Does it make sense to speak of artifacts in computer simulations?
"First, simulation is a kind of experiment, and as such brings with it problems of creating experimental artefacts, raising questions about how to distinguish genuine characteristics of behaviour from artefeactual ones created by the technology of manipulation." -- Mary Morgan, The World in the Model, p. 331
This distinction makes perfect sense when considering something like a telescope. So, there was nothing nutty about Galileo's doubters wondering whether those little blobs that appeared near Jupiter were really up in the sky, or were just productions of the telescope itself ("the technology of manipulation"). Galileo, in fact, in one of his observations, wound up drawing a moon that wasn't there: early telescopes were not easy to use! And I believe I recall reading that one of the recent "false positives" for creating cold fusion was due to just such an instrumental artifact.
But in a computer simulation, everything is a product of the technology of manipulation! There is nothing at all "really there" before the computer puts it there. I wonder if it is mistake to carry this distinction over from physical to computer experiments?
This distinction makes perfect sense when considering something like a telescope. So, there was nothing nutty about Galileo's doubters wondering whether those little blobs that appeared near Jupiter were really up in the sky, or were just productions of the telescope itself ("the technology of manipulation"). Galileo, in fact, in one of his observations, wound up drawing a moon that wasn't there: early telescopes were not easy to use! And I believe I recall reading that one of the recent "false positives" for creating cold fusion was due to just such an instrumental artifact.
But in a computer simulation, everything is a product of the technology of manipulation! There is nothing at all "really there" before the computer puts it there. I wonder if it is mistake to carry this distinction over from physical to computer experiments?
I had a computer game once where you would find ruins left by a different species who went extinct thousands of years previously. So there were artifacts in that computer simulation, for sure...
ReplyDeleteOn another note, my son was asking me how they program computers, and I got into the binary code / machine language / interpreter language stuff. He asked me why computers used base 2.
My instinct was to say because it corresponded to the on/off state of an electric circuit, but I stopped when I realized I was bluffing. Any insight, you who are master of several fields?
Your instinct was right. Vacuum tubes apparently also easily transmit an on or off state. You could do all this with plumbing, too: read water flowing as '1' and no water as '0': it would just be a LOT slower!
DeleteYes. Shannon mapped two valued boolean logic to on/off electrical circuits. It was probably the most influential master's thesis in history.
DeleteSee the Soviet Union's Setun which used ternary logic. Two states are much easier to represent than three or more since there are no intermediate values. We humans are used to base ten, but this came from the Arabs. Babylon had a base 60 numeral system. Rome, on the other hand, didn't have a base system at all.
DeletePerhaps we need to distinguish between three things here: reality, the model, and the technology that implements the model.
ReplyDeleteIs an artifact an error in how the model approximates reality, or an error in how the technology implements the model?
Yes, I think that is right, Matt. But we have a name already for the last category: a bug.
DeleteIf an artefact is an effect introduced by the method, why wouldn't computer simulations be full of them? Numbers are rounded, time measures corespond to intervals not instants, etc. that is why computer simulations should really be thought of as predictions rather than experiments.
ReplyDeleteYou make assumptions to create the model, then experiment to see how those assumptions play out. Certainly none of my recent models predict anything.
ReplyDeleteSure. Why not? Using binary floating point, for example, introduces errors in certain decimal calculations.
ReplyDelete