News for hackers. Stuff that matters.
Re:Sony’s racing AI destroyed its human competitors by being... by mrandish on Wednesday July 20, 2022 @03:28PM (#32170926)
I think this is material info which needs to be noted:

> GT Sophy doesn’t get the same view of the game that human players do. Instead of reading pixels off a screen, the program takes in updates about the position of its car on the track and the positions of the cars around it. It also gets sent information about the virtual physical forces affecting its vehicle. In response, GT Sophy tells the car to turn or brake. This back-and-forth between GT Sophy and the game happens 10 times a second, which Wurman and his colleagues claim matches the reaction time of human players.

The AI is getting precise 3D position data and physics data not only about its car but competitors cars - from a 3D physics simulation. The human competitors only get a 2D visual view based on what a race car driver can see from the cockpit. Maybe I'm misunderstanding but it seems like the AI is simply deriving the optimal parameters encoded in the 3D physics simulation through iterative testing based on precise 3D physics data supplied directly from the simulation. The humans aren't getting the same data. The human drivers must visually estimate the physics vectors and precise 3D positions from a limited yet rapidly changing 2D rendering. This is quite different from what I initially assumed based on the headline and opening paragraphs of the article.

Given the same precise 3D and vector data in searchable form along with sufficient compute power, it seems plausible a team of human statisticians could work out more optimal driving parameters than top human drivers limited to 2D visuals in nine 24 hour work periods (which is what the AI required to train).

Re:Sony’s racing AI destroyed its human competitors by being... by andsoitis on Thursday July 21, 2022 @11:29AM (#32180556)
That's why humans should insist on competing with AIs in our world not theirs (simulation).

When it matters, the playing field will be stacked heavily against the human who will not have direct access to the informed encoded in the virtual world. Putting it differently, any AI in a simulation is in some sense indistinguishable from the simulation itself (or at least, you should assume so).

Re:Sony’s racing AI destroyed its human competitors by being... by nomel on Thursday July 21, 2022 @06:02PM (#32185772)
> with AIs in our world not theirs

The benefit of keeping things in "their world" is that "our world" requires massive investment in sensors, while the exact same algorithms would be used. Letting them play in their world means you can focus on the algorithms. But, I agree that it should be fair. Putting a human in front of a screen with a steering wheel is in no way comparable to the sensory experience of being in a car. I'm fine around a race track, but racing simulators are nearly impossible for me.

Re:Sony’s racing AI destroyed its human competitors by being... by raz32dust on Thursday July 21, 2022 @12:57PM (#32181786)
It's hard to draw the boundary. We could say that AI should only be given access to exactly the same information that humans have access to. But AI will have magnitudes more memory and compute resources. So for example, an AI could only have the 2D view but it could store, access and process the 2D view in ways a human cannot.
Re:Sony’s racing AI destroyed its human competitors by being... by andsoitis on Thursday July 21, 2022 @03:53PM (#32184086)
> We could say that AI should only be given access to exactly the same information that humans have access to.

I believe that it would be naïve to believe the owner of the AI (whether a person, and organization, or another AI) would play by those rules. Better to assume an adversarial game board.

One approach would be to simply not compete against an AI even in our world. Another would be be to be in control of the framing or environment where you can outplay, outlast the AI. Do not assume friendship because its goals will be different from that of humans, unless you can be certain of mutual dependency.

The game theory around human/AI coexistence extends beyond these quick thought bytes, maybe someone is working one a treatise.

Re:Sony’s racing AI destroyed its human competitors by being... by EricMausler on Thursday July 21, 2022 @03:57PM (#32184139)
It has to do with input data. Part of the game is rationalizing what is on the screen and maintaining a memory or prediction of offscreen activity. When they get direct feeds to game data, it's cheating.

The AI should be given a game screen. I think it should also need to actually use a mouse and keyboard too if im being honest.

I would be way better at games if I only need to think about what I want to do in the game and have it done rather than use some method of physical controls to do so

Re:Sony’s racing AI destroyed its human competitors by being... by runnerup on Thursday July 21, 2022 @07:13PM (#32186422)
> I think it should also need to actually use a mouse and keyboard too if im being honest.

I'd go one step further and say those mouse inputs should be "fuzzed" a bit as a function of "actions per minute" because humans lose accuracy (both in the time domain and spatial domain) when they try to make more actions faster and faster. Obviously APM needs to be capped as well, preferably actually slightly lower than human APM.

The main reason for this to me is that it will help guide the AI's towards generating strategies which humans can actually emulate and learn from.

If you want to have an AI-only tournament, then by all means make up your own rules. If you want to say your AI "beat a human" then you have to be playing the same damn game.

Re:Sony’s racing AI destroyed its human competitors by being... by batmanthehorse on Thursday July 21, 2022 @02:57PM (#32183343)
Yes, but more memory and processing power might mean the AI is actually better at performing the same task. Cleaner, pre-processed input data is cheating in a different way.
Re:Sony’s racing AI destroyed its human competitors by being... by RajT88 on Thursday July 21, 2022 @11:56AM (#32180913)
It caught my eye as well.

If you imagine a real car, this is the sort of detailed sensor data you would get from it. The sensor data in the game is never wrong though, as it comes direct from memory.

The other thing which caught my eye was this:

> For example, GT Sophy often drops a wheel onto the grass at the edge of the track and then skids into turns.

This works consistently because (again) it's not the real world where the grass has subtle variations.

It would be very interesting indeed to see such an AI compete on a real track, but it may not be as clear cut a victory when the very finely tuned senses of a human racing driver are in play.

