-zag" -- doing a multi-stage action. An example would be a driver swerving left and right abruptly to avoid a zombie on the highway.

I can perhaps see a car or plane's AI going into a "ready" state, perhaps doing a fast check of control systems in preparation for some action of the driver/pilot.

Re:

[geminidomino]

Another issue is if the AI makes a wrong guess, assuming the pilot wants to "zig", when the pilot wants to "zig-zag" -- doing a multi-stage action. An example would be a driver swerving left and right abruptly to avoid a zombie on the highway.

Zombies kill more motorists each year than cancer, heart attacks, and murder combined.

Please, Don't be a Zombie...

Setup

[PhoenixOne]

It wouldn't act on the command before the pilot commanded, but it could do some "setup" (light up the buttons the pilot will want to use to make them easier to find, bring up data relating to the task, etc.). Some high-end cars have similar features (they tense up the seat-belt if you hit the breaks hard, and get the airbag ready).

Good computer tools do something like this already, just in a more primitive way. Imagine that when you are almost finished writing an email, the computer could bring up the sen

Meanwhile at a gaming tournament...

[jadin]

"Sir, you're not allowed to place sensors on your opponents."

Old news

[fred9653]

I saw such a programm a few years ago:
http://www.phi-t.de/mousegame/ [phi-t.de]
It uses a neural net to predict your next move.

It's bad enough

[gandhi_2]

...when MS word and excel keep trying to guess what I mean.

Health Warning

[kramulous]

News Anchor 1: The porn industry's new, forced feedback game, 'Long Time Coming', has reportedly achieved sold out status across the globe. The church continue to voice their opposition.

News Anchor 2: In other news, employers have noticed a sharp increase in the number of 'sickies' by employees. This absence from work is reportedly costing billions, a phenomenon being experienced worldwide. The cause of the epidemic is not yet known.

How to beat the bots

[adona1]

The researchers use neural networks to analyse several type of biofeedback signal -- heart rate, EEG and skin conductance

That should be easy to get around, just drink a helluva lot of Jolt & coffee, then play a very cautious strategy...

Re:

[Tzarius]

[i]then play a very cautious strategy[/i] I'd say that would give the sensors more time to pick up your intentions. Going hell for leather without thinking things through would definitely strain a system that needs a couple of seconds to pick up a change...

This won't be used in aviation

[AmiMoJo]

An aircraft that second-guesses the pilot? Sounds like a brilliant idea.

Frustration Games

[CodyRazor]

But it could also be used to create so-called 'frustration games'

That sounds like fun.

Ha Ha

[ElleyKitten]

Haha, computer, you will not catch me, I'm a button-masher! If I don't know what I'm doing until after I've done it, you'll never figure it out!

Seriously, while this might work for games like Mario (Hmm. That's a hole. Maybe I should jump?) I don't see it working on anything fast-paced.

premature ejection

[fuliginous]

I guess it will become an advantage to jump before you intend some time. Or any other action before you intend to.