Sony's Racing AI Destroyed Its Human Competitors By Being Nice (And Fast) (technologyreview.com) 28
An anonymous reader quotes a report from MIT Technology Review: Built by Sony AI, a research lab launched by the company in 2020, Gran Turismo Sophy is a computer program trained to control racing cars inside the world of Gran Turismo, a video game known for its super-realistic simulations of real vehicles and tracks. In a series of events held behind closed doors last year, Sony put its program up against the best humans on the professional sim-racing circuit. What they discovered during those racetrack battles -- and the ones that followed -- could help shape the future of machines that work alongside humans, or join us on the roads.
Back in July 2021, [Emily Jones], who is based in Melbourne, Australia, and races for the e-sports team Trans Tasman Racing, didn't know what to expect. "I wasn't told much about it," she says now, a year later. "'Don't do any practice,' they said. 'Don't look at its lap times.' I was like, it's obviously going to be good if they're keeping it secret from me." In the end, GT Sophy beat Jones's best lap by 1.5 seconds. At a level where records are smashed in millisecond increments, 1.5 seconds is an age. But Sony soon learned that speed alone wasn't enough to make GT Sophy a winner. The program outpaced all human drivers on an empty track, setting superhuman lap times on three different virtual courses. Yet when Sony tested GT Sophy in a race against multiple human drivers, where intelligence as well as speed is needed, GT Sophy lost. The program was at times too aggressive, racking up penalties for reckless driving, and at other times too timid, giving way when it didn't need to.
Sony regrouped, retrained its AI, and set up a rematch in October. This time GT Sophy won with ease. What made the difference? It's true that Sony came back with a larger neural network, giving its program more capabilities to draw from on the fly. But ultimately, the difference came down to giving GT Sophy something that Peter Wurman, head of Sony AI America, calls "etiquette": the ability to balance its aggression and timidity, picking the most appropriate behavior for the situation at hand. This is also what makes GT Sophy relevant beyond Gran Turismo. Etiquette between drivers on a track is a specific example of the kind of dynamic, context-aware behavior that robots will be expected to have when they interact with people, says Wurman. An awareness of when to take risks and when to play it safe would be useful for AI that is better at interacting with people, whether it be on the manufacturing floor, in home robots, or in driverless cars.
Back in July 2021, [Emily Jones], who is based in Melbourne, Australia, and races for the e-sports team Trans Tasman Racing, didn't know what to expect. "I wasn't told much about it," she says now, a year later. "'Don't do any practice,' they said. 'Don't look at its lap times.' I was like, it's obviously going to be good if they're keeping it secret from me." In the end, GT Sophy beat Jones's best lap by 1.5 seconds. At a level where records are smashed in millisecond increments, 1.5 seconds is an age. But Sony soon learned that speed alone wasn't enough to make GT Sophy a winner. The program outpaced all human drivers on an empty track, setting superhuman lap times on three different virtual courses. Yet when Sony tested GT Sophy in a race against multiple human drivers, where intelligence as well as speed is needed, GT Sophy lost. The program was at times too aggressive, racking up penalties for reckless driving, and at other times too timid, giving way when it didn't need to.
Sony regrouped, retrained its AI, and set up a rematch in October. This time GT Sophy won with ease. What made the difference? It's true that Sony came back with a larger neural network, giving its program more capabilities to draw from on the fly. But ultimately, the difference came down to giving GT Sophy something that Peter Wurman, head of Sony AI America, calls "etiquette": the ability to balance its aggression and timidity, picking the most appropriate behavior for the situation at hand. This is also what makes GT Sophy relevant beyond Gran Turismo. Etiquette between drivers on a track is a specific example of the kind of dynamic, context-aware behavior that robots will be expected to have when they interact with people, says Wurman. An awareness of when to take risks and when to play it safe would be useful for AI that is better at interacting with people, whether it be on the manufacturing floor, in home robots, or in driverless cars.
still very far away (Score:2)
Re: (Score:3, Informative)
I'm not sure that there's too much human gesturing that can be read by each other on a GP racetrack - the drivers are not visible from behind, and there's not a lot of hand gesturing going on when your hands are busy steering and shifting.
There may be some signalling between drivers https://youtu.be/GLRHnFUlyss?t... [youtu.be]
but it's use for indicating intent seems to be limited.
No way (Score:3)
So you're saying bouncing off an opponent's car on the outside of a corner to take a tight turn isn't the best tactic!?
You can get.... (Score:2)
"You can get much farther with a kind word and a V8 than you can with a V8 alone." - AI Capone
A Strange Game (Score:3)
Etiquette didn't become a thing for no reason (Score:3)
I find it fascinating that, as we experiment with neural networks, we are re-discovering many of the principles that we habitually live by as humans. Etiquette has developed among humans because it serves a mutual good. Toning down aggression is a smart strategy.
