On Tue, Aug 9, 2011 at 4:50 AM, Casey Ransberger <[email protected]>wrote:
> I think being able to point at things and see by the eyes and the angle of > the head what people are looking at (shared attention) are probably pretty > powerful in general. > I think you need to balance that against ability to actually see what the person is looking at. To see both an avatar and the target would imply an oblique angle on both, which could be a phenomenal waste of screen real-estate. In an avatar-less model, you might share attention by other means: tags, flags, subscriptions, RSS, other annotations. In bulletin boards, and mailing lists, we get shared attention by simply pushing to the top that which people have recently commented on. I think the main benefit of avatars would be support for facial language - recognizing irritation, sarcasm, surprise, et cetera. These benefits are not realized with today's technology, except in games such as Heavy Rain that make significant use of them. > > The best way to have a conversation with someone is in person, > I think it depends on the nature of the conversation. There are significant advantages to written conversations, such as: the ability to spend more time thinking about our responses, the ability to operate at different times, and having a written record you can search and reference. A slightly more formal language can be very valuable; I think an interesting experiment would be an argument forum where arguments are mapped out with premises and reduction rules, and the computer system helps us identify plausibility, internal consistency, and locate (and link) relevant arguments and counter-arguments for the premises. It could change how we build arguments. Speaking in person has advantages of body language (which is surprisingly expressive to most people) that helps the speaker recognize where clarification is necessary. But a state-of-the-art 3D avatar won't help much there. We should train a camera on the user and push facial twitches and gestures across the network. > Also, a hand drawn character looks less... creepy than the current state of > the art puppet, even if the puppet is more realistic now. Uncanny valley. > I wonder if cel shading can help a lot with bridging the uncanny valley. It gets you the look and feel of hand-drawn art while allowing state-of-the-art techniques in developing the 3D models. > > I like fractals. But it ends up being a mix in all likelihood. Voxels are > great for doing clouds at a distance, etc. But up close clouds made out of > little cubes aren't very convincing. Minecraft overcomes this by making low > resolution textures and huge voxels a kind of fashion statement. > I suspect you could do some sort of 'anti-aliasing' for voxels, perhaps using GPU shaders. Use of GPU shaders, for example, can turn an ugly bowl of triangles into a rather pretty tree ( http://the-witness.net/news/2011/06/witness-trees/). Beyond that, level-of-detail projections are also quite feasible. > > >> how to make it able to be used at *any* real scale on current HW >> (voxels+FEM is not exactly a lightweight combination). >> > > What if the "simulation overhead" could be distributed to every machine > currently participating? > For gaming, that can become a security/cheating risk that is rather difficult to reason about. I also think distribution-for-performance should not be a first choice. There are probably a ton of useful things we can do with level-of-detail, shaders, high-level culling, etc. > ocular occlusion in order to avoid simulating things that no one is present > to perceive World physics should be fully deterministic and cheap to compute in the absence of external influence. This includes NPC schedules and such. When a metaphor flaps its wings, we should know exactly what the 'game-state' consequences will be - if there are any. If we can reduce the world model to a fixpoint continuous integral - using algebraic signals of time - we can actually compute world state far more cheaply than would be possible with piecewise discrete-state simulations. But, either way, computing game state is often a lot cheaper than computing the animations. Consider, for example: when two NPCs converse (a sort of 'collision' event between NPCs) we might model them as exchanging information, objects, and germs. That would be the 'game-state' consequence of the collision. But the actual animation might involve handshakes, vocal exclamations, speech generation. All of that could be elided in the absence of an observer. The NPCs themselves could similarly be elided if they are part of the 'background' (i.e. to make a city look busy) rather than relevant to game state. The distinctions between game state modeling and animation can thus be a level-of-detail concern. We can carefully add a dose of indeterminism to the game by modeling an external controller - a game master, or more than one - and giving some of those to humans or expert systems. In a multi-player interactive fiction, NPCs but game-significant not in control of a player are in control of game masters. Regards, Dave
_______________________________________________ fonc mailing list [email protected] http://vpri.org/mailman/listinfo/fonc
