Not Really OT: AI

[herwin]

After spending a week cleaning up my research wiki, I will be disappearing for two weeks--without my computer--hopefully finding that AE has been released (and is working) upon my return. There are some pages on my wiki that are relevant to game AI.

There are two kinds of purposeful behaviour that we study--habitual and goal-directed behaviour. There is a problem with both that researchers find difficult--the state space explosion that occurs when you try to learn appropriate behaviour in a rich environment, but we can often solve it and currently have a model for learning habitual behaviour called the actor-critic model, where behaviour is learned by caching the values of actions in previous experience. Goal-directed behaviour involves planning: predicting the outcomes of actions and then back-propagating the resulting rewards to the current set of actions rather than learning by trial and error. It's much harder to programme. If you examine my page on prediction, you'll discover that most researchers currently assume away the difficulty and often come up with an actor-critic model for goal-directed behaviour. Unfortunately for that approach, in goal-directed behaviour, animals change their behaviour when their perception of the final reward changes, and an actor-critic model cannot do that. However, there was a Nature paper about a month ago that suggests that the way we model a plan is not by simulating it, but rather by graphing it. The hippocampus then stores the complete graph, and uses that to back-propagate rewards.

Why discuss this stuff here?

Later this summer, I will be writing a research grant proposal to explore this. It may lead to a better way for representing plans in games like WitP AE.

[Terminus]

[8|]

[51st Highland Div]

Oh dear...looks like not only do i need a copy of WitP:AE for dummies but AI for dummies book also to work that out..[:D]

[AW1Steve]

Or you could simply study the insanity and instability of WITP players and their reaction to delay of their gratification? [&:]

[Tophat1815]

ORIGINAL: AW1Steve

Or you could simply study the insanity and instability of WITP players and their reaction to delay of their gratification? [&:]

I'm thinking there is merit to this statement.[;)]

[Terminus]

ORIGINAL: 51st Highland Div

Oh dear...looks like not only do i need a copy of WitP:AE for dummies but AI for dummies book also to work that out..[:D]

Not really, no... Herwin is just spouting his usually wordage; nothing to worry about at all, and certainly in no way relevant to the game, not matter what he thinks.

[wdolson]

Herwin is an academic. 

Bill

[jmscho]

ORIGINAL: herwin

It may lead to a better way for representing plans in games like WitP AE.

So it really is off topic [:-], especially as we expect release very soon. Unless of course the research comes to any practical conclusions that can be implemented - maybe ready in time for WitP version 46.[:)]

[Tophat1815]

I think you should redesign the game. Just do it now,whats two more years.

[herwin]

ORIGINAL: jmscho

ORIGINAL: herwin

It may lead to a better way for representing plans in games like WitP AE.

So it really is off topic [:-], especially as we expect release very soon. Unless of course the research comes to any practical conclusions that can be implemented - maybe ready in time for WitP version 46.[:)]

It appears the way the brain structures plans is similar to the movement planning documents staff officers generate. So if you want AI, you may need to automate the generation and analysis of similar things.

[bstarr]

ORIGINAL: herwin

After spending a week cleaning up my research wiki, I will be disappearing for two weeks--without my computer--hopefully finding that AE has been released (and is working) upon my return. There are some pages on my wiki that are relevant to game AI.

There are two kinds of purposeful behaviour that we study--habitual and goal-directed behaviour. There is a problem with both that researchers find difficult--the state space explosion that occurs when you try to learn appropriate behaviour in a rich environment, but we can often solve it and currently have a model for learning habitual behaviour called the actor-critic model, where behaviour is learned by caching the values of actions in previous experience. Goal-directed behaviour involves planning: predicting the outcomes of actions and then back-propagating the resulting rewards to the current set of actions rather than learning by trial and error. It's much harder to programme. If you examine my page on prediction, you'll discover that most researchers currently assume away the difficulty and often come up with an actor-critic model for goal-directed behaviour. Unfortunately for that approach, in goal-directed behaviour, animals change their behaviour when their perception of the final reward changes, and an actor-critic model cannot do that. However, there was a Nature paper about a month ago that suggests that the way we model a plan is not by simulating it, but rather by graphing it. The hippocampus then stores the complete graph, and uses that to back-propagate rewards.

