Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!ogicse!milton!brucec%phoebus.labs.tek.com@RELAY.CS.NET From: brucec%phoebus.labs.tek.com@RELAY.CS.NET (Bruce Cohen;;50-662;LP=A;) Newsgroups: sci.virtual-worlds Subject: Re: The collaborative nature of VR Message-ID: <9315@milton.u.washington.edu> Date: 9 Oct 90 18:08:04 GMT References: <8374@milton.u.washington.edu> <8514@milton.u.washington.edu> Sender: hlab@milton.u.washington.edu Organization: Tektronix Inc. Lines: 236 Approved: hitl@hardy.u.washington.edu I'm falling behind a few days here; they want me to do some of the work they pay me for here! Of all the cheek! I think I'll handle these points out of order, since the one I want to handle first is the most fundamental: I think I see basic point of disagreement between Mike Moore and me , and I think it's based on a misunderstanding. When I say "modelling the real world", I mean that in the sense of a mathematical model, in which the most important aspects of the world (important to the particular model, of course) are abstracted and used in the simulation, and the rest ignored. I think Mike took me to mean a complete model in which all (or as many as we have computing power for) aspects of the world are simulated. And when he disagreed, I thought he was talking about *different* models, rather than less complex models. Consider computer graphics as an example: most realistic shading algorithms are based on physical models, but they all involve certain simplifying assumptions which make them physically incorrect, but computationally tractable. For instance, Gouraud shading gets good performance by ignoring specular reflection, and thus not being able to simulate highlights on shiny surfaces. Phong shading puts in the highlights at a cost of roughly an order of magnitude over Gouraud, but makes some simplifying assumptions about the smoothness of the reflecting surface. Neither of them is a perfect model of the real world, but they are points in a spectrum of techniques which range from highly abstract to highly realistic. But all of these techniques are based on the laws of physics as we know them, they just simplify them. There's another alternative in designing a model: choose one which simulates a world which contradicts the known laws of physics, chemistry, etc. Sort of like playing fairy chess: the pieces and the moves may be wildly different from the standard game; the challenge is to make the game internally consistent, that is, playable. Obviously, in this context "realism" is a relative, and almost useless word (come to think of it, that may be generally true; it's certainly true in the theatre and movies as well). I'm using it in the sense of the mathematical model I spoke of above. So to me, "non-realism" means, not something which is less than real, but something which is contrafactual. I would love to spend time exploring the design and use of contrafactual physical worlds, but I'm arguing against spending much time doing that at this stage in the development of VR. So I advocate starting with as good a simulation of the real world ("consensus reality" if you like) because we know it's consistent, and highly playable. Besides that, there are a lot of existing simulation and display techniques available to the VR programmer, often as publically available code, which are based on the laws of physics. Making them fit some arbitrary set of non-realistic laws is probably more difficult. Now, I think that Mike is saying that we should start with as small a set of "physical" laws as possible, and work up. So let me respond to the question: In article <8514@milton.u.washington.edu> mike@x.co.uk (Mike Moore) writes: : >> 5) VR designers need some experience too. Working incrementally from a >> robust, well-constructed, "real-world" model is a good way to get >> that experience. > > Hmmm.... I see what's happening. You are working on the assumption that > VR will *start* with a realistic 'real-world' reality (hey! that's good!) > whereas I see us starting with nothing (or very close to nothing) and > working up to a realistic 'real-world' reality. I must admit, I think > that my view is more realistic :-) But please explain more of > what you are thinking, I feel as though there's a huge chasm I've > missed somewhere. Why do you believe we'll start at realism and work > down? > As I said, there's a lot of art out there already. And ask yourself what's the mininum reality which a human user will accept as having some physical existence, that is, will give the sense of "presence" which we would like to provide. Do you need inertia for objects? Should they have flat color, highlights, shadows? Are there doors and do their hinges squeak? :-). Seriously, I think we need to spend a lot of time at this stage of the VR technology in answering questions about the modes and fidelity of sensory input and detection of motor output. I think that's easier to do if we keep the model we use as close as we can to the reality which the users already know. I agree that the reality we can generate in a reasonable amount of time (say with display frame rates of 10 / second, and stimulus-response delays of no more than 100 milliseconds) on comomonly available computers is not going to allow high-fidelity "reality", with radiosity-shading and force-feedback on every finger joint. And for the pioneering work which will be done in the next two or three years "reality" will be limited by the small number of people working in the field, and their inability to do it all themselves. But I think the minimum useful set of physical phenomena is larger than most of us here seem to think (one of my pet peeves is the way many people completely ignore sound, both as input and output), and I also think that a good choice of development platform for a VR system would make much of the existing display and simulation software easily available. But that's for another thread of discussion. > Bruce Cohen [brucec@tekcrl.labs.tek.com] writes: >>I think this is a very limited and limiting view of the way the designer >>and inhabitants of a virtual reality interact. The implication is that a >>world is finished and perfect construct, whose designer has considered >>every interaction possible, and knows the nature of every use to which the >>world will be put (or even most of them). This is not the way complex >>systems are introduced in the "real world". Often, particularly in >>systems with complex user interfaces, new interactions and new applications >>for the system come to light for years after the designer has gone on to >>other projects. > > I'm sorry if I gave this impression, that was not what I meant, and I > agree with