I've accepted at this point that PD just aren't giving us better AI no matter how much people keep bringing it up or harassing the Social media channels (the latter of which is not even remotely constructive or any more effective). Just seems like at this point it solely exists just to meet the bare minimum. As much as I want to say "Name one developer that has ever advertised their game with the Hardest AI difficulty setting in a trailer" (which to be fair, I don't remember seeing one myself even in anything recent), its a miracle to expect ANY real big changes.
Gonna share some experience here.
I play a lot of Assetto Corsa, and one day I felt like doing some short-track racing in it just to see how things went. Lo and behold, in the deepest, darkest corners of the Internet, I found an oval - Joesville Speedway from rFactor 2. Brilliant little track; a little bit like a shorter version of New Smyrna Speedway, and the conversion seemed mostly fine... until I raced against the AI. They were absolutely blind and never ever tried to pass. So I fired up AI Line Helper to both record a faster line (they were quite slow as well) and to set some new track boundaries (turned out that the AI thought the track was literally one car width, hence the blindness). For the line, I chose a Legends car - small, easy to set a decent line in, and the type of car you might actually find at a Saturday night short track like Joesville.
The AI worked perfectly with the Legends car. They were fast, clean, and a real challenge both to catch and defend against. Only one problem: that line that worked with the Legends? Yeah, it sucked for the late models I also had. They were far slower than they could have been because the line I set for the Legends was designed specifically for that type of car, meaning the AI was way too easy to be a challenge in anything else. And the opposite was true with the roles reversed: a line built for late models meant that the AI couldn't take full advantage of the Legends' unique abilities. That meant that the only
real alternative is to try to come up with a middle-of-the-road line where the AI isn't particularly fast in either car but also isn't particularly slow.
Now, that example covers just two cars and one track. When you also figure in hundreds of other cars
and the fact that GT is a game played by literal millions of all skill levels, you begin to understand why GT's AI is so slow - all the lines at the tracks have to work with a vast selection of cars, and the AI has to be quick enough to give a casual player a challenge without destroying them and frustrating them to the point that they stop playing.
Are there better approaches to this problem? Absolutely. FM's AI isn't particularly good, but it
is a bit better than GT's and doesn't usually fall victim to the same tedious chase-the-rabbit races as GT does. But now that I have experience creating AI lines, I can see where the problem lies when deciding what the AI should do at a particular track. And the only other method I can think of - creating a separate line with different braking and acceleration hints for every single car on every single variation of every single track - isn't an option either given how much extra data that would be, how much of a pain it would make it to update the game, and how much work it would be. So, again, there are better ways to handle the AI, but you really begin to see where PD is coming from with such slow opponents. They and other racing game devs really are between a rock and a hard place when it comes to this stuff.
