Richard, I've been too busy to participate in this thread, but, now I'll chip in a single comment, anyways... regarding the intersection btw your thoughts and Novamente's current work...
You cited the following 4 criteria, > > "- Memory. Does the mechanism use stored information about what it was > doing fifteen minutes ago, when it is making a decision about what to do > now? An hour ago? A million years ago? Whatever: if it remembers, then > it has memory. > > > > "- Development. Does the mechanism change its character in some way over > time? Does it adapt? > > > > "- Identity. Do individuals of a certain type have their own unique > identities, so that the result of an interaction depends on more than the > type of the object, but also the particular individuals involved? > > > > "- Nonlinearity. Are the functions describing the behavior deeply > nonlinear? > > > > These four characteristics are enough. Go take a look at a natural system > in physics, or an engineering system, and find one in which the components > of the system interact with memory, development, identity and nonlinearity. > You will not find any that are understood. Someone else replied: > > I am quite sure there have been many AI system that have had all four of > these features and that have worked pretty much as planned and whose > behavior is reasonably well understood Actually, the Novamente Pet Brain system that we're now experimenting with, for controlling virtual dogs and other animals, in virtual worlds, does include nontrivial -- memory -- adaptation/development -- identity -- nonlinearity Each pet has its own memory (procedural, episodic and declarative) and develops new behaviors, skills and biases over time; each pet has its own personality and identity; and there is plenty of nonlinearity in multiple aspects and levels. Yet, this is really a pretty simplistic AI system (though built in an architecture with grander ambitions and potential), and we certainly DO understand the system's behavior to a reasonable level -- though we can't predict exactly what any one pet will do in any given situation; we just have to run the system and see. I agree that the above four features, combined, do lead to a lot of complexity in the "complex systems" sense. However, I don't agree that this complexity is so severe as to render implausible an intuitive understanding, from first principles, of the system's qualitative large-scale behavior based on the details of its construction. It's true we haven't done the math to predict the system's qualitative large-scale behavior rigorously; but as system designers and parameter tuners, we can tell how to tweak the system to get it to generally act in certain ways. And it really seems to me that the same sort of situation will hold when we go beyond virtual pets to more generally intelligent virtual agents based on the same architecture. -- Ben G ------------------------------------------- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244&id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
