How accurate are these simulations compared to the real world ? I understand the reasoning behind it, and it makes sense, but it strucks as an oversimplification of really complex things.
Most of the simulation results for things like this are dubious at best. Seems like money better spent to me to focus research energy on more useful endeavors. However, some simulations are extremely useful and really do represent reality well.
Personally I think that this is likely just because it's often pretty easy to get realistic seeming results from simulations which are subtly unrealistic.
The trick here isn’t “accurate” simulation, it’s that they used a bunch of different simulations with randomly perturbed physics and the RL learned policies that worked across these wide range of “realities”.
Trying to use them to model real world scenarios would be useless in practice, due to the Ludic fallacy [0]. Real life is too complex to be modeled in any simulation.
Are simulations just as real? Or are they only culturally real? A simulation of the weather only makes sense when you have humans around to interpret the output from displays. Inside the machine, it's just a lot of 1s and 0s. It's not even really that. It's a lot of electrons moving about. A bunch of electrons aren't a simplification of the weather. It's only because human culture has computing devices that simulations make any sort of sense.
Physical systems aren't about anything, and don't represent anything on their own. It's the entire problem of intentionality.
Its interesting to see how the simulations looks like they are modeling society, and reminds me strongly on how game of life tries to be a model of artificial life. however, if someone want it to mean more then they really need to test the model and get data on how well it can predict.
Well yes, I mean, all simulations are, right? You put as much real-life data in as you can, but everything's an abstraction unless you're the team from the Devs tv series.
I think some of this falls into the simulation paradox: the more accurate the simulation, the closer the simulation is to the thing being modelled. But it's a quadratic relationship in most cases, so at some point meaningful increases in simulation accuracy cease to be economically viable.
It is more like comparing the game to real life to see how they hold up and teach how real life works than using a game simulation to say that's how life works.
It's like asking a master spy for his opinion on spy movies. They tell you what's real and what's bunk.
I worked with several academics and corporations as a contractor building their simulations. I’ve also read a lot of papers as well, etc.
Like most simulations, they pretty much only can match reality so closely. for instance, we don’t have many long term measurements, so we make (educated-ish) guesses. Academics also repeatedly run their models, tweaking parameters to get the results they want. Frankly, I don’t trust any research using simulations, after being paid to work on environmental, biological systems, simulations for drones and machines (trucks, construction, etc). (Drones and other machines at least had a way to quickly test).
I'm no expert but my understanding is that the word 'simulation' is a misnomer in this context. A better way to describe what they're trying to do is a proxy system. Effects observed in this proxy system would be used to predict interactions in reality. Whereas simulations are better at describing how real world interactions diverge from our models.
reply