The new AI (beta)

[rfrizz]

Anyone else getting kicked in the teeth by the new AI? Even on the St. Vith Tutorial, I have been having my butt handed to me. (I consider this to be A Very Good Thing, BTW.)

It seems to be far more sneaky, and more aggressive with the sneakiness. For example, the AI used to send a few isolated companies to probe the Lommersweiler area. Now they are coming in as a larger, more organized force. The units are still mostly non-crucial, they still work to deny full control of an objective AND they can snip supply routes.

In other words, this is the sort of thing I do.

[wodin]

This game now has to have the best AI out there. If it didn't already have before hand.

As it will only get better with the east front game it's scary.

[Phoenix100]

What is the AI, exactly, I wonder? (Great question, eh? - answers in ten words, please....) No, but seriously, what is it that goes into it? Is it really any different to what we compare it to, normally - different to scripted behaviour - ie; is what distinguishes the AI in this game just the sheer amount of effort that has gone into creating hypothetical Boolian type conditions and so on, which govern how the AI will react to given events and states? I know nothing about computer programing (as you can see), but I assume that this is what counts as AI, and that the more comprehensive the 'scripting' the more 'lifelike' the AI appears. But it's not actually AI, right? It doesn't actually create any responses of its own? (Though the same might be said of us...) Am I right? Is it actually just very complicated scripts that are at work here?

[Arjuna]

Basically we use a rules based approach - ie we write logical rules that the computer obeys. However, there are a number of things which seperates our partiular engine from others.

For one, most other games use scenario specific scripting whereas we use generic objective driven logic. So what does than mean? By scenario specific scripting I mean a set of rules ( scripts ) that only work for the designated scenario. Eg attack on the left flank and secure the following locations in sequence....1,2,3. In some games these can get quite sophisticated with conditional logic added. Eg If objective 2 is too heavily defended then go straight for objective 3. But regardless it only applies to the designated scenario. A whole new set of scripts have to be written for the next scenario.

In our engine the AI is situationally aware - ie it knows how to access info on the battlefield - friendly and enemy forces, terrain, weather, time of day, supplies, objectives and plans. It can analyse these factors and based on whatever objectives are designated by the scenario designer work out a plan of action and implement it. And it can do this no matter what the map or scenario. All it needs is one or objectives. Moreover, it does this each and every time you run the scenario and based on the unfolding conditions it will often choose a different plan. It is not following a set script it is generically determining its plan.

Another key difference with many other games is the fact that our entities are dynamic free agents - ie they are capable of making their own plans independently of other units. Thankfully more and more games are using this feature. In our engine each unit will go through its OODA loop ( observer, orient, decide and act ) and make decisions based on how it perceives the world around it. They don't just blindly follow the leader.

Probably the next biggest difference that comes to mind is that we have modelled the decision making process on the same proceses used by real military commanders. We model a chain of command that is completely scalable. Many games simply have two maybe three levels - a strategic set of rules for the on map boss and another set of more tactical rules for the subordinates. We do have some specific rules for the on map boss that mostly relate to victory conditions but by and large every force entity uses the same logic regardless of the size of force or what level of command it is at.

I could go on but it's getting late. Enjoy.

[Phoenix100]

Thanks Dave. So, you give the 'entity' a very large set of rules, and many sub-sets of rules, and give rules to individual units, and establish rules to determine conflicts between rules etc, and get them to 'look around' and 'decide' in the light of what they find. I think I get it - it's a little like giving it a rudimentary brain, I suppose. It's a more adaptable and flexible capability than the set of rules that govern 'scripted' behaviour. It's a much more sophisticated script set. You could argue that that's all we do, when we 'act', we humans, I mean, but at an even greater level of complexity. And we certainly have some 'objectives' hard-wired in to set the whole thing off. Only difference is that it doesn't 'feel' like anything when the 'AI' does all this decision making - there's no 'experiece' of the process. I seem to remember writing philosophy essays on this sort of thing at Uni.....

But it does sound like an enormous amount of work has gone into it. And - as has been said many times - it really shows. I agree also with the comment at the top of this thread about it now being harder to beat the AI. I'm not sure why it is, but I've played through Hofen three times now without getting near a decisive win. It seems more stubborn, harder to scare off.

[decaro]

ORIGINAL: rfrizz

Anyone else getting kicked in the teeth by the new AI? ...

It certainly seems to put up more of a fight, esp. for higher-value objectives, by arriving in force somewhat unexpectedly.