simulating large numbers of physical objects accurately, with friction
object, device, and human animation: how do you do it?
how do you deal with roundoff errors?
large amounts of computation needed for physics and agent simulations
handled both via multicore and distributed computations
tough issues dealing with latency, node failures, and cheating/security
a lot of physics is "faked" right now: objects are scripted instead of simulated etc.
human motion is generally handled via motion capture
landscapes etc. are visually approximated to keep the polygon count low; illusion falls apart if you get too close
detection and visibility (for combat simulation) is usually faked, leading to unrealistic behaviors both ways
interactions proceed through a series of scripted stages
NPCs are largely scripted, not intelligent
usually, little or no memory, no understanding or intelligence
usually, no natural language understanding (except for early text adventure games)
path finding; occasionally, some swarming behavior
only area of limited intelligence: combat strategies
great potential for AI, self learning
AI, learning and artificial evolution for better game characters
learning of motor behaviors for better human/animal simulations
better and more interesting combat/game strategies by NPCs
natural language and speech interaction instead of menus
more realistic perceptual systems on NPCs to make sneaking/combat more realistic
intelligent agent simulation
internal world models
brain machine interfaces
What was your (implicit/explicit) view of reality prior to hearing of the simulation hypothesis?
How does that differ from the simulation hypothesis? Contrast and compare the two views.
What is your conclusion? Do you believe that it is likely that we live in a simulation? That it is possible?
physics, all known physics is representable in terms of computations, otherwise: hypercomputation, in fact, it may be representable in terms of cellular automata, Conway: Game of Life, Game of Life, Game of Life in Game of Life, Game of Life in Minecraft, Fredkin: Finite Nature, Wolfram: A New Kind of Science
minds, "simulated minds" require strong AI, game-style simulations have "external minds" (players) plus NPCs, external minds / players may or may not choose to be aware of their presence in a simulation, possible positions, you and everybody else is a simulated mind, you are an external mind but not aware of it, you are not a simulated mind but there are external minds presents, you are an external mind and aware of it (please tell us all about it), Which would you rather be? Is the "real" world necessarily better than the "simulated" world?
entertainment, strong incentive to create "game worlds", game worlds are often oriented at / based on some historical context / events, together with non-historical elements, strong drive towards making NPCs smarter and more real: it's more entertaining that way, potentially big applications for AI research in creating NPCs and "game AI", example: WoW, Skyrim, eXistenZ
communication, virtual worlds serve as a communications medium, replaces physical travel, while still providing many of the benefits of physical presence, example: SecondLife
history, economics, sociology, business, multi-agent simulations, try to learn about, understand, and predict how real populations of people act under constraints, the more realistic the simulation of individual agents, the better the predictions, already useful for low-levels of individual intelligence
AI research, simulate agents, robots in a simulated environment, try out different kinds of architectures, assumptions, etc., the simulated environment can be quite different from the real world; the main goal is to gain insights
robotics, simulate agents, robots in a simulated environment, the purpose is to create an agent that can then function in the real world, the simulated environment needs to be very close to the real world, example: lots of robot simulators (including for Mars probes)
"internal simulation", or modern version of the dream hypothesis, we have seen that "simulation of reality" or a world model may be an important aspect of building an intelligent system, this world might be the internal simulation of a much more powerful intelligence, part of another mind, example: internal robot planning
Nick Bostrom paper kicked this off
the "reality" we experience is an actual simulation, running on some powerful computer somewhere
there are scientific reasons for believing so
civilization types, current ("human"): some technology, some computation, post-human: enormous computing power, capable of running full simulations of entire worlds
one of these three hypotheses must be true, EITHER civilizations never reach a post-human stage because they kill themselves first, OR civilizations aren't interested in running "ancestor simulations", OR we are living in a "ancestor simulation" created by a post-human civilization
Bostrom only believes (3) with probability less than 50%
the simulation argument implies a creator and a motivation for creating the simulation
note the connection with traditional religious views; however, theories about the motivations of the creator are very different
brains or people connected somehow to a simulated reality
external interfaces, direct brain interfaces
the world contains both "real people" and NPCs
World of Warcraft
primary motivation: entertainment, training
the entire world is digital
all characters in the world are "simulated" (and hence AI within a virtual environment)
character / minds may have initially been created through scanning and uploading
primary motivation: research in economics, AI
minds can both enter and leave the simulation
real mind -> simulation
simulated mind -> real body
primary motiviation: robotics