One of the interesting things about the AI research that gets funded is that most projects are centered around making tools, not intelligence. A great deal of money has gone into computational linguistics so DARPA can give it's Future Soldier program a universal translator, and Bayesian Reasoning is quite useful for intelligence analysis. The creation of intelligence, while an interesting intellectual curiosity, is second to fulfilling other needs.

If we want to map out innovations, we must first map out the desires of the agents in the world that we simulate. There are many psychological theories behind what motivates humans, like Maslow Hierarchy of Needs, but I'll put out one from a different realm. Drew Whitman created the Life Force 8 for advertisers to help them understand how to appeal to customers to sell products:

1. Survival, enjoyment of life, life extension
2. Enjoyment of food and beverages
3. Freedom from fear, pain and danger
4. Sexual companionship
5. Comfortable living conditions
6. To be superior, winning, keeping up with the Joneses
7. Care and protection of loved ones
8. Social approval

All humans are hardwired with the above basic needs and appealing to them works by default.

He also adds the following "Learned wants" which are not as strong but still a motivator:

To be informed

To satisfy curiosity

Cleanliness of body and surroundings

Efficiency

Convience

Dependability/quality

Expression of beauty and style

Economy/profit

Bargains

This is good for figuring out the desires of agents in the real world, but it doesn't mention the finite number of potential states a tool can come in to satisfy that desire. Mark Proffitt has does some interesting work in that arena:

http://www.slideshare.net/MarkProffitt/predictive-innovation-airbag-product-family-matrix?type=presentation http://markproffitt.com/media/

Any thoughts?