Re:Sony’s racing AI destroyed its human competitors by being... by richiezc on Thursday July 21, 2022 @02:08PM (#32182698)
>> For example, GT Sophy often drops a wheel onto the grass at the edge of the track and then skids into turns.

> This works consistently because (again) it's not the real world where the grass has subtle variations.

Well I mean maybe they just trained it on Initial D anime, Sony is a Japanese company after all

Re:Sony’s racing AI destroyed its human competitors by being... by ascagnel_ on Thursday July 21, 2022 @03:13PM (#32183567)
> It would be very interesting indeed to see such an AI compete on a real track, but it may not be as clear cut a victory when the very finely tuned senses of a human racing driver are in play.

Roborace[0][1] was an attempt at an all-AI racing series that would be run parallel to Formula E (the all-electric single-seater racing series); however, while they did get to the point that they could run cars on track, they never got to the point where they could run those cars in competition. It seems to be defunct, having been renamed "Arrival R" after the core technologies were purchased by one of the teams that had intended to enter the series.[2]

0: https://roborace.com

1: https://en.wikipedia.org/wiki/Roborace

2: https://www.autocar.co.uk/car-news/business-tech%2C-developm...

Re:Sony’s racing AI destroyed its human competitors by being... by cdkmoose on Thursday July 21, 2022 @03:43PM (#32183941)
F1 drivers do the same thing through a chicane or wide on a curve to maintain speed, leaving the smoothest road surface for that advantage (not the skidding).
Re:Sony’s racing AI destroyed its human competitors by being... by t0mas88 on Thursday July 21, 2022 @04:46PM (#32184804)
That's about taking a different line. Drivers will start as far out on the kerb as possible to make the overall turn less tight, allowing them to take that turn at a higher speed.

What this AI was doing is something that only works in sim-racing. It's done by humans sometimes in iRacing. Drop a wheel on the grass to make use of the fact that the sim will cause a consistent rotation (yaw) from it. Which can help turn a car into a corner.

It can't be done in reality, because the effect on the car isn't predictable on a real race track.

Re:Sony’s racing AI destroyed its human competitors by being... by MikeBVaughn on Thursday July 21, 2022 @07:46PM (#32186700)
The modeling in iRacing is really striking in that way. It's good enough in practice that driving in the sim is extremely convincing, which makes the odd methods of exploiting the physics model that much more striking.

I saw a video a few months ago of people running an endurance race in iRacing, and they'd consistently drop their left-side (IIRC, it could have been right) tires off the track on a specific straightaway. My understanding is that a) they didn't lose that much speed, b) because of the reasons stated previously, it's fairly safe in the sim, and c) it kept the front left tire temperatures closer to the optimum for the full duration of the lap. Thinking on it, one of the things I found cool was that, in a way, it's not entirely unlike what real drivers will do with wet-weather tires, where they sacrifice the best line to hunt for wetter stretches.

Re:Sony’s racing AI destroyed its human competitors by being... by rad_gruchalski on Thursday July 21, 2022 @06:14PM (#32185868)
> It can't be done in reality, because the effect on the car isn't predictable on a real race track.

Maybe a driver could work it out in training for a single race and in a certain corner but the cost of getting it wrong is not only not finishing the race but also having a race car in the barrier, so a damaged car.

On a race track, drivers usually know characteristics of every corner for selected weather conditions. For example: at the Nordschleife, exit of Foxhole into Adenauer Forst is safe to cut in the dry but a no no in the wet, unless you’re in a GT3 class car with enough downforce.

It is the same with kerbs, stay out of then when it is wet, use them in the dry. But it’s all touch and go when the conditions change and sometimes that leads to a crash.

Re:Sony’s racing AI destroyed its human competitors by being... by lowbloodsugar on Thursday July 21, 2022 @02:43PM (#32183171)
Fundamentally, despite the claims, Gran Turismo isn't a very good driving physics simulator.
Re:Sony’s racing AI destroyed its human competitors by being... by mjburgess on Thursday July 21, 2022 @11:26AM (#32180519)
It's always pseudoscience.

AI models correlation(Measures(X), Measures(Y)).

Marketing sells it as using a theory of how X causes Y.

In this case, we want a machine to acquire a causal model of how its actions produce effects, by measuring X and Y itself under its own actions.

There is no AI that does this. What we have is a machine "with the answers already" (ie., no need to measure) that has a correlative model.

Re:Sony’s racing AI destroyed its human competitors by being... by kernal on Thursday July 21, 2022 @12:16PM (#32181179)
>The AI is getting precise 3D position data and physics data not only about its car but competitors cars

GT Sophy is receiving the position of its competitors (which is rhetorical statement because of course it needs to know their position), but where does it say that it's receiving the physics data of its competitors? It does mention that it does receive 'the information about the virtual physical forces affecting its vehicle', but only the forces affecting its vehicle and not the forces affecting the other vehicles (which, IMO, would be irrelevant and, quite possibly, impossible to process given the number of cars on track and the limits of its game loop processing time).

Re:Sony’s racing AI destroyed its human competitors by being... by lowbloodsugar on Thursday July 21, 2022 @02:50PM (#32183257)
Kind of a straw man though. Any real-world simulator is going to need perception and then decision systems. They are mostly two separate problems. Certainly, the decision system will need to be adapted to deal with momentarily incorrect or absent data, but for the most part will remain unchanged. Perception (telemetry really) on a closed-circuit race track is far more achievable than on city streets. So hats off to Sony for working on the decision system aspect. Would love to see it paired with perception system and put on the track.

Ultimately it has no impact on professional sport at all. I watch motor racing entirely because of the humans. It is triumph and tragedy.

But as a nerd, I love seeing "AI" beat us humans. It is supposed to.