Why discuss this stuff here?

Later this summer, I will be writing a research grant proposal to explore this. It may lead to a better way for representing plans in games like WitP AE.

I felt the need to translate the above into 'Redneck Texan':

I’m gonna spend me a week cleanin' my garage, then I’m gonna go on a drinkin' binge for a week or two. When I get back I’ll be a-lookin’ for that there AE game.

I got me a couple of things I done noticed that probably ain't got nothin to do with AE but I’m gonna say it here anyhow. When women folks are a-drinkin’ real heavy-like they tend to act in two ways – horny or funny. Now when they’re actin’ all horny good ol' boys tend to get all excited and try to load’em up and take’em down some dirt road, but it ain’t that easy. Sometimes they act all horny, but they ain’t so we have to buy'em some drinks and figure out if they’re wantin’ to hop in the sack or if they just tryin’ to make some Bubba jealous. It ain’t always easy to figure out. Most good ol’ boys ain’t able to do it. They see some gal dancin’ all hot-like and they try to load her up; poor boys end up getting’ shot down or slapped right hard-like. Now me, I’m a cool cat. I done studied them women folk and I done figured them out – if they drinkin’ pretty drinks, they just messin’ with you; if they takin’ tequila shots, go a head and fire up the truck.

Why talk about that here?

I dunno.

(no disrespect meant; I just had to)

[Sardaukar]

[:D][:D][:D]

[stuman]

ORIGINAL: bstarr

ORIGINAL: herwin

After spending a week cleaning up my research wiki, I will be disappearing for two weeks--without my computer--hopefully finding that AE has been released (and is working) upon my return. There are some pages on my wiki that are relevant to game AI.

There are two kinds of purposeful behaviour that we study--habitual and goal-directed behaviour. There is a problem with both that researchers find difficult--the state space explosion that occurs when you try to learn appropriate behaviour in a rich environment, but we can often solve it and currently have a model for learning habitual behaviour called the actor-critic model, where behaviour is learned by caching the values of actions in previous experience. Goal-directed behaviour involves planning: predicting the outcomes of actions and then back-propagating the resulting rewards to the current set of actions rather than learning by trial and error. It's much harder to programme. If you examine my page on prediction, you'll discover that most researchers currently assume away the difficulty and often come up with an actor-critic model for goal-directed behaviour. Unfortunately for that approach, in goal-directed behaviour, animals change their behaviour when their perception of the final reward changes, and an actor-critic model cannot do that. However, there was a Nature paper about a month ago that suggests that the way we model a plan is not by simulating it, but rather by graphing it. The hippocampus then stores the complete graph, and uses that to back-propagate rewards.

Why discuss this stuff here?

Later this summer, I will be writing a research grant proposal to explore this. It may lead to a better way for representing plans in games like WitP AE.

I felt the need to translate the above into 'Redneck Texan':

I’m gonna spend me a week cleanin' my garage, then I’m gonna go on a drinkin' binge for a week or two. When I get back I’ll be a-lookin’ for that there AE game.

I got me a couple of things I done noticed that probably ain't got nothin to do with AE but I’m gonna say it here anyhow. When women folks are a-drinkin’ real heavy-like they tend to act in two ways – horny or funny. Now when they’re actin’ all horny good ol' boys tend to get all excited and try to load’em up and take’em down some dirt road, but it ain’t that easy. Sometimes they act all horny, but they ain’t so we have to buy'em some drinks and figure out if they’re wantin’ to hop in the sack or if they’re just tryin’ to make some Bubba jealous. It ain’t always easy to figure out. Most good ol’ boys ain’t able to do it. They see a gal dancin’ all hot-like and they try to load her up; poor boys end up getting’ shot down or slapped right hard-like. Now me, I’m a cool cat. I done studied them women folk and I done figured them out – if they drinkin’ pretty drinks, they just messin’ with you; if they takin’ tequila shots, go a head and fire up the truck.

Why talk about that here?

I dunno.

(no disrespect meant; I just had to)

You know, I found your version not only easy to understand, but I can actually attest to everything you just wrote. Not sure if that is good, or bad.

[stuman]

Bt the way bsatrr, I think most of us native Tennesseans have a sort of natural affinity to you Texans. Seeing as how we sent some of our best to go help you found that Republic of yours [;)]

[bstarr]

ORIGINAL: stuman

Bt the way bsatrr, I think most of us native Tennesseans have a sort of natural affinity to you Texans. Seeing as how we sent some of our best to go help you found that Republic of yours [;)]

Sure did. btw, my father's side of the family has roots in Tennessee. I used to have two greataunts who lived in Chattanooga - my accent is about as rural/hick as you can get, not pretty at all, but they had this southern belle accent that was pure music. I loved just to hear them talk.

[wdolson]

When my SO drinks, she just gets more surreal.  She has one of the biggest vocabularies of anybody I've ever known.  When she's been drinking, you need a Webster's Unabridged to understand what she's saying.  If she drinks any more she just goes to sleep.  Fortunately she rarely drinks.

Bill

[Reg]

ORIGINAL: herwin

Later this summer, I will be writing a research grant proposal to explore this. It may lead to a better way for representing plans in games like WitP AE.

Now if you succeed in getting a research grant to implement one of these experimental AI algorithms on the AE baseline, I would be REALLY impressed. [X(] Heck, if there is development money involved I'm sure the programmers even would help you implement it and you would have the fans (especially me) queueing up to play the aI (as opposed the Ai). We'll all be good little lab rats... honest!! [:D]

However, on a practical level it ain't ever likely to happen...... unfortunately. [:(] [:(]

[TOMLABEL]

ORIGINAL: wdolson

When my SO drinks, she just gets more surreal.  She has one of the biggest vocabularies of anybody I've ever known.  When she's been drinking, you need a Webster's Unabridged to understand what she's saying.  If she drinks any more she just goes to sleep.  Fortunately she rarely drinks.

Bill

Too funny!!!![:D] That fits my SO to a T.

But LORD help us if we have a difference of opinion during that period!! It's sad, but I'm just left with my mouth hanging open...[:D] Thank goodness for the 'sleep' effect!

TOMLABEL

[pompack]

ORIGINAL: bstarr

ORIGINAL: stuman

Bt the way bsatrr, I think most of us native Tennesseans have a sort of natural affinity to you Texans. Seeing as how we sent some of our best to go help you found that Republic of yours [;)]

Sure did. btw, my father's side of the family has roots in Tennessee. I used to have two greataunts who lived in Chattanooga - my accent is about as rural/hick as you can get, not pretty at all, but they had this southern belle accent that was pure music. I loved just to hear them talk.

Many of us had roots in Tennessee, but (at least according to family legend) they were pulled up right fast now because of the problem with the horse. Or the woman. Depends on who is telling the story. Of course, the best version is that great-great-...grandda had to take the horse to get across the river ahead of the woman's brother. [:)]

[stuman]

ORIGINAL: pompack

ORIGINAL: bstarr

ORIGINAL: stuman

Bt the way bsatrr, I think most of us native Tennesseans have a sort of natural affinity to you Texans. Seeing as how we sent some of our best to go help you found that Republic of yours [;)]

Sure did. btw, my father's side of the family has roots in Tennessee. I used to have two greataunts who lived in Chattanooga - my accent is about as rural/hick as you can get, not pretty at all, but they had this southern belle accent that was pure music. I loved just to hear them talk.

Many of us had roots in Tennessee, but (at least according to family legend) they were pulled up right fast now because of the problem with the horse. Or the woman. Depends on who is telling the story. Of course, the best version is that great-great-...grandda had to take the horse to get across the river ahead of the woman's brother. [:)]

[:D][:D][:D]