Philip wrote: > > Some people feel uneasy about trying to take reality and break it up > into pieces - since everything is all part of the seemless web of the > cosmos. > > But thinking of everything as one is too hard so we need to be able to > modularise reality to make it tractable.
This touches on something I've been thinking about, so I'll take this opportunity. Context is one of the things that really stands out when exploring memory processes of the type used in the evolutionary tract. I've seen context mentioned repeatedly within the Novamente online documents, but context is a rather messy word. I'm wondering if some clarification can be provided in naming subdomains of the concept of "context" as regards the discussion of AGI's. Consider the types of context that occurred to me on browsing the Novamente Q&A section: PsyNet Context: There are multiple types of collections nodes and links, different levels of the hierarchical mind network, attentional focus, etc Conceptual Context: The context surrounding a particular concept (person, Paris, running) Semantic Context: The underlying semantic context of a sentence Environmental Context: Is the AGI talking in a chat session, moving blocks in SpaceWorld, or solving a bioinformatics problem? Spatial Context: Assuming the AGI is able to interact with a 2d or 3d spatial domain, the context of spatial location is extremely important Temporal Context: This is an idea from recent neurophys research, indicating the brain maintains a strict sense of temporal context, which has potent effects on various kinds of memory. I see that Novamente has Context and NumericalContext Links, but I'm wondering if something more is needed to handle the various subtypes of context? As far as brains are concerned (and again, I know you're not building a brain per-se, but my AI philosophy involves taking lessons from evolution where we can), context is possibly one of the most important variables in any situation. A rat trained to do action X (push a bar for food) in context A will not perform it in context B, although can learn to do it much more quickly and will learn to generalize across contexts rapidly. The important thing is that they won't do it by default, but will instead explore and learn about the new context instead of pressing every bar they see in a new room. This is a powerful optimization technique by evolution for dealing with the array of situations that even something as simple as a rat will have to deal with over the course of their 2-3 years. I mentioned temporal context above. It appears to work like this (based on human verbal recall data). If I give you a list of words and ask you to recall them in any order, you will recall them in an order predicted by temporal context. If you recall word N, your next recall will most likely be word N+1 or N-1 (some assymmetry in the forward direction), followed by N+2, N-2, etc. Temporal context also extends across much larger spans in human memory, with an increased chance of recalling something from a given time point in your life, given a prior recall of something in the same temporal vicinity (something we've all experienced). Context even extends down to the pharmocological level. A large dose of heroin that an addict might find merely satisfactory in his home might be lethal when taken in a completely novel area. In summary, it's clear that context is a vital part of memory processes used by NGI's, and I was wondering to what extent context is emphasized in the design of Novamente. It's difficult to get a feel for it from the available documentation. I'd also like to explore the idea of creating some more concrete words for the various types of context that will be a necessary part of any AGI. The word "context" is too generalized to perform the many functions required of it. Agree/disagree? Am I reinventing the wheel? -Brad ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
