The AI Complaint

[geoschmo]

ToxicSlurpee said:
Until we create artificial sentience, AI will have to do as is. When that isn't satisfactory, we'll simply have to play each other.

I agree it's very difficult but I think it should be possible. The problem is a limitation of using algorithms itself. You quantify the variables, determine the best possible outcome, the most efficent means of achieving that outcome, and follow those steps. You continue on that course until the variables change and then you stop and reevaluate. It's all very logical, but isn't at all how most humans operate.

We often times ignore important pieces of data, make incorrect assumptions, reach incorrect conclusions, and don't always do what is the best in any circumstance. This intruduces an aspect of unpredictability that will sometimes result in success in spite of poor planning.

I think an AI could approximate this without having true sentience. Instead of always taking the "best choice" for instance, it could be programmed to rate various options and randomly choose one from all options that are at least within a certain percentage points of the best one.

I am sure it would be hard to code such a thing though. The key would be for it to remember what it's done in the past and what other players do in response, and factor that into it's future decision making. Other wise you'd have a bot that carreened from decision point to decision point with no clear stragetgy and constantly changing course.