Every generation, a new crop of entrepreneurs comes along that thinks they no longer have to follow the old rules. They quickly find out that short-term strategies don't lead to long-term success.
Re: Etiquette didn't become a thing for no reason (Score:1)
Re: (Score:3)
In the game it depends on the settings. You can turn off penalties, and some players prefer that because the penalty system can make mistakes. It's okay for stuff like ignoring track limits, but when it comes to assigning blame in collisions it's generally terrible.
Maybe they should develop an AI that can figure out whose fault a crash is.
Re: (Score:2)
In life, the penalties can't be turned off.
Re: (Score:2)
This is racing. 1st place is zero-sum. There's no "mutual good" unless enforced by other, arbitrary, rules.
It's only a smart strategy because the game has built-in penalties for the results of overly aggressive driving.
Turn off the penalty system and see how the AI behaves when it can be as aggressive as it likes at zero cost. I guarantee it'll be ramming the other cars off the road immediately, because why wait for an opportunity to overtake when you can just create one?
Re: (Score:3)
arbitrary
penalties for the results of overly aggressive driving
I guarantee it'll be ramming the other cars off the road immediately
Because penalizing overly aggressive driving is so arbitrary.
I guess the real world is arbitrary too, since it has physical laws that penalize overly aggressive driving, like death, fire, injuries, physical damage that impairs the vehicle in many different ways.
Re: (Score:2)
This is a game; all the rules are arbitrary. Remove the penalties and "etiquette" is no longer required. We're not "re-discovering" anything here; we're making an AI play optimally within a set of rules invented by a game dev to suit human players who don't like to be rammed off the road. Parallels to the real world do not apply here, and "etiquette" in this case is merely forcing a desired behavior by altering the rules. Remove those rules and then "etiquette" no longer applies, and toning down aggression
Re: (Score:2)
Specific rules of etiquette can be arbitrary, but the general need for rules of etiquette is not. For example, in real traffic, many places have a rule of etiquette that says that you can turn right on a red light after stopping. This is an arbitrary rule, and not all places allow this. But the important thing is that there is a rule that is understood by (most) people, and as long as they have this common understanding, they are able to share the road relatively safely.
In facing, even a racing video game,
Re: (Score:2)
I advise you to play some Forza Horizon 5 online against real human players to see what happens when there's no effective penalty system.
Spoiler: more often than not, the racers with the best "etiquette" are not the ones winning.
When driving against the AI "drivatars" they are less aggressive at lower difficulty settings which also means they're exploitable, since you can drive in front of them and they won't try to ram you out of the way. Just by staying in their path and slowing down a bit you can allow y
Re: (Score:2)
Of course, with no penalties, it makes sense that aggressive tactics would win.
Re: (Score:1)
Every generation, a new crop of entrepreneurs comes along that thinks they no longer have to follow the old rules. They quickly find out that short-term strategies don't lead to long-term success.
yet, this challeging of the stablishment is fundamental to avoid societal stagnation. Optimum short, medium and long term strategies change with time and the environment.
Re: (Score:2)
societal stagnation
What does that even mean? Who's measure of societal stagnation?
Sounds like the rambling of some teen in his basement writing a manifesto about how the world has become corrupt and needs to be cleansed.
Re: Etiquette didn't become a thing for no reason (Score:2)
yet, this challeging of the stablishment is fundamental to avoid societal stagnation
Not really. History has demonstrated that the challengers of establishments are not trusted. And following the (cosmetic) changes they bring about, they are the ones put against the wall by the "new" rulers. Witness what happened to Robespierre, and what Lenin, Stalin, Pol Pot and others did to those that empowered them.
Re: Etiquette didn't become a thing for no reason (Score:2)
Etiquette has developed among humans because it serves a mutual good.
This.
Even among animals, many challenges for dominance stop short of unrestrained physical conflict. Because the outcome is often the death of both combattants. Even the winner sometimes slinks off to die.
When the goal of the competition is to cross the finish line first and you have multiple adversaries, there are only so many you can bump if each encounter has a 50% probability of wrecking your own car.
If only (Score:2)
Great (Score:2)
Only because it confused other drivers (Score:2)
Did it beat male drivers? (Score:1)
You know, the ones with proven better hand-eye coordination skills ?
Re: (Score:1)
Or alternatively, is "Emily" actually a "woman" ?
Re: (Score:2)
Re: (Score:1)
Yes, and programs like Top Gear and "sports" like Nascar should be held accountable for road traffic deaths.
In other words.... (Score:2)
"Shake and Bake is back, Baby!"