Obviously in the past games we've been able to choose whether AI favours land/air/naval/turtle/rush/adaptive. I'm wondering whether there are plans to try and extend this further so it's even more variable and less predictable? For instance, an enemy changing it's preference based on the current planet's terrain, map control, enemy tactics, etc. I'm sure Sorian has many ideas down this path, just wondering whether there's any movement down this path, or any little gems we can hope for in the future?
Well of course, but I know you've got a deep interest in AI, I just wondered what grand plans you may have in this area.
Whatever constitutes a 'Good AI' is something I'm sure Sorian is striving for, he's had a lot of experience in crafting good AIs, even if up until now he's been limited in what can do. Mike
I don't doubt it, I'm just wondering whether he's got any thoughts or snippets. I'm not badgering for ETAs or complaining it's not in the latest build. Just trying to discuss the future. Additionally we could also discuss what sort of strategic personalities or behaviours we'd like to see or tried out. e.g. What worked and what didn't for these concepts in previous titles?
I suspect, considering the relatively low budget for the game, AI might be constrained by however quickly Sorian can implement new behaviours and options etc. Options for AI opponents with different play styles (rapid expansion, turtling, etc) would be super awesome but also quite a bit more difficult to implement, I would have thought. My limited experience in programming being what it is. AI in general is quite difficult, from what I know. I for one am really looking forward to seeing the AI being developed over time, as I tend to play the AI a lot when I haven't got someone to play against. I'm looking forward to seeing how the AI will challenge me!
Im no expert in AI, but I guess a static behaviour like turtle, rush, land, ect... would be easier to implement than one that adapts and changes behaviour. I would love that static options for sure for "easy" difficulty or training purposes. Adapting AI would then probably be the next level and way more difficult to beat. Maybe even good enough to act as dummys if you don't have enough players but want more opponents.
The community could also do a lot of stuff with AI, if they make it possible. Even in Starcraft 2 community has worked on some amazing AI, sadly the way SC2 works its limited for the maps with the AI implemented in. But someone made GreenTea AI, it would play like players, rush you, with the must absurd strats and such. I have a feeling that Planetary Annihilation is gonna be great in this respect.
I had always assumed that the difference between an adaptive AI in supcom, and one of the horde, tech, turtle, etc. "personalities" is that the adaptive one simply has a system for determining when it needs to switch between the other implemented styles based upon various input metrics. I'm not an AI expert though, so I could be completely wrong.
So 'Random' would of been a more apt name? 'Adaptive' kinda does imply that the ai might, you know adapt
It's not random, it adapted to the most appropriate play style for the map at the start of the game and maintained it throughout.
What carn1x said. The AI would look at several factors (distance to nearest enemy, nearby water, etc) and choose a personality based on that.
I just think the reason it confuses people is because Adaptation implies changing behavior based on concurrent circumstances. Which is what people are asking about right now. For PA. Is this possible?
I'd imagine you could get a very basic version going with something like if (being boned) then (turtle)
Failing that, it would be awesome if AIs were scripts or dynamic libraries with a documented API so we could easily implement our own.
If it's possible I would really want to see it scout regularly from several different directions. Based on that scouting information, the AI would effectively count the number of units, buildings, building type and generates a response based on that information. For example, if it scouts and only sees bots, there is a chance to trigger a response to build more bombers. If it sees a lot of base defenses, it might choose to have a relatively higher amount of artillery units. If you turtle and don't attack with a large amount of forces, it might choose to move up the production to the front line, and build lobbers, catapults and nukes. The AI would place radars in a wide ring around him to be able to shift defensive forces to the correct location.