Podcast Episode 8: In-Game AI

Yeah, but the podcast was about AI programming, not difficulty levels. We know that the AI will play smarter on higher difficulties and dumber on lower ones. What remains to be seen is if they need bonuses on the highest of difficulties in order to effectively challenge the best human players.
 
My guess is that there will be 'cheating', because there is no way for the Civ AI to compete at high levels without it, but that it will be less pronounced than in Civ 4, where the AI got bonuses at essentially every level.
 
My guess is that there will be 'cheating', because there is no way for the Civ AI to compete at high levels without it, but that it will be less pronounced than in Civ 4, where the AI got bonuses at essentially every level [...]
[...] which is a great progress in my books :)
 
Yea, it's a huge difference. If they can cut out most of the production/research mods in particular, it'll make high level gameplay less cheesy.
 
Based on what Ed said, it sounds like the AI will take longer to take its turns as you go higher up in difficulty. :(
We're also looking at kind of a different depth of analysis in terms of the military and tactical game when you go and you have a higher difficulty setting.

Very interesting information here (this podcast). My only fear with this new AI design is that the hardest difficulty be too easy to beat. I hope there are still enough handicaps to prevent that happening.
 
Thanks, Greg!

This AI sounds VERY promising,
only for me it is hard to imagine that all subsystems can give out
a coherent output that to cover a full successful game for the AI civ.

I fear of turns when AI will be mislead, confused, contradictory...
...but oh, maybe that fits well with the AI being "human-like" :)
 
I'm sure the Ai will be better than CivIV, but I don't think we have skynet yet. I'm sure the ai will still need to have bonuses to provide proper challenge at the highest difficulty levels. But smarter is better!
 
[...]
I fear of turns when AI will be mislead, confused, contradictory...
...but oh, maybe that fits well with the AI being "human-like" :)

I have no doubt that dedicated players after a while know what to do to confuse the AI, but that doesn't necessarily mean that's bad. Those same strategies would maybe work against human players too. It all depends on the way it is implemented. I'm looking forward to find out. :)
 
Very interesting podcast -- I am definitely feeling more positive about the AI and variability after reading this. (Thanks for the transcript!) Specific "flavors" for each civ, but with some randomness added to keep things from being too predictable -- very good news.

I also liked the bit about civs being smart enough to change course if their usual flavor will not work because of starting conditions, etc.

On bonuses for the AI at higher difficulty, I think they will still exist. Maybe they will not be as large as in Civ IV. But part of the point of the higher difficulty levels is the overall challenge of facing boosted opponents.
 
They say it gets smarter. But it's impossible that the AI won't cheat in the higher difficulties. Juyst think about it, really. No AI could beat good players on such a complex game without cheating. This is not chess where you can just use algorithms for checking many moves into the future
 
No AI could beat good players on such a complex game without cheating. This is not chess where you can just use algorithms for checking many moves into the future

Modern chess AI works different way. It has:
- Openings book.
- Endings book.
- Strategy layer.
- Tactical layer.
Only tactical layer just calculates the moves.

With good programming you could beat human player even in Civ game just by using better calculation. The only reason why I don't think it's the case - is the fantastic amount of work required for this.
 
Yeah, that was a surprising mention. My guess is there are still difficulty related bonuses and penalties (which, btw, is distinct from "cheating" imo), but perhaps they are less pronounced (or only take affect at the highest of difficulty levels).

Looking forward to the next one. How many are left?

I actually liked the CivIV system: you knew exactly where their advantages lied and could compensate. The way they described it in the podcast, the AI will be simply smarter in general, at the higher levels. At the same time, they aren't explicitly dumbing down the AI at lower levels, but rather giving it suboptimal choices it can make among the better options. It'll be interesting to see how the AI has changed.
 
People actually think that the AI won't cheat at higher difficulty levels :lol::crazyeye:
To do that you'd need YEARS of programming just for the AI after the rest of the game is (mostly) completed and testing vs. various highly-skilled players and then a super-computer to run the program.

Utterly ridiculous.
 
With good programming you could beat human player even in Civ game just by using better calculation.

No you couldn't. No amount of programming would change the fact that you'd need an absurdly powerful supercomputer. Civ, like most complex games with hidden information, is much harder to calculate than chess.
 
At the same time, they aren't explicitly dumbing down the AI at lower levels, but rather giving it suboptimal choices it can make among the better options. It'll be interesting to see how the AI has changed.

I don't think you'll notice if AI will be choosing wrong technology or building trading post instead of farm (and vice versa). Strategical combat mistakes (like not sending some troops to the key point) will also be not very noticeable, so I think you'll only notice direct tactical mistakes.
 
No you couldn't. No amount of programming would change the fact that you'd need an absurdly powerful supercomputer. Civ, like most complex games with hidden information, is much harder to calculate than chess.

A lot of things could be precalculated and set up up as priority values. Look at chess opening books as an ultimate sample.

From what I understand, the developers use this trick quite actively. But it has a drawback, since these values may require reconsidering with each balancing patch.
 
Yesterday someone posted a "lecture" by the AI programmer of Civ 3+4, which gives great insight. I just can't find it any more. Can someone link it here, please?

Basically, they had a huge conflict between making the AI competitive and making them roleplay as historic leaders (which added to the fun for a lot of players, but generally was dropped for a reason).

One example is religion and it's diplomatic modifier. The Civ4 AI role-played and therefore cared about the religion of the human player, but the human didn't need to.

In Civ5, they went away to a huge extent from this roleplaying, historic AI and made it simulate human behaviour, which makes it easier for the AI to act smart.



@stealth_nsk:

The lecture explained that the Civ4 AI did NOT have lists of build priorities ("build temple, then barracks"), but the AI thought about which basic items they need (culture, production,...) and built the buildings that provided this, no matter what they were called. That made the AI quite resistant to patches and mods.
 
The lecture explained that the Civ4 AI did NOT have lists of build priorities ("build temple, then barracks"), but the AI thought about which basic items they need (culture, production,...) and built the buildings that provided this, no matter what they were called. That made the AI quite resistant to patches and mods.

Yes, with things like buildings you could just compare cost/effect.
I was speaking about precalculating more strategic decisions, like beelining to Steel if you have enough Iron and want active war.
 
Top Bottom