At Autonomous Agents 97, Stephen Grand, at the time not well known to others at the conference, gave a presentation on a system, Creatures, that was in production, and in continued development, at Millenium Interactive / CyberLife [Grand, Cliff, & Malhotra1997]. Grand showed pictures of the relatively detailed, and dense, 2 1/2 D (e.g., with flat, but layered graphics) Creatures world, filled with cute imaginary animals, elevators that rise and fall, toys, food, and so forth, which he presented as objects with scripts that describe how they interact with other objects (see figure 4). The talk continued with details of how the biped creatures respond to positive and negative reinforcement, mature over time as they grow, and most importantly are influenced by their (symbolically represented, and effected) genetic makeup. By the time Grand began discussions of the genetic rules, user control of genetics, and adaptation between generations, ...of the creatures' ability to learn a simple verb-object language, and of a surprisingly robust (for a game!) neural-net ``brain'' model of a quasi-biological self-modifying response to stimulus, it would probably be safe to characterize audience reaction as one of, ``We have another mad scientist here, albeit one who certainly knows all the right buzzwords. Too bad such systems are so deadly dull when one looks at the code and sees them run.''
To the delight of many (might we venture all?), Grand's presentation ultimately DID convince, and decidedly so. This turned out to be one of those rare instances when a diverse set of content theories, after being filtered into similar levels of granularities, were synthesized into a real-time, temporally robust, system that works. Grand and his colleagues mix knowledge of genetics, biochemistry, object-oriented programming, engineering, artificial life, artistic craft, and a great sense of both humor and fun, into an inspired piece of small-scale software that not only holds up as entertainment, but is also fascinating to ponder from an intellectual point of view.
The end product cannot be safely characterized as a game, since there is a strong social fabric that develops not only between the agents themselves, but also between the user and the creatures that the user nurtures from an egg through old age.
Like other such products, it is clear that ethical questions might best be considered: how attached do participants become to their creatures? How do they feel when the creatures die - especially from neglect?
Because the mix of complex underpinnings and the very real constraints of a commercial, real-time, product are so tricky, it is difficult to suggest what might be the hard problems facing this group. One would hope that given the symbolic approach to many of the interesting schemes represented here we might see high-level control of some sophisticated social processes, and even personalities, as well. Further papers, and general information on this work can be had at http://www.cyberlife.co.uk/.