I think learning ai would work well for a single player game, but I think it would be quite difficult to make work in a persistent world with potentially thousands of players.
That would be one hell of a piece of AI scripting too - I tip my hat to anyone that manages to achieve it^^
Didn’t say it would be easy^^ But sure as hell damn amazing to have it^^
I do like that type of behaviour, could be cool having that
I’m pretty sure something like that is what leads to skynet and the apocalypse.
Yea probably but hey at least we had a fun time in Boundless^^
But honestly no^^ This won’t bring us any closer to skynet than a chess bot^^
I already asked them long ago if they plan to update the AI.
The answer is really clear, Yes. Your ideas in general are good but i’d wait for the new AI and more creatures before i could say what behaviors i’d like to see on what type of creature.
Learning AI for a MMO is a hard challenge. Do you heard about AlphaGo ? They need a super computer (1.2k cores) for teaching a computer the game GO 
I also see learning AI more in a singleplayer game.
TBH i’m not sure what’s the outcome of learning AI in a MMO because we all would try to kill them with different approaches and in as result the creature would be “just” shy like normal animals in the nature (they don’t even try to fight humans in normal cases they just run away).
I heard about it yes^^ But I think it’s a bit different.
Anyway. It would still be cool^^
Since it’s a fungus it most likely wouldn’t run away 
But I think it depends. It could as well just be ultra defensive/offensive. Maybe even learn and copy attributes from the enemies it kills and feeds on^^
But that would be even more advanced and might never happen^^
It shows what you need for “real” learning (that was my point). Of Go is not B< but if you like the AI to learn and adapt to the players you need a lot of computational force.
My primary argument is exactly this. This will (always) result in extreme behaviors (always run away, always attack, always fight till 10% and than run away) because we always try to kill them. So why don’t invest the time in a good but static AI instead of wasting a LOT of development time trying to make it dynamic.
Always depends on how deep you want to make the learning and how you program it but it sure takes some considerable ressources.
No it won’t always result in extremes because we adapt too.
As soon as it is so defensive that we won’t attack it as much, it will most likely soften up the defence because it doesn’t need such a strong defence anymore.
Especially the always running away, always attack part is bothering me. It’s dynamic like you said so there won’t ever be an “always doing x”.
And consequently because it is always changing and challenging it wouldn’t be a waste of development time at all.
The “lot” part also depends on how deep you make the learning.
I think you could still add a certain element of randomness to AI, but it would still be based around predefined rules… so “always fight till 10%” could result in “and then run away” or “and then attack with added force in a last ditch attempt” or “run to the nearest [same creature] and get backup”.
They’d be predefined responses sure, but they could be randomised on a per-encounter basis - you’d never really know what outcome you would get until the end of the encounter. You remove that familiarity of knowing how a creature will react every time.
That for me would be a pretty fun way to go for creature AI for a game like B<.
As far as i know now wolf on this planet attacks a human if the human is not treating him extremely. Otherwise it just run away (1000th of years ago we were the eat).
If a creature is weaker than you, you always attack them an kill them (that’s what i do in all MMOs). When do they experience an other behavior from us than that we try to kill them for their loot ? I don’t see how the AI behavior should change if we always act the same.
If the creature is more defensive i raise my attack damage till i smash it under my hammer. If it’s fast i try to trap it … we will always try to kill it and the AI can’t do something else than avoid contact with us under any circumstances (assuming the AI has no “cheat” skill to compare our stars against their own). Even if it has cheat skills it would always act the same and attack us if we are weaker or run away if we are stronger.
there is one other logical flaw in this approach. normally informations and tactics are shared from parents to youngsters. If we always kill the creature how should they teach the “error” to the youngsters (this is theoretical because they won’t “really” teach but it’s bothering me if the AI learns even if it fails [die] a lot).
That’s not a learning AI. This is a static AI with random elements as you described. So there is no difference to my other statement.
dynamic means “learning” and “adapting” in this context. Not always the same (static but maybe random) behavior (random elements are of course part of every good AI). you also need to be really careful with random elements. It’s not useful if the creature starts attacking if it’s at 10% and we are at 90% (logical context is important).
I answerd that question in the next line!
It changes it’s behavior because we change ours because it changes its etc.
you can’t raise your attack infinitely (atm) and even if you could it might still outsmart you. As soon as we reduce the frequency of killing it, it will change again because we changed.
That’s because it’s a hive mind. You can’t completely kill a fungus that spans thousands of kilometers. But it can learn from the mistakes individuals make and adapt. Even if one dies.
Also, if you kill it so much that there are almost no visible extensions of it left, then you won’t find anything to keep killing it at the same rate anyway. It could adapt while you kill other stuff (if you’re so much into killing) and come back completely different.
What you describe is not a hive mind which makes almost concious decisions. It’s the behaviour of a single individual which follows it’s instincts. Namely fight or flight.
Erm… I was running off your last statement…
… giving ideas as to how it could be achieved, but not make it predictable!
