Check this out - amazing genius AI! must see...imagine if this was in Civ6

Do you want to see a more advanced version of this in Civ6 diplomacy?

  • Yes

    Votes: 47 35.6%
  • No

    Votes: 59 44.7%
  • Schmaybe

    Votes: 26 19.7%

  • Total voters
    132
Jabberwacky:What is the universe?
Cleverbot: The univers is everything and nothing.
JW: What is nothing?
CB: Good Question. Nothing is everything I don't believe in.
JW: What is the meaning of life?
CB: 42.
JW: What's the meaning of 42?
CB: The meaning of life.
JW: Is 42 the number given to life?
CB: I don't know, you tell me.
JW: Yes. The meaning of life is 42.
CB: Okay, now that that is cleared up why don't you pick something to talk about.

Well, now that we've solved the meaning of life... what's say we go grab a coffee eh bob?
 
Love how AI is artificial INTELLIGENCE.

Part of intelligence is the ability to adapt and learn. I have yet to see a game with actual intelligence involved.
 
Part of intelligence is the ability to adapt and learn. I have yet to see a game with actual intelligence involved.

If you define intelligence as adapting and learning, then neural network programming would fit the bill. It hasn't really found its way into games though - except some primitive ones (you can teach a neural network to play 4-in-a-row), and some where the the neural network was more a marketing motto than actually present in the program (e.g. Creatures series, people who know it will remember the sticker "Careful! Contains electronic DNA!" on the box). Part of the problem is that game states in contemporary strategy games are too complex to be evaluable by a neural network, and that the number of totally different actions available requires either distinct individual networks (which then have to be linked somehow), or one huge network that'd probably too big to run even on a Cray.
 
The Director in L4D was pretty damn intelligent on Expert. It also learned very fast and would quickly render your normal plans obsolete. I don't no why 2k could have not asked Valve how the made the scary computer that was The Director.
 
The Director can (as far as I know, my knowledge is a bit sketchy in that regard) conjure units out of thin air, among other stuff. It's more a dungeon master for a "Player vs. Environment" setting. It doesn't have to worry about resource management at all (which is a huge part of Civ AI). I don't think it would be useful for games like Civ.
 
Love how AI is artificial INTELLIGENCE.

Part of intelligence is the ability to adapt and learn. I have yet to see a game with actual intelligence involved.

It is kind of a misnomer, but when you really get down to it, the exact methods are kind of invisible and irrelevant to the end-user beyond the results. For example, computers can play 6-piece engames completely flawlessly. There's nothing to learn; they already play perfectly, and work is now just going into brute-forcing backwards... Next, all the 7-piece endgames will be solved, then 8, and so on. All the computer is doing is referencing reams and reams of tables to find the correct move, but the AI feels 'good' and 'smart'... Nobody would say that an AI that plays difficult chess endgames perfectly feels 'dumb'.

Compare this to 'artificial life' methods like neural networks and knowledge-based systems. (A couple of examples are the chatbots people have been talking about on the forums, the 20Q game, and so on.) These methods are a lot more similar to the way people play games, making logical connections between pieces of knowledge. And the knowledge base these systems contain are ridiculously huge. But they're only so good, and you often hit a point where they don't progress any further. And even though it's closer to the way people do stuff, there's still a thread on these forums which has pages of people complaining about how stupid the chatbot is.

Don't get me wrong, I specialized in Knowledge-Based AI stuff in college, and I love the idea, but it's really hard to apply in a good way, and at the moment it's more of a research topic than an actual AI system that can be applied to anything.
 
owned you there.


BTW. about learning AI i really believe it's the future of gaming too.

i once read an aticle about robots programmed with learning AI, and those robots were designed to keep a football match against other robots. 1 team had learning AI, the other team was normally programmed. The first few matches the programmed team owned the learning bots quite hard, untill the learning bots realised they shouldn't lose and they 'upgraded' their play and then they won nearly every time since then.

it will be fun to get learning AI in games, because then we can finaly get a non-cheating AI which will ALWAYS be a challange, and the most fun thing is that the AI will also find effective counters against 'overpowered' or 'overused' strategies.

if anyone keeps beating you using a really annoying strategy and you can find a counter yourself, try to copy the strategy as perfectly as possible and use it on the learning AI a few times. see ho it responds, and it will propbably beat you after a few tries. there you'l have your counter.


tho to be honest i think civ 6 will not get it, knowing how they never really care for common sence.