you that any system as complex as a world-wide network of > VR machines would be impossible to analyse completely. I was looking > more from the point of view that when a new 'room' comes on-line, it is > the responsibility of the designer to ensure that only the people he > wants in are allowed in (see further below). > This is more an issue of security than object design. I make that distinction because I believe that it should be possible to design an object to work in an unsecure environment, and then, as a completely independent operation, install one instance of it in a secure environment, and another in a less secure environment. In other words, objects should not contain policy which controls access to them, they should only contain mechanism which provides the access. > But take another example I came across the other day (from > a famous book I borrowed of a friend over a drink for 10 minutes, so I'm > not accurate [from Mind Children by Hans Moravek]) that is basically a > school, or a VR, AI text book. It gives you individual tuition (on a > time-sharing basis) and even allows you to interact with other students > so you can choose to learn together or on your own. Not in a school, but > walking along an orchard path with Isaac Newton to learn about gravity > and mass, or in a train to learn about the Doppler effect, and relativity > with Einstein (or, at least, VR AI interpretations of them). Now you > *don't* want people messing around, they have a limited set of allowable > actions. They won't bring anything with them, and they probably won't > take objects away. Each VR environment will have it's own set of > constraints on what can and cannot be done, and collaboration is merely > one example of one set of 'constraints'. > Again, I consider this a use issue, not a design issue. In other words, the designer provides a way for you to allow or disallow others from using your instances of the objects, you decide who gets to use them and how. In the meantime, I get to make different decisions about my instances of the objects. > To explain my point in > this context, if I were denied access to a room by some security software, > it would be in the best interests of the software to not even inform > me of the existence of the room. i.e. on entering an area I must > identify myself and then the software informs me of what it wants me > to know. (see below) > That may be the way you want your doors to work. But if we both buy doors from the Acme Virtual Door Company, and you want yours only to appear to your friends, whereas I want mine to appear to everybody, but be locked between 12 and 4 everyday, and have a sign on it to that effect, I don't want to have your behavior wired into my door. > 1) The universe (as we know it) is consistent only on a macro level. > Half of the study in nuclear physics is currently revolving > around Quanta (i.e. the *probability* that if cause A occurs > effect B will seen rather than effect C). This gets a little astray, but I don't think this statement is true. QM is perfectly consistent, they're just not deterministic. The difference is between whether the underlying world works so that one set of rules never causes something ruled out by another set (consistency), and whether the outcome of some set of events is completely controlled by the past history of events along any worldlines entering the events (determinism). > 2) To model the real-world we would need to model the non-perceptible > parts of the world as well. i.e. Solar radiation (non-perceptible, > at least without instruments) acts on the upper atmosphere > (non-perceptible, unless you happen to be there) and the Magnetosphere > (non-perceptible) to produce the Aurora Borealis, which *is* > perceptible. As I said before, I think this is where we got confused. I'm not advocating modelling everything. I may take up the question of what constitutes the appropriate granularity of the world we need to simulate in a later posting, but right now, let me just say that we don't have to even model everything which is perceptible, and those things we do model may be simplified. > True, but when computers hit the market for the first time, there were > very few experienced users. When cars hit the market for the first time, > no one could drive. People *do* learn (when motivated). > True, but what motivates them? Why should the average computer user (who according to everything I've seen uses at most one word processing program, one calendar, and one spread-sheet, running on DOS) use VR? I contend that the reason for developing VR in the first place is to find ways of better presenting the computer's power to the average user, and I specifically reject the argument that they'll learn to use it because they have to. Which is to say that a kind word works better than a gun :-). >> : >> 6) I suspect we'll learn more about what we can do to modify VR models >> from letting the VR inhabitants mess around with the rules than any >> other way. That means starting with a baseline, and giving users >> the tools to modify it incrementally. Along the way, we have to >> learn how to reconcile conflicting world-views as well. > > Now this I don't understand. From your point of view, if it is a good > model of reality, how is it possible to have conflicting world-views > (other than psychologically). From my point of view, every gets there > own individual world view, just the same as you get your own individual > copy of this article. Please elaborate. > What I meant here was that each user could customize the interface to the system so as to see common objects in different ways, like having multiple views of the same text, with different fonts and formatting, in different windows. The problem is reconciling the effects of actions by the users on the objects, such that an operation which one user can perform has a visible and comprehensible result on the view of the object which another user has. What makes this complicated is that the operations may be abstract actions which affect the meaning of the object to the user, and be translated into the object's attributes in subtle and indirect ways. A poor example that comes to me offhand is one user opening the manila folder which represents a container of data files which in this view are pieces of "paper", while being watched by another user who sees data files as simulated terminal screens. -- --------------------------------------------------------------------------- Speaker-to-managers, aka Bruce Cohen, Computer Research Lab email: brucec@tekcrl.labs.tek.com Tektronix Laboratories, Tektronix, Inc. phone: (503)627-5241 M/S 50-662, P.O. Box 500, Beaverton, OR 97077