In the 1966 movie classic Fantastic Voyage[1], a group of humans were shrunk and inserted inside a human body to cure a disease. Sixteen years later, Tron [137] shrunk film heroes into a new fantastic domain: the world inside the computer. This new pop icon sprang up from the massive impact that the 8-bit microprocessor had in our culture, bringing us personal computers and an explosion of arcade video games.
Virtual worlds as studied by Artificial Life are inhabited by creatures that reproduce, learn and evolve -- with varying degrees of physical realism -- without human participation. For the virtual worlds of video games, instead, humans are invited but the emphasis is on the visual appeal: artificial opponents are present but they rely on access to the internal variables of the game more than artificial adaptation -- they are not artificial life beings.
Robots interact with physical reality and live creatures: they are embodied. But robotics research is difficult because the real world brings with it limitations of space, budget, engineering and speed. A video-game instead, is a simulated world where human intelligence can meet artificial adaptation, through the immersive experience of virtual reality.
Here we posit that the science of Artificial Life should employ the metaphors of virtual realities and video games to attain knowledge about adaptive behavior, putting artificial agents in contact with live creatures, by introducing them into a simulated world. Virtual worlds could enable ALife researchers to study how artificial and natural intelligence coevolve, adapting to each other. Moreover, with the revolution of the Internet, these worlds can reach thousands of human subjects and artificial learning can arise from the combined contributions of many individuals.
We present the first work that evolves agents by having them play against humans. The recreational nature of a simple video game attracts people and creates a niche for mutual adaptation between people and agents, providing the substrate for the first experience in learning a game through the massive training by thousands of human subjects.
The physical features of our model are limited to the simulation of a two-dimensional space with walls; but they are sufficient to observe the emergence of navigational behaviors such as wall following and obstacle avoidance.
Parts of this research have been reported on the following publications: [50,51,44,125].