But realistically, the only reason we don't exploit other planets or antarctica is that its too difficult compared to expoiting something on our own territory. Believe me, if they discover massive amounts of oil in antarctica, and an easy way to extract and ship it, we'll be doing it as fast as possible.
Yeah clearly that got people into disbelief mode when my point was really that the work wasn't going on in some niche game no one had heard of or played.
The sources talked to them on condition of anonymity. Believe me, as a developer, you don't want MS/Sony to come take your development platform away for violating your nondisclosure agreement.
Well, maybe I don't know enough about pennies and light sockets, but wouldn't that have about a 10% chance of killing them, and about a 30% chance of permanent scarring that will impact their long term earning potential and ability to give you n-children?
You're not a cannibal if the 'person' you're eating isn't human, but rather god.
Re:So Call Me Old And Cranky
on
Effective C#
·
· Score: 1
On the other hand, if you're new enough to programming to need to read this book, perhaps you're only middling competent in 2 languages which are likely to be java and c# and trying to read and understand the techniques being presented in a language you don't know is just heaping on unnecessary difficulty.
What's missing is the genetic algorithms built neural network deciding which action to take.
It looks more like MonsterAction action = myNeuralNet->pickAction(inputs); doAction(action) ;... myNeuralNet->feedBack(feedBackInputs);
where the first set of inputs is a bunch of parameters picked from the environment (direction and distance to each player, health % each player, class of each player, etc), and the second set of inputs is some information about the success of the doAction (number of players hurt, how much, how much I got hurt etc).
The second set of inputs retrains the neural network a little bit, and theoretically the doAction gets a little smarter over time.
The problem of course, is that it takes a pretty big neural net to play any game very smartly, and a lot of training time. That's why most games only bother with pre trained neural nets (at least among people I've talked to) and in my case I have dynamic training, but its mostly a toy as there isn't enough time to learn anything meaningful, and the neural net can't be big enough for performance reasons anyway. To really play a game like D2 effectively (at a guess) the AI would need probably a few hundred inputs, many thousands of nodes, and in the tens of thousands of connections. I could happily use up several gigabytes of memory and roughly 100 p4s worth of cpu just to do a competent learning AI.
It'll give you an idea of the computing resources necessary just to play something like checkers effectively.
My original point in posting on this topic was just to note that game devs aren't ignorant of the computer science behind AI, and that in fact we've been developing and using 'real' AI techniques for a long time. It's really just that they don't work very well in the context of such complex game environments that you don't really see more talk about the subject.
Thanks, that was easily the most fun thing I did for the game.:-) The script was great. Stuff like: Now tell a funny joke in cow: Moo moo moo moo, moo moo: Moo!
There's obviously vast room for improvement in the field of AI in games. My point was only that people in games do keep up with what's going on in the research community, and in some cases are actually inventing things that dont get published.
I realized this morning that to some extent I actually can do this.
Have a look at:
" Analysis of State Exposure Control to Prevent Cheating in Online Games" Kang Li, Shanshan Ding, Doug McCreary , and Steve Webb. In ACM Nossdav 2004, Kinsale, County Cork, Ireland.
Actually the point of that was to emphasize that doing 'real' AI in games is not outside the mainstream. People who make games love to learn new technologies (you have to) from sound to graphics, to AI techniques. It's not like we're sitting there ignoring what is going on in the research field.
Exactly. This stuff is not rocket science. There's plenty of documentation out there, I wouldn't even call it hard. A neural net learning AI is trivial to implement, the problem you run into is a training time vs complexity vs capability trade off. To make a really effective neural net AI for something as complex as a modern video games takes a huge neural net with a lot of inputs. I could easily see using gigabytes of memory and the processor power of about 100 p4s to make a better AI for action games like D2, but that level of AI obviously won't be seen for another few years. In the meantime we make do with poor substitutes that just obviously aren't as effective. But it's not like the original poster implies that we aren't trying this stuff out, or even using it in more limited ways.
I implemented learning AI in a couple of popular video games (including at least one multi million unit PC title) more than 5 years ago, and I'm pretty confident I wasn't breaking any new ground.
Sadly, I'm familiar with my neighbors, and about 1 in 3 is in fact a bloodthirsty robot (or unthinking person as you prefer). And as it happens, about 1 in 5 or so of those seems to be in the military, yet amazingly none from the other category.
They'd make plenty of money based on economy of scale. It's just that they'd sell it at approximately 1/100th or less of the current street value. Still plenty of room to make money, just like any common vegetable people buy at the store rather than grow themselves.
Slashdot Mirror
50 miles high in dollar bills. 2700 feet in hundreds.
There are legal restrictions based on the space treaty:h tm
http://www.oosa.unvienna.org/SpaceLaw/outersptxt.
But realistically, the only reason we don't exploit other planets or antarctica is that its too difficult compared to expoiting something on our own territory. Believe me, if they discover massive amounts of oil in antarctica, and an easy way to extract and ship it, we'll be doing it as fast as possible.
Yeah clearly that got people into disbelief mode when my point was really that the work wasn't going on in some niche game no one had heard of or played.
The sources talked to them on condition of anonymity. Believe me, as a developer, you don't want MS/Sony to come take your development platform away for violating your nondisclosure agreement.
