next up previous
Next: Experimental Model Up: Intelligence on the Web Previous: Intelligence on the Web

  
A space where agents can thrive and evolve

It has been suggested that intelligence should be present on Internet sites in the form of intelligent agents. According to this hypothesis, such environments will contain software agents that interact with human users and adapt according to the behavior displayed in those interactions [93].

With Tron we are exploring the hypothesis that one of the forms in which this idea may be realized is through the presence of species of agents evolving through their interactions with the rest of the web. From this perspective, the Internet is seen as a complex environment with virtual niches inhabited by adaptive agents.

Here we propose that learning complex behaviors can be achieved in a coevolutionary environment where one population consists of the human users of an interactive software tool and the ``opposing'' population is artificial, generated by a coevolutionary learning engine. A niche must be created in order for the arms race phenomenon to take place, requiring that:

1.
A sufficiently large number of potential human users must exist.
2.
The artificial population must provide a useful environment for the human users, even when -- in the early stages -- many instances perform poorly.
3.
An evaluation of the performance of the artificial population must be measurable from its interaction with the human users.
The experimental learning environment we created for the game Tron met these requirements. First, the game is played in a Java applet window on our web site. As Tron was being launched, Java was a new thing and there was a great interest on any applications, particularly games. So by advertising our site in Java games lists we were able to attract visitors. Second, our earlier experiments with Tron had shown us that, by self-play, we could produce players that were not entirely uninteresting when faced by humans. And third, each round of Tron results in a performance measure: a win, a loss or (rarely) a tie.


next up previous
Next: Experimental Model Up: Intelligence on the Web Previous: Intelligence on the Web
Pablo Funes
2001-05-08