Gran Turismo Sophy: Sony AI x Polyphony Digital

  • Thread starter Magog
  • 1,675 comments
  • 175,712 views
But given how PD has bungled executions in the past, I'm not holding my breath...
So negative :eek:

Would you say that's a fair assessment or would you say it's annoying being labled without proper context? Interesting to think about.
 
Last edited:
So negative :eek:

Would you say that's a fair assessment or would you say it's annoying being labled without proper context? Interesting to think about.
You take a person's words out of context and label them...then you ask if this is a fair assessment?

Hmm, let me get back to you because I need more time to contemplate this elusive question...but don't hold your breath.
 
You take a person's words out of context and label them...then you ask if this is a fair assessment?

Hmm, let me get back to you because I need more time to contemplate this elusive question...but don't hold your breath.
Yes, the point clearly went over your head.
 
I don't think Sophy is intended to replace human drivers or competition. It's a training tool.

I would compare it to one of those tennis machines that launches balls so you can practice different types of shots. If you love playing tennis, you'll enjoy it and it will quickly improve your game. Of course, it will always be more fun to play a real game against a real person, but that's not the point of the machine.
If Sophy only becomes a training tool like a tennis ball machine as you say, then I for one will not be keen to engage with it. But, if there are personalization options, along with different events and challenges within a deeper and longer single player campaign, then I would be enthusiastic.

Good luck PD, this could be a beautiful addition that separates you from the pack, or another underwhelming hyped feature that drives fans away.
 
I'm kind of surprised at how negative people are. At worst, it will be better AI to race against...or customizable AI that is just a tad faster than you at the moment so you have a 'live ghost' to race against...or even, because it's a reinforced learning algo, it may have the ability to watch your laps and adopt your specific faults/habits so that you can race against a full grid of drivers exactly like you. Or if you prefer, customize your AI to have certain traits that you want to learn -- like trailbraking.

Why all the sour faces?
I personally don't see it as negativity. It's completely justified skepticism. We still don't know how Sophy will be implemented in the game, and Polyphony's track record doesn't exactly inspire confidence.
 
I personally don't see it as negativity. It's completely justified skepticism. We still don't know how Sophy will be implemented in the game, and Polyphony's track record doesn't exactly inspire confidence.
You're right about Sophy not being implemented yet and PD's track record.

But having just read the latest article from Jordan who seems to be impressed with it, is a credible signal that points to something that many of us may enjoy. I'm not saying Jordan's word is final, but he seems like someone who takes his journalistic profession seriously and has given an interesting -- and perhaps a very innovative -- part of GT a fair assessment.

Might be more interesting to engage in this development with curiosity and a sober level of optimism rather than just negativity/skepticism. I personally do not value comments from people who offer little else than a constant no, never, not gonna happen.
 
You're right about Sophy not being implemented yet and PD's track record.

But having just read the latest article from Jordan who seems to be impressed with it, is a credible signal that points to something that many of us may enjoy. I'm not saying Jordan's word is final, but he seems like someone who takes his journalistic profession seriously and has given an interesting -- and perhaps a very innovative -- part of GT a fair assessment.

Might be more interesting to engage in this development with curiosity and a sober level of optimism rather than just negativity/skepticism. I personally do not value comments from people who offer little else than a constant no, never, not gonna happen.
Other than insulting people you deem to be negative and fighting peoples right to opinions you dislike, you've really not contributed very much in this thread let alone anything of value.

@Tyrone B is absolutely spot on, we do not know enough about the implementation or final form of Sophy to blindly believe it will be amazing. We do know that Kaz has declined to suggest it will replace the games AI in general and be a competitor to race against and instead will be more of a training tool which the article implies will more likely be the case too. Combine all of this with PD's habit of over promising and under delivering on features and there is plenty of room for skepticism. More information may fuel that further or lessen it, but time will tell which it will be.

You may dislike all forms of negativity, but scrutiny is an important part of life and it's not going anywhere, in fact it's often neccessary for the betterment of a product or service.
 
Last edited:
I just think it's more FUN to think about the possibilities than to constantly dig the same grave of no, not now, never. Isn't it better to try and help PD get their act together than to just keep kicking them in their little nuts?

Maybe you and I need to have a beer, Dave?
 
I’m pretty sure in an earlier article someone from PD or Sony AI confirmed that Sophys difficulty level will be scalable
They've said that. Getting Sophy into GT7 seems like a similarly ambitious project to the GT6 course maker, and given how the goalposts shifted on that project I wouldn't take what saying about Sophy features right now as anything other than optimism.
I want to know if it's possible to use the same machine learning algorithm to create a Sophy AI race steward that can judge contact penalties correctly. Because the current Sophy already learns some degree of sportsmanship, it wouldn't be a stretch to basically reprogram it to judge racing incidents and then decide if it's "wrong" or "right", and give correct amount of penalty to the right driver.

Making a perfect penalty system is basically impossible but with machine learning super AI, I think we can do much better than the current system. It would be beneficial for players of all skill levels too, instead of just challenging for the super fast drivers. Also it would help prevent such a controversial ending as the final race in the Nations Cup.
It's harder than it sounds. Sophy has a goal, go fast, and some negative consequences for doing so in a manner than results in contact, off-tracks, and such. With enough training and appropriate weightings, this results in an agent that is quick and naturally mimics some of the behaviours of a clean driver.

With an AI steward, what is the goal that you're trying to optimise towards to determine fault? How are you training these agents, and how do you tell if they're getting better or worse?

Racing incidents are something that highly educated and experienced humans can disagree on. Good luck trying to explain them to a computer in a clear enough manner that it's able to take over and be consistent.
I just think it's more FUN to think about the possibilities than to constantly dig the same grave of no, not now, never.
So you want to fantasise? Sure, but you're then gonna rag on people for being realistic with regards to this realistic racing game? Odd choice.
Isn't it better to try and help PD get their act together than to just keep kicking them in their little nuts?
None of us are helping PD, because they don't pay attention to the community unless it's about to tank their profits. We can all say what we like, and the chances of Polyphony or Sony AI paying any attention are microscopic. So it really doesn't matter whether you're kicking or licking their little nuts. "Better" doesn't come into it at all.
 
So you want to fantasise? Sure, but you're then gonna rag on people for being realistic with regards to this realistic racing game? Odd choice.
See this finally for what it is, while I understand his contempt (because I've said so many time about how PD has underdelivered), these kinds of comments, from these kinds of individuals that are constantly negative amount to one thing and one thing only...it just doesn't have any value...because for one, it just turns everything into a fight and kills any kind of dialogue that may very well lead to improvements we may all enjoy.

None of us are helping PD, because they don't pay attention to the community unless it's about to tank their profits. We can all say what we like, and the chances of Polyphony or Sony AI paying any attention are microscopic. So it really doesn't matter whether you're kicking or licking their little nuts. "Better" doesn't come into it at all.
I rest my case.
 
See this finally for what it is, while I understand his contempt (because I've said so many time about how PD has underdelivered), these kinds of comments, from these kinds of individuals that are constantly negative amount to one thing and one thing only...it just doesn't have any value...because for one, it just turns everything into a fight and kills any kind of dialogue that may very well lead to improvements we may all enjoy.
Says the guy who gets bitter every time someone doesn't agree with him. If you want dialogue, you've gotta start by not writing off anyone with a different opinion as worthless. Because as it stands it seems like what you want is an echo chamber, not dialogue.
 
If you want dialogue, you've gotta start by not writing off anyone with a different opinion as worthless.
When have I done this? Your specific accusation must be called as an example of some who just want to fight incessantly. Think about what you are saying, as it pertains to the civility you are also citing from others.

Edit -- on a good day, without sounding patronizing, you have many some good points on various shortcoming about GT...but, I must profess, perhaps you are drinking tooooo much from your Coolaid?
 
Last edited:
What's there really to say at this point besides "I hope Sophy doesn't suck like all the other half-baked lofty ideas that PD have hyped up over the years"?
I feel like it has a bit more promise behind it because we've actually seen it in action many times.
 
What's there really to say at this point besides "I hope Sophy doesn't suck like all the other half-baked lofty ideas that PD have hyped up over the years"?
Yea, this is a solid view after all we have gone through together. Thanks for keeping your comment civil...I hope PD listens to all of their devoted fans, in all our shades, colors, demographics, enthusiasm, etc.

But I'm not holding my breath...call me impartial.
 
I feel like it has a bit more promise behind it because we've actually seen it in action many times.
We've seen it taking absurd lines abusing the physics and track limits to set alien times and beating top GT players in 1v1s.

What does that promise for the average player?
 
When have I done this?
Uh, in the post I quoted?
...these kinds of comments, from these kinds of individuals that are constantly negative amount to one thing and one thing only...it just doesn't have any value...
"Doesn't have any value" is another way of saying worthless. You were pretty explicit and it's right there.
Your specific accusation must be called as an example of some who just want to fight incessantly.
Is it? Is that what it's called when I say "maybe don't write off other people's opinions as worthless" after you've said "these comments from these people have no value"?

Take a breath, dude.
...but, I must profess, perhaps you are drinking tooooo much from your Coolaid?
The Koolaid that implementing Sophy on a PS5 is a non-trivial problem? That alien level AIs aren't innately fun to race?

If you look at my posts in this thread I'm fairly positive about Sophy as a technology, it's not as groundbreaking as some people would like to make out but it's a very clever and highly useful implementation of machine learning. I just think that getting it running on the same hardware as GT7 is a relatively difficult proposition given the information we know about it so far. If they do get past that hurdle, there's then the question of what gameplay can be made with this new AI. Having very, very fast AI isn't fun all by itself unless you're an alien, you have to do something with it that users are going to enjoy.

Polyphony have their problems as a studio, but you don't even need to get into that to see that there's some stuff around putting a Sophy-type AI in a racing game that needs a fair bit of work. But saying that would apparently be negative, so feel free to go off.
I feel like it has a bit more promise behind it because we've actually seen it in action many times.
Sort of. It's promising because we've seen it actually function, but it's also not that promising because it's been a long way from running on the same system as the game itself.

With something like the course maker we never saw it working before release, so the question was whether these things they were promising were possible at all. With Sophy we know that it can be done, but it's less clear whether it will work on a PS5 that's also running the game.

Given that Jordan's instance was just running through a separate Playstation, there's also always the option that they just run Sophy in the cloud and treat it like an online player so that there's no extra load on the PS5. But one would assume that they would only do that if they couldn't get it running locally, which probably means that the hardware demands (and therefore costs) would be reasonably high. I doubt Polyphony would want to be paying extra money for servers for a feature that doesn't bring them in any extra money. This assumes also that Sophy is a free update, if they charge for it then they could well afford server hardware but I think that it would probably be pretty unpopular with the users.

If it doesn't end up working or if it comes in with limited features or capability, then I suspect it's going to be largely because of hardware and not because Sophy is fundamentally incapable. The technology is fairly impressive and adaptable, but the PS5 is finite in what it can do.
We've seen it taking absurd lines abusing the physics and track limits to set alien times and beating top GT players in 1v1s.

What does that promise for the average player?
To be fair, abusing physics and track limits is what top human players also try and do. That's not inherently bad, and if you want a human-like opponent it's probably a good start.

At worst it's an interesting new(-ish) approach to AI. Traditional methods are usually slow at first and get faster as the AI code gets more refined. Sophy seems like it might have the opposite problem - it's entirely too fast at first and they're going to have to figure out how to slow it down while keeping the racecraft and awareness.

I do agree though that I'm struggling to see what they're going to do with this gameplay-wise. Other games have had significantly better AI than Gran Turismo for years, and while it makes the AI more pleasant to race against it's not a magic bullet. Race formats and grids still need to be well designed to be fun. And if they can't figure out how to slow Sophy down, there are few formats that I can think of where drivers with major speed differences match well.

Endurance racing, perhaps? Sophy could be a co-driver or something, but if they're heaps faster than you then the game is to get them as much seat time as possible so it becomes basically B-Spec. My fear is that they'll just make you the rabbit and you have to get to the finish before Sophy catches you. Which could be cute a couple of times, but that's probably about it.

1670382719727.png
 
I wonder if training this AI with tire wear, tire temperatures and dirty tires on would help with the physics exploitations. I don't know pretty much anything about machine learning but setting up some kind of condition where the least amount of tire wear or general grip loss is the most desirable should be possible right? That would probably iron out the preemptive sliding into corners it does.

Don't know what they would do to penalize curb abuse though.
 
Uh, in the post I quoted?

"Doesn't have any value" is another way of saying worthless. You were pretty explicit and it's right there.

Is it? Is that what it's called when I say "maybe don't write off other people's opinions as worthless" after you've said "these comments from these people have no value"?

Take a breath, dude.

The Koolaid that implementing Sophy on a PS5 is a non-trivial problem? That alien level AIs aren't innately fun to race?

If you look at my posts in this thread I'm fairly positive about Sophy as a technology, it's not as groundbreaking as some people would like to make out but it's a very clever and highly useful implementation of machine learning. I just think that getting it running on the same hardware as GT7 is a relatively difficult proposition given the information we know about it so far. If they do get past that hurdle, there's then the question of what gameplay can be made with this new AI. Having very, very fast AI isn't fun all by itself unless you're an alien, you have to do something with it that users are going to enjoy.

Polyphony have their problems as a studio, but you don't even need to get into that to see that there's some stuff around putting a Sophy-type AI in a racing game that needs a fair bit of work. But saying that would apparently be negative, so feel free to go off.

Sort of. It's promising because we've seen it actually function, but it's also not that promising because it's been a long way from running on the same system as the game itself.

With something like the course maker we never saw it working before release, so the question was whether these things they were promising were possible at all. With Sophy we know that it can be done, but it's less clear whether it will work on a PS5 that's also running the game.

Given that Jordan's instance was just running through a separate Playstation, there's also always the option that they just run Sophy in the cloud and treat it like an online player so that there's no extra load on the PS5. But one would assume that they would only do that if they couldn't get it running locally, which probably means that the hardware demands (and therefore costs) would be reasonably high. I doubt Polyphony would want to be paying extra money for servers for a feature that doesn't bring them in any extra money. This assumes also that Sophy is a free update, if they charge for it then they could well afford server hardware but I think that it would probably be pretty unpopular with the users.

If it doesn't end up working or if it comes in with limited features or capability, then I suspect it's going to be largely because of hardware and not because Sophy is fundamentally incapable. The technology is fairly impressive and adaptable, but the PS5 is finite in what it can do.

To be fair, abusing physics and track limits is what top human players also try and do. That's not inherently bad, and if you want a human-like opponent it's probably a good start.

At worst it's an interesting new(-ish) approach to AI. Traditional methods are usually slow at first and get faster as the AI code gets more refined. Sophy seems like it might have the opposite problem - it's entirely too fast at first and they're going to have to figure out how to slow it down while keeping the racecraft and awareness.

I do agree though that I'm struggling to see what they're going to do with this gameplay-wise. Other games have had significantly better AI than Gran Turismo for years, and while it makes the AI more pleasant to race against it's not a magic bullet. Race formats and grids still need to be well designed to be fun. And if they can't figure out how to slow Sophy down, there are few formats that I can think of where drivers with major speed differences match well.

Endurance racing, perhaps? Sophy could be a co-driver or something, but if they're heaps faster than you then the game is to get them as much seat time as possible so it becomes basically B-Spec. My fear is that they'll just make you the rabbit and you have to get to the finish before Sophy catches you. Which could be cute a couple of times, but that's probably about it.

View attachment 1213980
That last sentence, we have that in GT Sport. Especially, the dang Porsche Cup races. Watch ya back! ;)
 
I just think it's more FUN to think about the possibilities than to constantly dig the same grave of no, not now, never. Isn't it better to try and help PD get their act together than to just keep kicking them in their little nuts?

Maybe you and I need to have a beer, Dave?
I feel you on that, but that will just set you up for inevitable disappointment. I learned that lesson from GT5. Also, constructive criticism is the best way to help PD get their act together and there's plenty of it here. The problem is PD doesn't seem interested in actually addressing the criticism and improving the game.
 
Last edited:
I feel you on that, but that will just set you up for inevitable disappointment. I learned that lesson from GT5. Also, constructive criticism is the best way to help PD get their act together and there's plenty of it here. The problem is PD doesn't seem interested in actually addressing the criticism and improving the game.
It's understandable that someone might have a different perspective on Polyphony if they didn't play through the PS3 era. People who started playing with GTS probably have a fairly positive impression of them as developers. There were issues, but the quality and level of support was high if you were into the type of game that it was.

I think it's easy for those of us who have been around for a while to forget that GT6 was 9 years ago, GT5 was 12 years, and GT4 was 17 years. There are adults playing GT7 now that were too busy filling their diapers to play when GT4 came out. While it hasn't been that many releases in real terms, the slow pace at while Polyphony puts games out means that there's always going to be a decent amount of people who have heard of Gran Turismo but haven't been around for a game release.

Also, damn kids, skateboards, rock music, yadda yadda. :cheers:
 
If you look at my posts in this thread I'm fairly positive about Sophy as a technology, it's not as groundbreaking as some people would like to make out
Well, Sony AI are calling it a breakthrough, and an article in Nature about it is not just empty words on an internet forum. Besides, for their part, it's not just research into "racing game AI" – hop on to https://ai.sony and take a good read.

And, even if there are games with great "AI", very few of them pose a challenge to top-level drivers the way that Sophy does. Remember, many "AI" in games cheat in various ways to accomodate the human driver's level of expertise. Sophy doesn't need to cheat as such in the game.
Having very, very fast AI isn't fun all by itself unless you're an alien, you have to do something with it that users are going to enjoy.
Come on, are you seriously implying that PD wouldn't have thought about turning it down from 11 to accomodate drivers starting out in Gran Turismo?

As it has been shown publicly thus far, Sophy is a proof-of-concept, to prove that it actually can take on, and in many cases beat, the top level drivers in Gran Turismo. This by no means implies that the same level of "awareness" will be applied to all human drivers, because, as you correctly state; it wouldn't be much fun if the AI zoomed off into the distance never to be seen again.

---

As an aside, having dabbled a bit with machine learning personally, there are various ways to accomodate varying levels of proficiency. One way, as has been mentioned, is for how long a model learns (100 laps vs 1000 laps). Another way is to limit the number of features available to the model, or how far into the future it can "look" (I think they are looking 6 seconds ahead at the track).

This also makes it possible to create ML models that are smaller and faster (but less intelligent and more prone to mistakes) for races with more cars, and models that are more intelligent and takes more features into account for higher-level drivers in a one-on-one challenge.
 
I agree that you can’t beat racing real people, but most public lobbies in any racing game are pretty chaotic. This ai looks faster and cleaner than most players if you just want to hop on and race.

That is very true. But these chaotic lobbies force us to find decent people who can race. And then we add them to our friends list, they push us to even better laptimes and we respect then for racing clean and fast. Its called "Friends".

The ai is always able to beat my best times. Always. Any day, any time of day. It doesnt laugh, it doesnt curse. It never comes to Hockenheimring to meet after years of online racing.

The ai might be a nice addition, but theres way too much work, attention and importance given to it. Its a headline and click count thing.

Theres a robo race series, where ai controlled EVs race each other. No ones watching. The human factor, human interaction, thats what its all about. Imperfections make character.
 
Theres a robo race series, where ai controlled EVs race each other. No ones watching. The human factor, human interaction, thats what its all about. Imperfections make character.
You definitely have a point, but the huge difference between the two (as an AI comparison) is that Sophy can afford to crash thousands of times to find a model that works. Roborace et al needs to keep their equipment as safe as possible for budget reasons if nothing else, coupled with the difference between a digital version of the track ahead compared to the computer vision version available to a real life AI controlled vehicle.

Edit: Want to make clear that I am not criticizing your point, just wanted to compare the two approaches. AI as a concept, at least at this time, is pretty unpersonal in its nature. Though if big brand names attached themselves to something like Roborace, especially in light of self-driving cars being on the horizon, the competition between brands (brand X avoids collisions better than brand Y) may well create another sort of enjoyable entertainment. Too soon to say.
 
Last edited:
I just think it's more FUN to think about the possibilities than to constantly dig the same grave of no, not now, never. Isn't it better to try and help PD get their act together than to just keep kicking them in their little nuts?

Maybe you and I need to have a beer, Dave?
That fine, but it doesn't make it wrong to be a realist either. As for helping PD, the most PD have acted in the interest of the consumers regarding GT7 was in response to the review bombing that took place several months ago when they cut race payouts. So being all positive and la-de-da about the game isn't necessarily going to help them or anyone playing the game.

In fact, I want people to critique my ideas in work because my ideas impact a lot of people and once I give the green light they are final, if everyone just tells me how wonderful they are yet there is a glaring flaw that really should be remedied that can cause big problems, I want people to tell me that. If my idea makes someones work harder, I want them to tell me that, if it adds unneccessary steps to a process, I want people to tell me that. The worst feedback is blind feedback be it positive or negative.

I do at times think how great a game GT7 would be if they did this, that or the other, and you can find those posts by me in other threads where I discuss how great a base GT7 is to build a game that's really special. But as a realist, I don't expect many (if any) of those changes, and at the same time, they would make the game perfect for me, but there's plenty of other people with different tastes who might want something different again.

So keep kicking PD in the nuts until they do something about improving the game they've had the pleasure of charging us all to buy, it's your right and mine as a consumer and products that receive backlash either improve or die. GT7 isn't on the cusp of writing off the GT franchise, but receptipon isn't what it was, there are only so many missteps that they can afford. My positivity leads me to hope that they sort it out with a Spec II update or something, but the realist says they might sort it out for GT8 but we'll have to wait and see.
 
Last edited:
Right! But I dont get how that matters to me as a player? How does this affect my playing experience?
Different people, different wants and needs. I have personally taken the journey from playing GT purely offline (all the way since GT1) to more or less only playing online with friends. Nevertheless, there is a "threshold" so to speak, which can sometimes be difficult for some to overcome; "Am I good enough to play online against real people?"

Having the option to play against an AI that can challenge you at your own level may well be the decider that allows someone to take the step and play online, having learned along the way what it means to race cleanly against an AI that properly defends instead of "on-the-rail-kind-of-AI" we have now. Too brutal when passing (or defending for that matter) and you fail. Doing things like this in public lobbies or daily races sadly often results in insults and negativity (in some respect @dabz343 have a point too, with respect to this thread).

Personally, for me the game's "AI" (I prefer not to use that word really) is too much of just being "in the way" rather than any kind of challenge. Only challenge they pose is when the game places you last and P1 drives like a bat out of hell until you catch up.

One (playing with friends) does not preclude the other (playing against the AI).
I do at times think how great a game GT7 would be if they did this, that or the other, and you can find those posts by me in other threads where I discuss how great a base GT7 is to build a game that's really special. But as a realist, I don't expect many (if any) of those changes, and at the same time, they would make the game perfect for me, but there's plenty of other people with different tastes who might want something different again.
Hear, hear... There have been many a time that one would have wanted to have a day with Kaz (and translator-san, I guess, my japanese is not up to scratch) to discuss the possibilities of what could be done.
 
Back