With the reciprocal learning environments presented in this thesis, we have described new ways to exploit the enormous potential of feedback loops between human intelligence and evolutionary methods. The ``blind watchmaker'' algorithms [30,86], and TD-Leaf's induction of the relative values of chess figures from games against humans [7] were previous indications of this potential.
EvoCAD (section 2.9) shows how a design tool can employ AI to generate creative solutions for engineering problems. Researchers in design usually reject the concept of artificial creativity, yet the diverse characteristics of computer and human problem-solving lead to solutions that are radically different. If our experiments can be described as generating surprise and innovation [116] then the goal of artificial creativity might not be impossible, after all.
In EvoCAD, human and machine take turns in trying to bring a design closer to a final goal. But Tron puts human and machine in a collaboration of a larger scale.
The cooperative effect of large groups of humans working together on the Internet is known, but Tron is the first example of an adaptive system that harvests voluntary contributions of intelligence from its users. The technique also induced learning in humans, suggesting that the coevolutionary dynamics can produce new kinds of educational and entertainment environments through coadaptation between machines and humans. Sklar and Pollack [126] have begun an effort to create educational environments for people based on the type of mutually adaptive challenge first demonstrated by Tron.
The Internet is a new kind of environment where people and software interact in an unprecedented scale. Thus the potential for a new kind of intelligent software that -- as exemplified by our Tron domain -- thrives on the niches of a virtual ecology that mixes natural and artificial life forms.