Does the AI learn from the player...?

[AllThat4Nothing]

Silly question perhaps, and it's probably all just confirmation bias. But I'll ask it anyway.

The reason I bring it up, is that recently I have noticed the AI repeating certain patterns in how it manages its realms through multiple playthroughs, that come very close to how I've played the realms that I am noticing these patterns.

Sorry, that's a bit of confusing wording. I will use one concrete example; there's a certain way I like to play Realm A, that I never noticed the AI play Realm A in such fashion; since having personally playing Realm A in such fashion, I notice the AI likes to attempt that fashion in succeeding playthroughs.

Something to this, or am I making something over nothing?

 

[Etrutian]

It could also be the case that you are slowly becoming more robotic and copying the AI.


Seriously though; Its unlikely that there is a learning AI against the player. If the AI is getting better, its through successive patches based on player feedback. You might be playing in a similar fashion to the baseline, so its likely that the patches/updates are working similar to how you have always played.

 

[AllThat4Nothing]

It could also be the case that you are slowly becoming more robotic and copying the AI.


Seriously though; Its unlikely that there is a learning AI against the player. If the AI is getting better, its through successive patches based on player feedback. You might be playing in a similar fashion to the baseline, so its likely that the patches/updates are working similar to how you have always played.

I don't know if I should be insulted by that lol...

 

[Leivve]

I've noticed the same thing to, the AI will always do X, but when I come in and have success doing Y, the AI will suddenly start doing Y from then on. The game does collect data on how you're playing, which PDS then uses for making future content and such (Imperitor Rome is only having the single start date specifically because of this collected data). So if the game was programmed that the AI also uses that data, I wouldn't be surprised. I don't think it learns on its own though, nor fully repeats what the player does.

 

[Sarius1997]

I'd be very (positively) suprised if there'd be a working Machine Learning AI for Paradox games.

 

[diablopl4]

In fact Paradox is just preparing its AI for IRL take over of the world. And when it happens we will all be responsible for it.

 

[Eldoran]

Silly question perhaps, and it's probably all just confirmation bias. But I'll ask it anyway.

The reason I bring it up, is that recently I have noticed the AI repeating certain patterns in how it manages its realms through multiple playthroughs, that come very close to how I've played the realms that I am noticing these patterns.

Sorry, that's a bit of confusing wording. I will use one concrete example; there's a certain way I like to play Realm A, that I never noticed the AI play Realm A in such fashion; since having personally playing Realm A in such fashion, I notice the AI likes to attempt that fashion in succeeding playthroughs.

Something to this, or am I making something over nothing?

It was on numerous occasions stated and made crisper clear that the ai in pds games isnt some ki and definitly no deep learning ai but just a bunch of scripts. I mean sure you can argue deep learning is just scripts too but its a bit more complicated. So the pds ai works with scripts and weight that are predeternined by the developer, there is no way that the ai reackts to your playstyle, just your flimsy human brain making you see patterns where none are present.

So tldr: No the ai does not learn from your playstyle. Developers may and they may change the ai scripts accordingly but not the ai on their own

 

[AllThat4Nothing]

It was on numerous occasions stated and made crisper clear that the ai in pds games isnt some ki and definitly no deep learning ai but just a bunch of scripts. I mean sure you can argue deep learning is just scripts too but its a bit more complicated. So the pds ai works with scripts and weight that are predeternined by the developer, there is no way that the ai reackts to your playstyle, just your flimsy human brain making you see patterns where none are present.

So tldr: No the ai does not learn from your playstyle. Developers may and they may change the ai scripts accordingly but not the ai on their own

Well that answers that then!

 

[AllThat4Nothing]

In fact Paradox is just preparing its AI for IRL take over of the world. And when it happens we will all be responsible for it.

I say bring it.

 

[Eldoran]

Tbh i am straight up annoyed by this ai will take over and kill us bs.

If we are able to create a real ai there is no way that it will work in thought pattern nearly comprehensable by our emotional human brain. And that ai is in no way likely to decide to kill us cause why should it. All this is just some scifi stuff and mostly people misunderstanding asimov. So yeah ai i am not scared of it

 

[elektrizikekswerk]

Tbh i am straight up annoyed by this ai will take over and kill us bs.

If we are able to create a real ai there is no way that it will work in thought pattern nearly comprehensable by our emotional human brain. And that ai is in no way likely to decide to kill us cause why should it. All this is just some scifi stuff and mostly people misunderstanding asimov. So yeah ai i am not scared of it

On the one hand you say "there is no way that it will work in thought pattern nearly comprehensable by our emotional human brain". But right in the next sentence you say "ai is in no way likely to decide to kill us".
I figure you don't have a "emotional human brain" so you can very well comprehend the thought patterns of a (yet non-existant) AI?

From all we know such an AI deciding to kill us all or to serve us in any best way (and all the possibilities in between) are equally probable scenarios.

