Harry,
Obviously this is an issue any intelligent AGI has to deal with. However, at high level I don't think it is that mysterious, although, like most things in AGI, in detail it would have quite a few wrinkles, most of which a properly designed AGI should learn to deal with automatically. At a high level, the concept of an individual physical object that has a continued path through space time, is one the system learns from grounded experience --- and the system learns to label certain sets of perceptions as corresponding to such an individual object, based on experiential knowledge about what perceptions are likely to correspond to an instance of such a type of object. This is a refinement of the concept of the persistence of objects which babies learn by, I think it is, six months --- made more sophisticated by understandings of the probabilities, under differing circumstances, that what appears to the same physical object, might actually be a different one. When I did most of my thinking on this I thought about diet coke cans, since they are often a common object in my environment, and since individual instances of this type share so many similar traits. Yet still there are might be attributes, which one might associate with one particular set of diet coke can perceptions which can convince your mind to different degrees that they correspond to the same physical object, rather than to two or more very similar objects. Such information can include something obvious, like a particular dent, or something less direct, such as a memory of placing a can in the location diet coke can perceptions are currently coming from, in an environment where there are believed to be no other things that could have replaced it with a similar can in the same location. The more exactly it matches your recollection of the position and orientation with which you remember last placing it, and/or the more exactly it matches having the same amount of coke in it, the more likely you are to believe it is the same physical object, even if you are at a crowded party where there are multiple agents capable of having replaced it since you last saw it. An object like a single large tree in the front yard of a house is much more likely to have multiple perceptions of it at different times be labeled as being associated with the same physical object, since the chances that such a tree would be replaced by a roughly similar try in most human time spans is very low, even if the memories of the trees properties are rather vague. Interesting experiments have been done showing the extend to which generally, but not necessarily, reliable assumptions, often play a larger role than accurate perception, in our guesses about continuity of identity. I attended a lecture, where they showed video clips of multiple repetitions of the following amazing experiment. A person in a construction outfit, including hard hat, near a construction site, asks a passerby for directions. While the passerby is pointing in the direction of the asked for path of travel, two other pretend construction workers, similarly dressed, walk between them carrying a 8x4 piece of plywood or wallboard. When this happens, the pretend construction worker who asked the question, grabs the end of the plywood, and is replaced by one of the similarly clad construction workers previously carrying the plywood. This new pretend construction worker stands in the same location with the same stance and expression as the original questioner. In the vast majority of cases, when the passerby looks back to where the questioner was, her or she, fails to notice he was talking to a different person, even though they are separated by only two to three feet. And the passerby continues the brief interchange without any look of surprise or other evidence of noticing the switcheroo. This is true even when the new construction worker was obviously, to any one who looked with any care, of a different sex. So probabilistic reasoning is often involved when thinking about identity is done. Ed Porter -----Original Message----- From: Harry Chesley [mailto:[email protected]] Sent: Friday, January 09, 2009 12:10 PM To: [email protected] Subject: [agi] Identity & abstraction I'm trying to get an idea of how our minds handle the tension between identity and abstraction, and it occurs to me that there have probably been human subject experiments that would shed light on this. Does anyone know of any? The basic issue: On the one hand, we identify two objects as being the same one (having the same identity), even when encountered at different times or from different perspectives. At least a part of how we do this is very likely a matter of noticing that the two objects have common features which are unlikely to occur together at random. On the other hand, over time we make abstractions of situations that we encounter repeatedly, most likely by removing details that are not in common between the instances. Yet it's these very details that let us derive identity. So how do we remember abstractions that are dependent on identity? It seems that there must be experiments or evidence from brain-damaged individuals that would give clues. Example: I may notice over time that whenever object A is smaller than object B and object B is smaller than object C, then object A is smaller than object C. Note that I have to give them names in order to even state the problem. Internally, we might do likewise and assign names, in which case there might be a part of the brain that performs the naming and could be damaged. Or we might go back to the original cases (case-based reasoning). Or we might store references to the original object instances from which we abstracted the general rule, which would provide unique identity. The later two may be distinguishable experimentally by choosing clever instances to abstract from. Anyone know of any research that sheds light on this area? ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?&; Powered by Listbox: http://www.listbox.com ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
