This thread is going to discuss the future for AI. In general, I'd say I have two issues with AI. First, is how they conduct basic on track driving, how the AI cars appear to behave on track with respect to the physics. Second is how they perform whilst in competition with either other AI cars or human opponents. When I review an AI car driving around the track in front of me, I see several concerning issues in rF2. First is the apparent application of braking and acceleration. For the former, you often see the cars lock up wheels and leave black tire marks on the road, even at relatively low 'strength' levels. The brake application seems very abrupt and digital, and not at all smooth. They don't seem to have a great concept of things like threshold braking and trail braking. For the latter, throttle application is also very digital, on-off, and not smooth. You can hear this in the engine notes. I've never seen a human be quite so reactive/jumpy in inputs, and review of telemetry shows that people do spike and have quite large fluctuations on inputs quite rapidly, but not to the same extent as the AI cars do. Possibly it's down to the physics 'tick' rate and not having sufficient chances to modulate inputs. But the net result is the car doesn't seem to be operated in the same fashion as a human would. Further, the behavior of the car under AI control is also different to a human-driven car. My understanding is they run a simplified physics model. And at times, this clearly shows especially as the car say rides curbs or comes off curbs, where there are very noticeable jumps. The cars also seem to have almost infinite grip, or at least, very different grip characteristics to the non-AI cars. Whilst they can make mistakes, often hilariously overshooting corners/apexes if the AI "line" isn't well constructed, they don't seem to make the same mistakes as humans make, and often we end up in "lazy spins" into and out of corners as we try to regain control and fail, where the AI most usually makes immediate corrections and continues on. Finally there is competition strength, and under what circumstances they try to take a different line, overtake, or defend an overtake. Far too often the cars seem to run in a "train" and not try alternate lines/looks, or run into the back of you as they intend to hit their pre-programmed brake markers whilst sticking to that ideal line. They do a good job generally of avoiding side-to-side contact and sticking to a line once selected within a corner. We all appreciate how a human will "fake" a move one way with full intent to make the move a different way: the "over-under" and similar as opposed to just making a pass due to superior traction into or out of a corner. I have to figure part of these limitations are based on the fact that the AI physics are in the same physics thread as the calculations done for the human player. In other words, that one processor thread has to do all the work. Not sure if programming wise the AI 'logic' can be separated out into separate threads, but it seems a bit silly that many of us have machines with 4/8/16 threads possible and only 1 or 2 being used, CPU being completely under-utilized in total as a result. Discuss!