This is a really cool (and in contrast to the learning AI also realistic) idea that could lead to some pretty interesting encounters and also guarantee that encounters remain interesting.
both is realistic, just not in the same timeframe^^
The result is the same … we try to kill it because of loot. Nothing really changed at all even if we adept our “way of killing”. If it want to survice it need to avoid us. It’s completely irrelevant “how” it would fight us because the best surviving strategie is avoid us and hide.
Why should we reduce (this is not RL and we can’t find other increadiants for our recipes) ? If i need the loot i go out and kill … i’ll also get another benefit of killing it because if a lot of players give up the prices will rise and i’ll make more profit which will attract other players who try to kill too.
I don’t thought you really mean a “hivemind”. i saw this as an example. also a motionless fungus would by definition not be able to adapt in a proper ways because a lot of possibilities are missing.
But can you see something … we are talking a lot about this topic and we don’t even scratch the surface. How do you think are the devs able to get the concept of a real “learning” AI in shape with a tolerable performance. There is no “simple” AI that is able to “learn” (it’s call “strong AI” btw). You can only try to simulate “intelligence”, like @Stretchious said, with a lot of static elements that are nested or switched in a clever and not obvious way but that’s not “learning”.
PS: I was assuming you meant learning when you said adapting. If this is false assumption excuse me that i missed your point.
No, it could also defend itself or attack us.
Do you even read what I write???
Who says Boundles fungi are motionless or lack the possibilities our earthen fungi do?
No, we talk about the same things because it doesn’t seem to get through to you what I say. That’s not a lot though.
I don’t know. But that doesn’t mean the can’t or won’t do it in the future.
Yes, you can simulate it or you can let it learn. Both are possible the first one is certainly the easier one.
I meant that yes.
I didn’t mean adapting as in having a set of behaviors which it will display when triggered with the right stimuli.
Who says they are motionless, you did:
What possibilities do our earthen fungi have ? Can you name me some which prevent them from being eaten (except toxins).
that was my attempt to close the theme (AI) because if you really like to talk about complexity of AI development we should talk in teamspeak about it (this is not the correct place).
Most likely no and no … did you every tried to write a “simple” AI ? There is a reason why AI’s in games are “predictable” (weak AIs). The complexity of a strong AI is far beyond the frame of a game (it’s a scientific problem).
You can’t let it “learn” that easy. That’s what most of my posts are about. This is highly complicated. If it would be easy we would see intelligent and learning AI’s all over the place but in fact all learning AI’s are scientific at the moment. As said before, if you like to deepen this topic PM me a time and we talk about it in TS 
EDIT: The other half of my posts is about the behavior change of types of creature (like your hivemind AI). While this is a nice idea it won’t justify the time spend to develop it. Why ? Because there are “stable states” (Read Nash equilibrium for more informations) which are as boring as static / weak AIs. After some time there will be stable states and if one side changes their strategie (player or AI) the other side will adapt and force a new stable state (the adaptation is significant faster compared to the stable states duration). Furthermore if it’s a hivemind and if all player driven influences are used as inputs this will act like an “average”-method in statistics. A bad player will force the AI to attack while a strong player will force it to hide. In the result (average) it will most likely stay around an just do nothing (hypothetical). There are far too many variables in this scenario. You could write a game or a PhD about this topic alone without any problems.
There are frameworks and companies who dedicated work on game AI for games (like xAItment) and even their AIs are only weak AIs (highly developed but weak) and not really learning at all.
EDIT2 (That’s why TS is a better solution for such a discussion): If we talk about strong (learning) AI we talk about an AI able to learn NEW strategies independent from a developer pre-set. If you take this scenario it would be possible for the AI the learn the behavior “trap the enemy in lave” from the combined behaviors “tease the enemy” and “run away”.
To learn this without any hints from the devs the AI needs to be able to
For one final strategie there is the need for a lot of “sub-behaviors”/sub-strategies all need to be worked out independent from the developers. they only produce a minimal subset of behaviors and the ability to learn.
looking into the nature you may answer this questions according to learning:
If this would be possible within the time of game-development all wonderstrucks devs would work for the DoD / NSA or Google and / or would be stock market billionaires.
… and call them Marvin? … NO WAY!!1
I can understand bullets one and two here, but how would bullets three and four make sense?
EDIT: something I thought of while reading another post:
ghostly creatures that hunt in packs and can turn invisible.(alterntively, they could just hide in foilage or use active camoflauge) they stalk their prey and herd it into a kill zone (inside a ravine where it can’t escape or on top of a cliff where it would be easy to push it off) one turns invisible until it’s in the Ideal position. it appears in the direction that would scare the prey into the kill zone. once the prey is in the zone, they all appear and attack. they would never attack unless the prey was in the kill zone. if you move toward them when they’re in scaring mode, they retreat.
it would make for an interesting situation where the safest thing to do would be to move toward the creatures instead of away. it would also be really spooky.
alternatively, a less spooky (or even cute) creature could taunt an adventurer into following him into the kill zone. that may work better with our videogaming brains hardwired to kill and loot as many creatures as we can find, or if the player is merely curious about where this tiny creature is leading him.