Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uunet!wuarchive!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!dali.cs.montana.edu!milton!wex@dali.pws.bull.com From: wex@dali.pws.bull.com (Buckaroo Banzai) Newsgroups: sci.virtual-worlds Subject: Re: "Space" Message-ID: Date: 7 Aug 90 22:29:33 GMT References: <9007250107.AA01311@hitl.vrnet.washington.edu> Sender: hlab@milton.u.washington.edu Organization: Bull Worldwide Information Systems Inc. Lines: 84 Approved: hitl@hardy.u.washington.edu for virtual-worlds@milton.u.washington.edu (from news@pws.bull.com (Remote NNTP postings)) id ; Tue, 7 Aug 90 18:25:46 EDT Nntp-Posting-Host: dali.pws.bull.com In-Reply-To: schraudo%beowulf@ucsd.edu's message of 6 Aug 90 22:38:17 GMT In article schraudo%beowulf@ucsd.edu (Nici Schraudolph) asks some very good questions: Buckaroo's suggestions don't strike me as useful for either visualization, a navigation or manipulation of virtual objects: Well, you're welcome to come up with your own. That's the point of a theory after all. What I have is a theory and some experience with small-scale implementations of it. I'm always ready to be proven wrong. To visualize an object in it you have to somehow embed the semantic space in our plain old 4-D spacetime. The entire cognitive engineering problem resides in the design of this embedding, yet you are very vague about it. Could you tell us specifically what sort of embedding you had in mind? You're right, I was pretty vague. The mapping is really pretty arbitrary. What you do is take the three most interesting properties and map them to X, Y, Z. In doing this, of course, you want to take advantage of peoples' cognitive skills and expectations. Usually you have things increase left to right, you try not to clutter the display, etc. The basic semantic space theory provides for visualization of property triples. In order to do more than that, you have to move to the automatic- icons world. This idea is described in a paper I published with Kim Fairchild and Greg Meredith in Interacting with Computers. It's also available as an MCC tech report. Anyway, for those who don't want to track down the report: it's easy to see that X, Y, Z position are only three of the representational properties one can use. Size, shape, color, and texture are all obvious. Velocity and acceleration are less obvious, but usable. With more advanced output devices, we can also map to flexibility, softness, and so on. In general, what you're doing is producing a mapping from object space to icon space. Our paper describes the mathematical foundations for this and gives some examples from a software reuse system. Humans have intuitive navigation abilities only for 2-D spaces, and fare very badly in higher dimensions and alternative geometries. Ever played "wumpus"? It's played on a 3-D dodekaeder, and even though that's a comparatively simple space you usually get lost after a few moves. Humans can navigate complex spaces only with the help of cognitive artifacts (maps, paper & pencil, ...), and using them in VR therefore defeats the whole idea of VR as a user-friendly interface. I disagree on two points. First off, people do pretty well with flight simulators, so we *can* go to 3D (in fact, we spend most of our waking lives navigating in 3D). Wumpus is a kind of weird space; not very intuitive. Second, who says cyberspace should be devoid of cognitive artifacts? I think there should be lots. Again in a past paper (with Kim Fairchild), we talk about navigating cyberspace. To do this, we hypothesize the construction of whole new classes of artifacts to act as navigational aids. For object manipulation the data structures used should accommodate the most frequent operations used on the data. A vast sparse product space (which is what Buckaroo suggests) strikes me as the LEAST useful organization of data. You're confusing theory with implementation. One can implement sparse arrays in a number of compact, useful forms. Why not do the same for spaces? Remember, the prime purpose of semantic-space theory is to provide a natural locative system for abstract data. Why not use VR implementations of such time-honored computer science data structures as trees, inverted lists, priority queues, and so on? Don't confuse *presentation* with *representation*. You can look at your data any way you like. That's like comparing the tabular output of an SQL query with the record structure of the queried database. But to more directly address your point: if your tree contains 10,000+ nodes it's going to be *damn* hard to find any particular node you're interested in. Time-honored computer science data structures just don't deal well with huge volumes of data, and hierarchical divide-and-conquer strategies only go so far. I'm attempting to suggest something to supplement these things. If we are going to connect up all the world's data, we're going to have to invent new ways of dealing with it. -- --Alan Wexelblat phone: (508)294-7485 Bull Worldwide Information Systems internet: wex@pws.bull.com Today is Hiroshima Day. Rest in peace 200,000+ innocents