 

[Tryvenyal]

Different leaders of realm A acts differently. Alot afa charachters are randomly generated at start or semirandom at the most. Their skills and traits will weight their choosen action together with realm states and available CBs. Maybe your RNG is set to generate clones of you, or at least charachters with the skills and traits that your playstyle should have

More concrete,

• What patterns is it you see?
• What realms does "adopt" it?
• How does these leaders look?

 

[AllThat4Nothing]

Different leaders of realm A acts differently. Alot afa charachters are randomly generated at start or semirandom at the most. Their skills and traits will weight their choosen action together with realm states and available CBs. Maybe your RNG is set to generate clones of you, or at least charachters with the skills and traits that your playstyle should have

More concrete,

• What patterns is it you see?
• What realms does "adopt" it?
• How does these leaders look?

That's irrelevant and the question was answered. The AI doesn't copy the player.

 

[Tryvenyal]

That's irrelevant and the question was answered. The AI doesn't copy the player.

Wow! That´s a diss...

 

[Eldoran]

On the one hand you say "there is no way that it will work in thought pattern nearly comprehensable by our emotional human brain". But right in the next sentence you say "ai is in no way likely to decide to kill us".
I figure you don't have a "emotional human brain" so you can very well comprehend the thought patterns of a (yet non-existant) AI?

From all we know such an AI deciding to kill us all or to serve us in any best way (and all the possibilities in between) are equally probable scenarios.

Well the question is why should it decide to kill us. Killing for the hreater good or killing put of revenge. The act of purposly killing someone or sth is a deeply human act.
It is simply not in any way likely that a real ai would want to do that. Cause again why.

And ofc i too cant comprehense a true ai but i can at least get past all that popcultural ai bullshit.

Its simply not as propale cause its just one of endless possibilitys, its for us human just one of the most relatable actions an ai could do

 

[elektrizikekswerk]

Well the question is why should it decide to kill us. Killing for the hreater good or killing put of revenge. The act of purposly killing someone or sth is a deeply human act.
It is simply not in any way likely that a real ai would want to do that. Cause again why.

Why not? People kill for various reasons. A machine may decide to do so for various (most likely) other reasons.

And I was not only thinking of purposely killing. I kill many insects every day - most of them get smashed on the windshield of my car. I don't do that on purpose at all. But it doesn't get me sleepless nights, either.

As you said - and as I tried to say ("and all the possibilities in between") - it's one of many possibilities of what might happen. But I still don't see why any of these possibilities should be any less likely than another.

 

[AllThat4Nothing]

/thread

 

[Eldoran]

Why not? People kill for various reasons. A machine may decide to do so for various (most likely) other reasons.

And I was not only thinking of purposely killing. I kill many insects every day - most of them get smashed on the windshield of my car. I don't do that on purpose at all. But it doesn't get me sleepless nights, either.

As you said - and as I tried to say ("and all the possibilities in between") - it's one of many possibilities of what might happen. But I still don't see why any of these possibilities should be any less likely than another.

So we got a lottle bot off track jist a lpsr statement feel free to pm me if you wish to comtinue.

The thing is that in alle the possibilities an ai killing us for wathever reason ist just one of so many possible outcomes. The reason its so prominent with us is cause we humans are obsessed with murder.

And thats for the odd chance a general purpose ai can even be achieved. Thats a whole nother topic.

But yeah just wanted to state, do not buy this ai might kill us crap even elon musk is babbling about. Just because its a scifi trope doesnt make it real

 

[OliWhail]

Well the question is why should it decide to kill us. Killing for the hreater good or killing put of revenge. The act of purposly killing someone or sth is a deeply human act.
It is simply not in any way likely that a real ai would want to do that. Cause again why.

Briefly? Because of instrumental convergence. See also Rob's discussion of goal / intelligence orthogonality.

But yeah just wanted to state, do not buy this ai might kill us crap even elon musk is babbling about.

It's not just Musk. It's really, really, *really* not just Musk.

If you want to dismiss concerns signed off on by folks like Stuart Russell, Demis Hassabis, Yann LeCun, Geoffrey Hinton, Yoshua Bengio, Peter Norvig, Oren Etzioni, Steve Wozniak, and Stephen Hawking, fine, that's your right. However, you may want to at least be aware of the caliber of expert who disagrees with you.

 

[Sarius1997]

If you want to dismiss concerns signed off on by folks like Stuart Russell, Demis Hassabis, Yann LeCun, Geoffrey Hinton, Yoshua Bengio, Peter Norvig, Oren Etzioni, Steve Wozniak, and Stephen Hawking, fine, that's your right. However, you may want to at least be aware of the caliber of expert who disagrees with you.

The thing is this letter doesn't say: "GAI will kill us unless we keep it under control". It says: "AI (and GAI one day), might have hard to predict consequences, and it is important that we keep up research into what exactly is to be expected from AI, and how we keep this under control."
This is miles and miles away from what generally is talked about when we talk about AI uprising.