Advances in Artificial Intelligence for next generation, & advanced AIs in videogames

[Deleted member 11852]

At the same time, that could be frustrating for gamers who like to snipe from afar and don't want to be forced into a different play style. Almost as if progressive challenges needs to be an option, leaving the option to cheese enemies for the more casual player.

There are definitely downsides. This is where difficulty settings are sorely under utilised frequently being a bullet-sponge scaling setting.

 

[iroboto]

Units are obviously not magically spawning body armour, the changes happen between missions so as to give the appearance of the enemy adapting to the player's tactics. Civ doesn't make much use of fixed data trees for anything much, it's why the AI is infuriatingly unpredictable when it ought to be based on the core traits of any given civilisation.

Pretty sure it's the same AI running decision trees but with biases to make them act differently and to somewhat adhere to what that civ would do.

All civilization biases indicated here: http://civdata.com

 

[Deleted member 11852]

Pretty sure it's the same AI running decision trees but with biases to make them act differently and to somewhat adhere to what that civ would do. All civilization biases indicated here: http://civdata.com

Decisions trees are very different, these are merely variables that impact AI behaviour. You use decision trees to simplify AI choice when you wish to avoid waiting 90 seconds between player turns. That's what Civ actually does.

 

[iroboto]

Decisions trees are very different, these are merely variables that impact AI behaviour. You use decision trees to simplify AI choice when you wish to avoid waiting 90 seconds between player turns. That's what Civ actually does.

Imo i don’t think they are all that different.

We are likely looking at a combination of behaviour trees with decision trees and likely some form of state machines at the lowest level.

 

[Deleted member 11852]

Imo i don’t think they are all that different.

They are completely different. Exactly how much AI code have you written?

 

[iroboto]

They are completely different. Exactly how much AI code have you written?

Little, most of it in university when we were studying the different forms of decision making. I don't recall it being that unrelated. FSM had failings of being only operating in 1 state. Decision trees could be rather long IIRC.

I don't see why you wouldn't use them in conjunction with each other. They aim to serve the same purpose but have different pros and cons.

 

[iroboto]

Decisions trees are very different, these are merely variables that impact AI behaviour. You use decision trees to simplify AI choice when you wish to avoid waiting 90 seconds between player turns. That's what Civ actually does.

just curious but are you referring to a scoring system when you mean decision tree?
because when I read 'tree' I see a diagram that looks like a state machine.

 

[Deleted member 11852]

Little, most of it in university when we were studying the different forms of decision making.

Figures. Decision trees are something you consider falling back on when there are a limited limited number of variables, clear critical path outcomes and which are also computationally intensive.

So not Civilisation, which has way too many variables.

 

[iroboto]

Figures. Decision trees are something you consider falling back on when there are a limited limited number of variables, clear critical path outcomes and which are also computationally intensive.

So not Civilisation, which has way too many variables.

oh I read your comment wrong, I thought you said it makes use of decision trees.

oh right. Yea I guess it wouldn't. It would be computationally too expensive I think. Really old way of doing AI. Likely they are going with some scoring system perhaps something that lets the AI do fuzzy logic I guess. Not a clear defined path as you said.

 

[Deleted member 11852]

Likely they are going with some scoring system perhaps something that lets the AI do fuzzy logic I guess.

Nope. "Fuzzy logic"? Just stop posting on this. Seriously.

 

[iroboto]

Nope. "Fuzzy logic"? Just stop posting on this. Seriously.

Why? And don’t say it’s because I’m not qualified. Barely anyone here is qualified to talk about most of our topics.

It’s a great topic to discuss that we don’t do enough on this forum. If you know exactly how it works I’d like to know. I have 1 game out. It was as basic as AI comes.

Anyway, yea fuzzy logic.

Like defining a action and assigning points to each action. You let the scoring system define what the action the computer will take. It allows for some flexibility in the decisions the AI could take, it’s not anywhere close as rigid as tree and you can quickly tabulate multiple actions quickly to see which one gains priority.

