v3xt wrote:How hard would it be to make an AI that can manipulate and betray you? A program that plays to win, and is able to use you to meet its own goals.

Extremely hard.
In fact, I don't think anyone's been able to do something of that level yet, except for maybe Google.

If that's the case..
To 'communicate' with the AI, can we have it change its behaviour through keywords in chat? For example, I want to ally with Wintermute the AI, instead of ticking boxes, I could type '/ally' and Wintermute will ally with me. In the 'ally' state, Wintermute will not attack me, and will automatically send subs to outposts you are attacking if it's in range, with a random chance that it will become yours, or Wintermute's when the battle has finished.

If I no longer like Wintermute and would like to attack it, I could either send subs to its outposts to automatically trigger the 'foe' state, or type 'oppose'. In 'Foe' state, Wintermute will attack me, but only when it leads to a decisive victory if it goes on offensive; on defensive, it will try to hold its outposts.

The third state would obviously be the 'neutral' state. In neutral, Wintermute's behavior mimics that of an opportunist: if it predicts a good outcome, it will attack anyone within range. Which shouldn't be easy to come by. Alternatively, it will do nothing; only expand on dormant outposts, or outposts belong to elimnated players, and concentrate its forces on mining and self-defense.

@v3xt at our current tech level, impossible. The best we could do is what I described I think.

Wouldn't everyone ally with the computer player?

What's the name of that AI that beat Ken Jennings at Jeopardy? That's who we need.

Bigredsk10 wrote:Wouldn't everyone ally with the computer player?

What's the name of that AI that beat Ken Jennings at Jeopardy? That's who we need.

I think that was IBM's Watson, I'm not too sure though.

I don't really see a good diplomatic AI anytime soon, but a simple battle one perhaps. Here's my take on it:
Combat stage:
Every X hours (where X is a random time between 6 and 12 hours), it will choose the weakest outpost within its sonar range and attack with its closest outpost. Before sending it, if the AI calculates a win, there is a 80% random chance he will attack, and 20% to recalculate for another enemy outpost within its range. If no guaranteed calculated win, the AI will go through the reinforcement stage.
Reinforcement stage:
Every Y hours (where Y is a random time between 2 and 4 hours), it will send 66% of all its drillers from one factory to reinforce the closest/weakest outpost (calculated independently for each factory).
Specialist stage: If/Else statements could be put in place for the AI to choose where and how to send its specialists. Here are a few examples: Smuggler [will send it to factory and shuttle drillers during Reinforcement stage], Helmsman [will accompany any drillers leaving outpost it was last at], Pirate [during Reinforcement stage, travels to a random friendly outpost. during Combat stage, if closest outpost has a Pirate, any nearby subs are also taken into consideration to attack]
Defense Stage: If the AI sees an incoming attack on one of its outposts, it will send drillers from the 2 closest outposts to the attacked outpost if the reinforcement will arrive in time. The AI will send enough drillers so that the attacked outpost will survive the attack. If the reinforcements are too late or cannot recapture the outpost within 6 hours, the AI will evacuate all of its drillers from the outpost to its closest friendly outpost.

What are your opinions on this?

If you happen to be the weakest neighbor, I think it puts you at a huge disadvantage to everyone else in the game. If you happen to be the strongest neighbor, I think it gives you a huge advantage. Once the AI attacks, you can swoop in and claim their relatively unguarded outposts.

One positive about the current system is that anyone can take a resigned players outposts, but there's a cost to it.

I just thought of an idea which I have seen in other online games - if a player auto-resigns, the game will be relisted on "Open games" and would allow another player to look at the game and choose if they want to step in for the person who left. However, that person will not have a win listed on their record if they do win, only some sort of note that they have finished games.
This could be abused of course - a player joins and attacks everyone, but we could limit access to people who have L2, or finished lets say 4 games.
Opinions?

There's always the option of taking out the 'resign' and 'autoresign' option. This has been discussed by the Devs before. I think this would be a great idea

How would removing the auto-resign feature change anything? Most people that auto-resign just never play the game again, so their drillers would still end up sitting there. Also, the point of this thread is not to think of a way to make auto-resigning less problematic, it is to discuss how a Subterfuge AI would function and how you could program it.

And Panda, your idea of stages seems logical, it would better simulate a person as it wouldn't be active all the time.

I think defensive NPCs are a good idea that way the AI can defend itself and make taking resigned players outposts more risky. However, I don't think offensive NPCs are a good idea. Who wants to have to move troops because a computer is attacking you? I certainly don't.

Who wants to move troops because anyone is attacking you? When someone resigns from the game, the neighbours of that person don't have to defend those borders anymore. That gives certain people an advantage that others won't have. If an AI controlled the resignee's outposts, it would be more fair for everyone involved. As long as it wasn't programmed to be entirely hostile there wouldn't be much of a disadvantage to neighbouring it. It would just make the effects of people leaving less noticeable.