Steve Grand works for CyberLife, and was the lead programmer for the famous Creatures program. He has many years of experience with both neural networks and genetic algorithms, and many other aspects of evolutionary computing. Please visit his homepage to learn more about his interests and background.
1.) Creatures uses a combination of Genetic Algorithms and Neural Networks, in possibly the most complex and realistic simulation of life ever created. What sort of research and development was required for such a massive project?
Er, I sat on a hill for three days and thought about it! (Plus twenty years of assorted thinking and experimentation with neural nets, evolutionary systems, etc.). Actually, Creatures doesn't use a Genetic Algorithm - GA's are a rather abstract thing; Creatures uses something far more similar to real genetics in real organisms.
2.) Why did you choose the Hebbian neural-network?
I didn't exactly *choose* it - the design of the Creatures network is unique, and based on biological and psychological principles (rather than Connectionist principles, which underlie most NN research). It's an action-selection system which uses (approximately) Hebbian learning, but I don't think anyone has used a network quite like this anywhere else. It's only "Hebbian" in the sense of using synaptic weights to control learning and behavioural expression (no excitation patterns, etc.). In fact it uses a fairly odd synaptic weight mechanism, which has in-built habituation - more like real neurones than I think Hebb had in mind.
3.) Many believe that the secret to a fully autonomous program will be one that is *relatively* simple in design - yet is capable of evolution and memory. What is your opinion on this?
I've no doubt that's true, but only if you're prepared to wait around for a bit (say 4 billion years or so) while it evolves into something more complex! High levels of intelligence necessarily require high levels of complexity (as I'm sure I could prove, given a bit of thought), and people do tend to underestimate how much time it takes to evolve things that complex, no matter what clever tricks you use. Personally I think the only practical approach to the problem is to
think hard and do some engineering.
4.) How does the object-orientated programming methodology help in AI programming?
That might take a long time to explain! However, OOP is all about splitting systems up into small pieces that interact with each other in limited and controlled ways. Human beings are machines made from roughly a trillion moving parts (cells)! Nature makes such machines possible by a similar splitting-up and limit on interactions. Object-orientation (or
more strictly component-orientation) is mostly important for managing the complexity you need to make artificial life forms.
5.) CyberLife has recently got into a defense contract, utilizing technology similar to that in Creatures in UAVs. What advantages would such technology have in the military?
Adaptibility, robustness, broad-spectrum intelligence, understandability, self-repair... Much the same reasons as the Military still finds a use for soldiers!
6.) Taking the human out of the interaction loop has its good points - less attrition etc., but also has its bad points. Moral and ethical issues will be brought up. What is your stand on such issues?
Nothing pleases me more than when I stir up moral and ethical questions! I think people are generally very naive and simplistic in their moral judgements and make some terrible mistakes. Artificial Life throws a big spanner in the works that causes people to question their moral attitudes more carefully. This can only be a good thing.
7.) Hollywood takes such issues to the extreme, and often scares even the most thoughtful people - do you think this will have an adverse effect on the advancement of AI?
Probably not. The first question journalists usually ask me is "are these things going to take over the world?", but why should they? They're just another life form, and there are millions of other species out there who haven't "taken over the world". People are bound to be resistant at first, but then they're also scared of impersonal technology, and one of the good things A-life can do is make machines more personable and friendly - people will soon learn to love them!
Steve Grand works for CyberLife, and was the lead programmer for the famous Creatures program. He has many years of experience with both neural networks and genetic algorithms, and many other aspects of evolutionary computing. Please visit his