 

[Deleted member 11852]

Why? And don’t say it’s because I’m not qualified. Barely anyone here is qualified to talk about most of our topics.

You seem to have a little bit of knowledge of some dated approaches to AI techniques and are posting as though it's fact. That got my goat. I should have been more diplomatic but I wasn't.

Let's move on. If you are genuinely interested, go read some books on AI. I'd strongly suggest starting with books written in the 1990s because these explored techniques which were considered as insightful at the time but now, with the application of experience, are recognised as flawed, but cover techniques can be employed to solve non-AI problems. Part of understanding any discipline is not just knowing how to do something but also knowing how not to do something, and why.

 

[iroboto]

You seem to have a little bit of knowledge of some dated approaches to AI techniques and are posting as though it's fact. That got my goat. I should have been more diplomatic but I wasn't.

Let's move on. If you are genuinely interested, go read some books on AI. I'd strongly suggest starting with books written in the 1990s because these explored techniques which were considered as insightful at the time but now, with the application of experience, are recognised as flawed, but cover techniques can be employed to solve non-AI problems. Part of understanding any discipline is not just knowing how to do something but also knowing how not to do something, and why.

AI comes in two flavours though. Game AI and no game AI (you're asking me to read on game AI? or industry AI?). Those are not the same as far as my understanding of things and I have a feeling you're referring to non game AI if that's your industry practice.

edit: MCST! interesting, has some similar feels to A*

 

[Deleted member 11852]

AI comes in two flavours though.

 

[iroboto]

Lol. Okay. Let me rephrase. When you say Is your goat, do you mean it’s what you do professionally ? Or just the biggest part of your interest when it comes to games?

IIRC you said you did Chem? But you are now in data sciences or related field ?

 

[Deleted member 11852]

Lol. Okay. Let me rephrase. When you say Is your goat, do you mean it’s what you do professionally ? Or just the biggest part of your interest when it comes to games?

Both!

IIRC you said you did Chem?

Umm.. what?

But you are now in data sciences or related field ?

We don't consider it 'science' but close enough. AI without parseable data is useless though, so they're closely related. I'd definitely recommend reading some AI literature because while a lot of delivers nothing, it covers processes and techniques that are invaluable for solving non-AI problems. When we recruit new programmers they are required to read a varied amount of research material not directly related to their specific field. Generally people stick around and months/years later they'll come back with some problem they solved predicated on some technique discussed in a paper on AI.

 

[Tkumpathenurple]

Can you recommend any books in particular? Failing that, any authors or publishers?

 

[Deleted member 11852]

Can you recommend any books in particular? Failing that, any authors or publishers?

Anything AI-related by Jack Copeland.

 

[iroboto]

@DSoup can you elaborate on the solving non AI related problems using AI?

What exactly does your conpany do? And how would it differ from say a data science/BI shop?

 

[Shifty Geezer]

AI comes in two flavours though.

I think that's a little unfair. Games can use all sorts of cheats and hacks to get around an AI requirement, whereas a true AI solution needs to be complete. So it's akin to saying there's two types of computer graphics - realtime and CGI. Both are based on the same fundamental principles explored in depth in CGI, but realtime uses whatever cheats and approximations it can. Or saying there's two types of car - racing and consumer. Both are based on the same engineering principles explored to their zenith in performance racing cars, but consumer level cars will require more grounded, robust solutions.

Gaming is tolerant to shortcuts and hacky fudge-factors to save cycles that AI problem solving isn't, if you're being true to it.

@DSoup can you elaborate on the solving non AI related problems using AI?

DSoup didn't say 'solved using A'I, but solved using principles ("techniques") learnt from studying AI. I guess he means getting a different set of mental models to view a problem from and being able to express it as Big Data or relationship models or a branchy tree structure or a car analogy...

What exactly does your conpany do?

IIRC he works for the British Government and can't really talk about it.