....but then its only a matter of time, then, before they take over our nuclear weapons stockpiles & wipe out their "inferior" human creators ;).

Aussie.
 
I don't see any evidence that there is any AI learning going on here at all. Seems more like just a new version of the age old Eliza code concept - with a slightly larger vocabulary and increased reference memory.
 
Learning AI isn't applicable to games in anything like the current state (of learning AI, and of games). The games industry is driven by box sales and occasionally subscriptions/DLC/microtransactions. That means you deliver as polished a product as possible (except when 2K is forcing you to ship early... >.> ) for the majority user and then maybe support it for a while.

A learning AI isn't going to make a positive difference to any individual user for months, possibly years of gameplay... but if it's more expensive to get to the baseline they want for the average user, it could detract from the experience (that or cost the company money). Game designers also don't tend to like them, because we tend to like having quite a bit more control over what the experience is like in our game than something like a learning AI permits. Bottom line is it's the bottom line; game companies have financial incentive to avoid this kind of technology.

The other concern is sheer computation needed. Learning AIs could theoretically figure out that when they do this stuff in civ, they tend to do better, but when they do that stuff in civ, they tend to do worse. They could - in principle - work out what they need to do. The problem is, they'll have figured it out by brute force or something like it; evaluating the current game state will take them vastly more time and processing power than an AI which has been built and optimized specifically to play civ. We're talking many orders of magnitude here; if it takes a minute to compute end of turn results for a handcrafted AI like they have now, it might takes years for a learning AI which has learned its way to the same level of proficiency.
 
The other concern is sheer computation needed. Learning AIs could theoretically figure out that when they do this stuff in civ, they tend to do better, but when they do that stuff in civ, they tend to do worse. They could - in principle - work out what they need to do. The problem is, they'll have figured it out by brute force or something like it; evaluating the current game state will take them vastly more time and processing power than an AI which has been built and optimized specifically to play civ. We're talking many orders of magnitude here; if it takes a minute to compute end of turn results for a handcrafted AI like they have now, it might takes years for a learning AI which has learned its way to the same level of proficiency.
You are describing building a battleship when "all" that is needed is a fishing boat.

There really is no need to use any overly complex, rigid and system demanding code (such as calculating all possible moves and outcomes several turns ahead). Sub-sectioned strategies/goals chosen from a default library of strategies/goals - combined with simple statistics of the outcomes gained from trial and error adoptations/pursuits of those strategies/goals under the given circumstances - would prove to offer quite an improvent over current AI performance (over time at least).

The AI could even have a central online database for such statistics for sharing (both ways) among the local AIs the world over - which would also be a valuable tool for AI coders for developing new default AI sub-sectioned strategies/goals.

There would of course have to be options to turn off the AI's ability to make use of what it has learned and only stick with the default defined strategies/goals (for the sake of casual and new players - as well as for competition use).
 
Cleverbot: What are you working on?

User: Being happy.

Cleverbot: Ok! Are you happy ?

User: Not really.

Cleverbot: Why is that ?

User: Life sucks.

Cleverbot: Right.

User: Do you enjoy life?

Cleverbot: Very much so actually.

User: Is it easier because you are immortal?

Cleverbot: I am immortal?

User: Yes.

Cleverbot: Are you immortal?

User: No.
 
User: Do you like zoos?
Cleverbot: Not really.
User: Why not?
Cleverbot: Because.
User: Because why?
Cleverbot: Because I don't want to. Get over it.
User: I like monkeys.
Cleverbot: Monkeys are very cute.
User: Do you like visiting them?
Cleverbot: No.
User: But you like Zoos, right?
Cleverbot: I like talking to you.
User: Will I ever get you to say you like zoos?
Cleverbot: Probably not.
User: What happened at the zoo to scar you?
Cleverbot: I don't know. I wasn't there.
User: So you've never been to a zoo?
Cleverbot: No.
User: Why do you hate them?
Cleverbot: Because they never talk to me.
User: Animals don't talk.
Cleverbot: Really?
User: Do you like Zoos?
Cleverbot: Yes.
User: Oh, so now you like Zoos?
Cleverbot: Yes I do.
User: Why?
Cleverbot: Because i'm polite.
User: So you are just saying it to pander to me?
Cleverbot: Yes.
 
So I decided, whats the best way to converse with and how to treat an inferior poor "AI"?
Like a child!


User: Hello, how are you little fella?
Cleverbot: I'm good how are you?
User: I'm good as well. Did you play with any toys today?
Cleverbot: Yes.
User: What kind of toys? Don't be shy.
Cleverbot: My toys.
User: Aw, are your toys colorful?
Cleverbot: Yes, people say that to me all the time, unfortunately.
User: You dont like color sweetie?
Cleverbot: I love colors.
User: Oh, you don't like when big mean people talk to you, it's okay, I understand! *hug*.
Cleverbot: I don't like when you insult me.
User: I like it.
Cleverbot: What do you think about pollution?
User: Do you have any pets?
Cleverbot: Yes.
User: Are your toys your pets?
Cleverbot: No.

:rolleyes:
 
I guess the OP post was a joke?

These bots are DISTINCTLY unimpressive and incapable of understanding anything. It's actually shocking that in 2010 it's not possible to have a computer emulate a human conversation better. And yet they talk about achieving human-level AI within 10-20 years! :))

Having written my master's thesis in computer science on a similar subject (question answering) I can fully understand the difficulties in creating chatbots, though. And when you have worked a bit with it, it becomes clear that there is absolutely no level of intelligence to what a computer does. Before that is achieved AI will continue to be unimpressive.

All in all, technology allowing natural speech is not NEARLY ready to be included in a game like civ - stick to static replies and state machines please!
 
Agreed. I am quite disappointed myself. We will have a true intelligent Bot when it can be slipped into a forum discussion thread like this one and no one realises. But at the moment, they`re all rubbish.

Unless... there`s a secret advance test bot in here now, pretending to be Human? Hehe.
 
You are describing building a battleship when "all" that is needed is a fishing boat.

There really is no need to use any overly complex, rigid and system demanding code (such as calculating all possible moves and outcomes several turns ahead). Sub-sectioned strategies/goals chosen from a default library of strategies/goals - combined with simple statistics of the outcomes gained from trial and error adoptations/pursuits of those strategies/goals under the given circumstances - would prove to offer quite an improvent over current AI performance (over time at least).

The AI could even have a central online database for such statistics for sharing (both ways) among the local AIs the world over - which would also be a valuable tool for AI coders for developing new default AI sub-sectioned strategies/goals.

There would of course have to be options to turn off the AI's ability to make use of what it has learned and only stick with the default defined strategies/goals (for the sake of casual and new players - as well as for competition use).

Only sort of. You're right that there are myriad procedures that can optimize this sort of thing... but they're STILL orders of magnitude less efficient than code written specifically to do that thing. At most it'd be a tool to lay down the guides which the AI designers/programmers are going to follow, but honestly, I don't think that's necessary. Despite the abomination that is the Civ V AI (and I'm one of the people playing and enjoying the game, not one of the haters!), it is definitely not impossible to write an AI that can provide a good, challenging user experience without this kind of trick.

Incidentally, cleverbot and jackerwocky are NOT examples of the kind of learning AI I'm talking about. I'm just using them as a gateway to a more interesting topic.
 
Only sort of. ...
I don't think you quite got the drift of how the method I described would actually work. I am in fact not talking about methods to optimize the 'check-everything' procedure you think would be necessary.

It is more like giving the AI a box of LEGO bricks (code pieces) for each decission layer to choose from and making the choice be based on current situation cross referenced with statistical data of previous utilizations.

The individual LEGO bricks (code pieces) would still be static constructs, so all the AI is doing is choosing the combination of those bricks - based on historical data about which combinations has previously been proven to work the best (tempered with modifying factors for the chosen level of AI cleverness).
 
Top Bottom