Well, maybe I don't know enough about pennies and light sockets, but wouldn't that have about a 10% chance of killing them, and about a 30% chance of permanent scarring that will impact their long term earning potential and ability to give you n-children?
While bronze swords have fallen by the way side, our (US) economy would suffer a serious disruption if the railrounds were to vanish.
You're not a cannibal if the 'person' you're eating isn't human, but rather god.
On the other hand, if you're new enough to programming to need to read this book, perhaps you're only middling competent in 2 languages which are likely to be java and c# and trying to read and understand the techniques being presented in a language you don't know is just heaping on unnecessary difficulty.
What's missing is the genetic algorithms built neural network deciding which action to take.
) ; ...
l lapillaAndDBFogelText.pdf
It looks more like
MonsterAction action = myNeuralNet->pickAction(inputs);
doAction(action
myNeuralNet->feedBack(feedBackInputs);
where the first set of inputs is a bunch of parameters picked from the environment (direction and distance to each player, health % each player, class of each player, etc), and the second set of inputs is some information about the success of the doAction (number of players hurt, how much, how much I got hurt etc).
The second set of inputs retrains the neural network a little bit, and theoretically the doAction gets a little smarter over time.
The problem of course, is that it takes a pretty big neural net to play any game very smartly, and a lot of training time. That's why most games only bother with pre trained neural nets (at least among people I've talked to) and in my case I have dynamic training, but its mostly a toy as there isn't enough time to learn anything meaningful, and the neural net can't be big enough for performance reasons anyway. To really play a game like D2 effectively (at a guess) the AI would need probably a few hundred inputs, many thousands of nodes, and in the tens of thousands of connections. I could happily use up several gigabytes of memory and roughly 100 p4s worth of cpu just to do a competent learning AI.
For more information, have a look at this:
http://www.comp.nus.edu.sg/~cs6211/papers/TNNKChe
It'll give you an idea of the computing resources necessary just to play something like checkers effectively.
My original point in posting on this topic was just to note that game devs aren't ignorant of the computer science behind AI, and that in fact we've been developing and using 'real' AI techniques for a long time. It's really just that they don't work very well in the context of such complex game environments that you don't really see more talk about the subject.
Thanks, that was easily the most fun thing I did for the game. :-)
The script was great. Stuff like: Now tell a funny joke in cow:
Moo moo moo moo, moo moo: Moo!
See my reply to jacknm and email me if you're curious enough.
There's obviously vast room for improvement in the field of AI in games. My point was only that people in games do keep up with what's going on in the research community, and in some cases are actually inventing things that dont get published.
I realized this morning that to some extent I actually can do this.
f
... well, I think it will be pretty hard to convince you.
Have a look at:
" Analysis of State Exposure Control to Prevent Cheating in Online Games"
Kang Li, Shanshan Ding, Doug McCreary , and Steve Webb. In ACM Nossdav 2004, Kinsale, County Cork, Ireland.
http://www.cs.uga.edu/~kangli/src/nossdav_2004.pd
This is some of my research work at University of Georgia.
Then email my slashdot user name @uga.edu to receive final convincing.
If that's not sufficient evidence
Actually the point of that was to emphasize that doing 'real' AI in games is not outside the mainstream. People who make games love to learn new technologies (you have to) from sound to graphics, to AI techniques. It's not like we're sitting there ignoring what is going on in the research field.
I'm the cow king.
Exactly. This stuff is not rocket science. There's plenty of documentation out there, I wouldn't even call it hard. A neural net learning AI is trivial to implement, the problem you run into is a training time vs complexity vs capability trade off. To make a really effective neural net AI for something as complex as a modern video games takes a huge neural net with a lot of inputs. I could easily see using gigabytes of memory and the processor power of about 100 p4s to make a better AI for action games like D2, but that level of AI obviously won't be seen for another few years. In the meantime we make do with poor substitutes that just obviously aren't as effective. But it's not like the original poster implies that we aren't trying this stuff out, or even using it in more limited ways.
Now that's a challenge! Seriously, how would you establish yourself given a similar demand?
I have one solution, but i'd like to hear yours, maybe yours is better than mine.
So look me up in the credits already!
If you can suggest how better to prove that I worked on it, let me know.
Diablo II, I'm Doug M.
I implemented learning AI in a couple of popular video games (including at least one multi million unit PC title) more than 5 years ago, and I'm pretty confident I wasn't breaking any new ground.
Sadly, I'm familiar with my neighbors, and about 1 in 3 is in fact a bloodthirsty robot (or unthinking person as you prefer). And as it happens, about 1 in 5 or so of those seems to be in the military, yet amazingly none from the other category.
It's not necessarily clear that b results: much of the interest in pot exists because of the illegality. You have to consider:
How many people who would really like to smoke pot are not doing so because of legal issues?
How many people who currently do smoke pot do it because it is cool to do something illegal?
They'd make plenty of money based on economy of scale. It's just that they'd sell it at approximately 1/100th or less of the current street value. Still plenty of room to make money, just like any common vegetable people buy at the store rather than grow themselves.
Presumably it's as free including international as the best available land service can be.
Nahh, it can be a lot of fun, especially if you're obsessive compulsive.