OK, this time fer shurr: I am going to come up with the AI for actors in dream combat in this essay. The very first decision is that actors will not make a decision based on their own strengths; instead, they will make a decision solely on the perceived weaknesses of their opponents.
I will not base the actor’s decision on the lopsidedness of aura counts; the single most powerful consideration is the weakest aura count; every actor will avoid using that.
However, I will add a consideration for the degree to which an actor likes the opponent; actors will refrain from attacking their friends, because those friends are an important source of information.
But one nasty problem arises: how do I take into account the uncertainty of my information. Suppose, for example that I am choosing between attacking Joe and Mary. Joe has (2.8±1.9, 0.9±1.5, 1.6±1.2} while Mary has {2.6±0.9, 1.7±0.5, 1.1±0.4} Joe has the lower count (0.9 ± 1.5), but I’m more certain of Mary’s 1.1 ± 0.4. Now, if I merely base it on simple probabilities, then even the greater uncertainty of Joe’s count does not alter the fact that I’ll have a higher probability of beating Joe than of beating Mary.
Ah, but there’s a catch: the distribution of net benefits is lopsided. That is, if I beat Joe, then I gain from his loss. But if I tie with him, then I lose an aura, and if I lose to him, then I lose an aura. I must therefore calculate the losses and gains I obtain from these different outcomes.
The first approximation is simple: an outcome earns 1 point for every opponent aura I destroy, and loses 1 point for every aura I lose. This yields the following table of results:
win: +1
tie: 0
lose: -1
However, this first approximation isn’t right, because, from my point of view, losing an aura does not balance destroying an opponent aura. I would not consider a tie to be of zero benefit; I would consider it to be an undesirable outcome. So I should place more weight on my own loss than I place on an opponent’s loss. If I place half a point of extra weight on my own loss, then the { win, tie, lose } benefits set becomes { +1, -0.5, and -1.5 }
To hell with it!
I’ve been going back and forth on this all day today, and I have come to a decision: the actor will choose the opponent with the smallest aura count, and will give no consideration to the uncertainty. It’s just low aura count plus how much the actor likes the candidate opponent. That’s all. Simple and easy. I can tune it up later if I see a need to do so.
I do this too often: try to overthink things, try to come up with a brilliant solution when a good solution is ready at hand.
