I was a bit scared about how the AI looked in the Gamescom Mugello videos but now with all that new reviews out it looks that the new machine-learning powered drivatars are even worse.
I mean, for what we can see they are not even able to accelerate at full speed at the start of the race on straight line. In high difficulties (6/8) every reviewer agrees that is pretty easy to overtake from 10 to 15 cars in the first lap as they get stuck on piles or cannot find a line.
And then the fast guys in front open like 10 seconds gap in lap 2, and if you compare laptimes at the end of the race there are differences in laptimes like 15 or 20 seconds between first driver and 6th or 10th positions.
This is nothing like a competitive and clean AI, and obviously nothing to do with real motorsport in any category of it.
Is really that hard to make an AI that has not only the pace, but the racecraft and the possibility to make great competitive real racing?
One of the “reviewers” said that they have since fixed the A.I since giving it to the reviewers but I find that hard to believe, like this close to launch and you hand out a playable demo to a selected few with A.I that bad and then say we have suddenly fixed it… yeah nah
To me it seems the AI has serious issues with overtaking, that’s why it’s so easy to go past the “crowd” and then chase two or three guys who can drive fast because no one is hampering them. Also opponents seem to have varied skill level across the field (which is not a bad thing, adds authenticity).
Other than that, I like the fact that AI has multiple driving lines and makes mistakes every now and then. It makes racing less predictable and more interesting. I think it all goes down to tinkering with available options and finding perfect combination of difficulty and starting position (and I appreciate the fact there are so many factors to customize).
Time constraints. Someone up above had said they MUST have something out so the selected “reviewers” could do a realease that coincided with the last monthly, warts and all. For all we know they’ve been sitting on their reviews for 2 months and, after pointing out things out then changes are being made since, for example, initial feedback came from those 2 months ago.
This is why Forza always ends up being bumper car racing. You’re either forced to drive in an unrealistic manner, trying to overtake at every corner, pushing the AI aside if you want to win, or the AI does it to you, turning into your car or deliberately spinning you out even on straights. There’s nothing realistic in this, it’s bumper cars. Watching all the previewers struggle just confirmed this. “From the ground up” this is not.
I wonder how car upgrades will affect the AI as well. Like in most racing games once your car gets better parts it should and it will perform better. But then you end up destroying the AI completely.
So how will the spikes in performance and handling translate over time as the player is progressing through and unlocking more parts? How will the AI react to this? Balance itself out on the same PI as the player? Or go above? So far i don’t see any signs that suggest some significant difference compared to how most racers deal with this.
Even in gt7 as a comparison from what i see once you upgrade your car properly you tend to be able to overtake with ease corner after corner. By the time you finish lap 1 you’ve overtaken most of the AI so then you’re left fighing for the last 3 poisitions. There’s no drama. And it starts to become a bit static.
How is that massively different from the previous gen motorsport titles? I do see some genuine changes in the AI but they seem so little judging by the latest preview (even if the current issue is acknowledged and possibly fixed ). It just doesn’t sell to me the idea so far that it was " built from the ground up".
I feel free to be proven wrong once the full game comes out.
From my observations: in that initial cup opponents have PI 550-570, while player is limited to PI 600 when upgrading car, and AI don’t change PI to match yours. So it’d mean that car tuning is another way to customize difficulty. I like it
If the player’s car is really limited to be upgraded to the top of the class whitin the bulders cup then i dont know how well does that speak about the upgrade system.
I suppose what would happen is there will be parts which you’ll just be unable to put on because they simply take you out of the class? Which could limit greatly the building potential. Especially if the car is already close to the top of PI class.
If I had a dollar for every time a clearly broken thing was defended/explained away as “Oh, we’ve fixed that already” before a game released, I could afford to spend half of a year eating at five-star Michelin restaurants.
The “best” part is when the game does release and the fix was subpar at best or completely ineffective at worst, and then the developers spend the next three months showing how truly innovative they are in trying to clean the egg and rotten tomatoes off of their faces.
I think this AI is what it’s going to launch with. You can’t fix those kinds of systems in months, let alone weeks, and if they had a fix like they claimed they do, the demos would have been delayed until the new AI was coded in. At minimum, I don’t expect this or any of the other situations facing the game to even see attempts at fixing until well after the holiday season passes. At that point, though, the damage is going to be done.
They claim to fix AI every new game but never do. This time they went really in depth of what they fixed but not so much how they fixed it. They always said Drivatars were based on AI adaptive learning. Effectively what they’re doing here. But it was never good, And seems to have gotten worse over time, not better. This time, they’re going in harder on the generative AI schtick, But from the sounds of the hands on, Its the same as its always been. You’re still going to have to lunge up 10 places in the first corner to actually compete with pole. And then pole AI are going to run away with world record pace or faster and gap the entire field by quarter miles like they’ve always done. I’ve seen 1st and 2nd ai in Horizon 5 lap the pack before. And the rest of the pack get the same laptimes and brake way too early.
It vaguely sounds the same here. But time will tell, the real issues with the AI tend to bare out when taking them out of the games tailored scenarios, and thats the only thing reviewers are getting to look at right now. So endurance racing, multiclass, mixed weather. If they’re having similar issues they’ve always had under ideal conditions set by the demo, Imagine what they’re going to look like when you actually put them under extended observation.
The worst part about the Forza AI is that they never compete against themselves.
All the AI cars are going to basically finish in the grid order they start in. No matter how long the race lasts. Barring player intervention its almost a guarantee. 1st and 2nd are the only AI that are actually running competitive times that will actually give you trouble, The rest of the grid are little more than rolling obstacles. If this game hasn’t solved this, They’ve completely lied about the new game. Not even misrepresented, just an outright lie.
It’s all so familiar as well. They promised the AI in Forza Motorsport 5 would be ‘powered by the cloud’ and would be innovative and realistic and so on. They always make bold claims and use buzz words but it never amounts to anything. It looks like they tried to do it again for this game but people can see through it this time. If it releases with AI anything like as bad as we see in these recent videos, the game is going to be DOA.
Well, opponents at front potentially drive just as quick and clean as GT Sophy
As for other factors, we still don’t know how Sophy handles driving in the crowd and with speed lower than maximum (i.e. at lower difficulty levels). And after my experiments with it I’m not convinced that Sophy is more aware of player’s car in their path than the new AI in Forza. In FM I saw opponents leaving place for player and trying to avoid contact so it still could turn out better than what Polyphony Digital is cooking
I had a very good experience with Sophy, but it does need more time in the oven with regards to tuning for different speeds. In their monthly event, PD faked difficulty by using cars with different PPs, not adjusting the AI’s level. When it comes to driving in crowds, the quickest one (Sophy Violette) was able to move up from last to first at a decent pace.
The issue I see with T10’s AI is the implementation of it (same old rabbit chase) and also the praise it gets for things like the AI making random mistakes, which they used to do in FM5 and 6. Mixing dumb and smart AI to simulate different skill levels may not be a good way to showcase the feature either, as dumb AI may be “smartly” playing dumb!