Your correct. Lag is bad, adding more lag is going to make the experience a little more frustrating and potentially unplayable. What I was trying to say was that games already have lag, although that doesn't excuse adding more lag, but it does show that if the lag were reduced enough if could be negligible in comparison. Now perhaps, on this platform, hypothetically speaking, lag could be reduced in the game so that the physical update and rendering did not take a standard 16ms/33ms (60fps/30fps) amount of time, but instead took a much shorter amount of time, because in hypothetical land they have big servers with vast processing power. You could at least reduce the lag a little between providing the input and producing the associated rendered frame. Although it's getting a little late to engage my logical brain, but maybe this is possible using some clever time slicing. I would say that a game would be to be specifically coded to support such a feature, although I don't think it would be too much hassle (thinking along the lines of an update then render, rather than running both in parallel but rendering last frames information, which is currently the norm).

In the video he talks about having a sub 20ms ping. I think the idea is that they would setup lots of smaller servers spread out geographically to reduce the amount of lag as much as possible. What people perhaps overlook is that games naturally have quite a large lag already, once you've pressed a button it takes up to 1 frame for that change to be registered, another frame to update the physics / animation etc, and finally a frame to render based on the previously calculated physics information. In a 30FPS game that's between 66-100ms, and that's assuming a really damn good engine which is responsive, which a lot of game engines aren't. There was an article on Gamasutra on this very topic about a year ago, if you want to read more. If the check out the third page of that article you'll see the response times for some popular games, and you might be suprised!

Game logic and physics are both required to be deterministic, I think random error in these systems would be bad. Some parts of rendering need to be error free, such as the vertex shaders (you don't want your triangles wobbling all over the place), but their accuracy becomes less important with distance (although z fighting may be an issue), whereby a lower level of detail on calculations at the cost of speed is currently done in algorithms already. Graphical rendering and physical simulation of choatic systems, such as weather (wind, rain, snow, heat haze), water, and systems whereby choas wouldn't be so noticeable like reflection maps would not have a detrimental effect on visual quality. I can't see some parts of an AI system not being hurt to badly by this error either.

Sure, but people play simulation games because they're fun :)

Games aren't meant to be simulations, they're meant to be fun.

Visual Assist does a good job of making Visual Studio code a lot easier to read by applying custom colours to various syntax. It also does a load of other stuff, as a C++ developer it reduces the amount of typing I have to do by half by having a really good intellisense system. Now all they need to do is write a program to automate the other half :)