Creatures 2

Dateline: 10/20/98

First generation of Creatures, a virtual world of fuzzy little characters called Norns, were introduced in December 1996, and subsequentially described in one of the previous feature articles on this site. Creatures were the first commercially available entertainment title to deliver new artificial life technologies to the general public. Modelled on real biological principles, Norns can think and act for themselves, while their behaviour is not specifically programmed, but emerges from their particular needs and drives. Norns can also learn to communicate with each other and learn from their experiences.
This game is entirely different from the "Quake-style" products that focus and death, guns and destruction, creating a kind of "anti-Quake" culture where players can exchange their pets and their "pseudo parenting" experiences. But don't compare Norns to ordinary virtual pets such as Tamagotchis: the technology behind them is far beyond ordinary game or simulation programs. These little creatures have digital DNA that can be mutated, so they are developing in unpredictable ways. They have a neural network for a brain, which means they can learn; they also have a simulated biochemistry.
Douglas Adams, author of "The Hitchhiker's Guide to the Galaxy", once noticed that the software used to create Creatures is more exciting than the discovery of evidence of life on Mars. More than half a million copies of "first generation" Creatures has been sold, and the game was named a finalist for a prestigious Codies Award under the category of "Best New Software Program of 1997" as well as being voted "Best Strategy Game of 1997" by MacWorld.

Creatures 2 features new version of Norns' virtual home, Albia, that now has weather, seasons and a more complex food chain. Norns now have 120 facial expressions, so players can tell what they are thinking and how are they "feeling". People from CyberLife say that some particularly intelligent Norns whose DNA had mutated into a form that was much better than the original program were incorporated into this version of the game.

But there is much more in this than an ordinary computer game. Stephen Grand, Director of Technology at CyberLife, says: "The only way way to create living systems is to create models of the building blocks from which existing life-forms are constructed." There is an important distinction between cyberbiology and other adaptive methods (neural networks, genetic algorithms): cyberbiology is not about the general application of adaptive techniques, but is specifically about the creation of synthetic life forms. The art of cyberbiology is in extracting the "essence" of biological systems to construct similar, but not identical, organisms in cyberspace. There are no blind copies or duplicates of identical natural processes. Single-mode approaches won't work here: real living organisms are too "messy" for them.

To ease the task of creating new A-Life form, the tricky process of visual pattern recognition is skipped, so every Creature has fairly abstracted sensory data feed to its brain. On the other end, "action scripts" are used to sequence the individual muscle movements and environmental interactions, rather than attempting to develop a neural system of linked oscillators or other means of motor control. That leaves us with "only" focus of attention and action selection tasks.
Scientists at CyberLife started by designing rich and highly configurable neurons to act as a flexible toolkit from which to construct the various parts of the artificial brain. Hebbian weights are used to modulate input signals, employing the chewing gum strategy to define a system that can learn strongly from a single reinforcement episode but that quickly forgets most, but not all, of that experience. This results in a short-term memory that prevents embarrassing repetition, while the long-term memory enables the system to learn from the probability of reinforcement as well as its intensity. The Norn's brains is divided into two clusters of neurons (lobes): one for the episodic-memory (concept lobe) and the other for action-selection (decision lobe) mechanisms. The punishment-and-reward mechanism is implemented by means of chemical reactions between drive chemicals and drive reducers or increasers.
So, at the end we have a complex system for designing autonomous, intelligent creature, capable of learning by its mistakes in a complex, noisy world. The details of it are far beyond the scope of this article. But don't worry: just follow the links below, and you'll find piles of clear and concise articles that will make things very clear...

CyberLife's long-term goal is to create human-level intelligence by the year 2020. Great! It will be fun to play with this and new versions, and to learn so much about A-Life, cyberbiology and related areas in such easy way.

Similar site(s) of interest: