I've been mulling over it a bit, but I can't help but to think that Sophy, at least, as we know it right now, isn't the answer to Gran Turismo's problems in the long run.
This is my first race with Sophy AI, and I got spun out by Sophy out of the hairpin at Suzuka (at around 1:08 in the video), and then forced off track in the braking zone going into Casio Triangle (2:14). Despite all these mishaps, I still finished 2nd place in a 3 lap race. I was thoroughly unimpressed by this new AI. I have heard many things about how technically impressive Sophy AI is behind the scenes, how inhumanely quick it is, and how resource intensive it is to train it up to handle different cars and tracks, but to the end user (me), all I can see is still stupidly dangerous and woefully slow AI that still struggles to overtake each other in spite of an overwhelming straight line speed advantage, and is thus functionally no different from the existing AI we have in the game.
By now, we all know that Polyphony Digital applies an artificial slowdown onto Sophy opponents to make them beatable. There is no reason my 2002 RX-7 Spirit R will vastly overpower a 1980 930 911 Turbo on the straights. From my experience
racing a real person with bone stock cars with BoP/Settings Disabled turned on, I can tell that the straight line acceleration between the FD and 930 is roughly equal until about 205km/h, when the 930 has to upshift into 4th gear. Yet, when I race a Sophy controlled 930 in that exact same car (at around 4:26 in the video), my FD blew past the 930 as though the Porsche was standing still well before I reached 200km/h. It can be seen via replay HUD that the 930 was going flat out, and it cannot even use a lean fuel mix in a race with no fuel consumption. The only explanation here is that there is an unseen, artificial slowdown applied onto the Sophy AI cars, which would mean that it is Polyphony Digital's conscious decision to structure the races in this exact "chase down the exceptionally fast lead car while navigating through slow roadblocks" fashion.
As can be seen from the various Clubman Cup+ "Chilli Races", the existing AI in the game is more than capable of putting up a thrilling battle with the player. I opine that the main problem with Gran Turismo 7's Single Player campaign isn't that the AI is bad; but rather, it's that PD keeps insisting on these asinine "chase the rabbit" style races, where the player only has one true opponent to beat in a time trial, and the rest of the cars on the field are annoying road blocks. Even with Sophy AI, if PD were to keep applying this artificial slowdown and stupid style of "racing", I hardly think that the AI model would make much of a difference to the experience of the game. That is to say, I opine that the main problem with GT7's lacklustre single player campaign isn't so much the bad AI, but moreso the way the "races" are structured, if those charades can even be called races.
The worst part about this artificial slowdown is that,
according to Peter Wurman, "Sophy can adjust by literally driving like a newer driver instead of just artificially slowing down", and yet Polyphony Digital chose the artificial slowdown route instead. Perhaps that driving like a newer driver bit still has some bugs and kinks that need working out, which also ties into my next point.
My main concern however, is that, if it takes so much time and resources to train Sophy AI to handle a given car and track combination, to the point where we can only race against Sophy AI on a select few tracks under fixed conditions with completely stock cars, then I fear that Sophy AI will become the bottleneck in overall progress for the series. I imagine that we will be stuck with our current physics model for a few good years, because to revamp the physics model would involve all the time and resources in re–training ALL existing combinations of Sophy to handle a physics change. I fear that in the next Gran Turismo title, we won't have races with variable time and weather, tyre wear and fuel use, or even tuned opponents, simply because Sophy cannot handle any of that.
I'm failing to see why we should make all these large compromises, if the end result is that Sophy AI becomes functionally identical to the regular AI.