Readings (Eliza / Expressive AI) – Kyna

Expressive AI

This reading, while very technical in nature, forced me to consider a few things that had never occurred to me before. It never occurred to me that the subtle differences in the methods used by AI developers would manifest themselves into separate schools of thought. The interactionists use a model based on reacting to changes in an established environment, while the GOFAIs use the sense-plan-act model. These two groups have similar goals; they both desire to create an AI with a thought pattern close enough to reality to be classified as actual intelligence and not mere coded reaction. The interactionists in particular seek to make a being that can be considered alive. Next to these lofty goals, the goals of the artist seem significantly less noble. Artists, according to the writer, are simply after a believable AI system, not one that actually embodies intelligence but that merely seems intelligent. For the more technical research groups, an AI system is the end product of their labors. For an artist, the AI is just a tool, and the interaction with an audience is the end goal.

Eliza

The Eliza effect is a common occurrence when someone utilizes and open-ended AI personality. At first, you play right into the expected patterns, and the AI responds in a very believable way. However, as the conversation continues it soon becomes apparent that the AI follows a discernible pattern of speech that can be easily broken with the right word choice. Some other systems of conversation AI employ obvious road blocks that break the immersion of the player as well. Galatea for instance will simply tell you that you cannot do something. I think perhaps Facade has the best method of getting around this barrier. The characters of facade, if you say something they don’t understand or can’t form a response to, will just move on with the conversation as though you never said anything, or change topics to one they can respond to. This comes close to the response that some people actually give when they’ve misheard you or don’t know how to react to what you just said.