The Deep Dive AI is not what we need right now
HungaryHippo
Member Posts: 1,144 ★★★★
As a long time players, we've built an "understanding" of the AI based on a long history of fighting game experience (hysteresis). Many of us have noticed the AI changing over time. With that perspective, the AI deep dive did not address many of the issues that plague the game currently and I have no idea why the player base finds it so appealing. So here are a few issues that are glazed over and why it doesn't take AI 2.0 to fix them:
A frame perfect button masher with random outputs ( More on this later )
This is a huge deal. I've said on many occasions that the AI appears to be frame perfect (just frame) and this means that 60 times per second (used to be 30 times a second), the AI can read your input and react accordingly. This is in no way a reasonable standard for a fighting game. The average human reaction time is 15 frames which is why it feels like the AI is so smart. I remember stating that I hope Kabam would adjust the AI for 60FPS due to concerns of the AI essentially doubling in speed... it appears they didn't. Why do you get parried by the AI? It's able to make a decision to block multiple times in the parry window. The AI is now blocking, then letting go of block, then special intercepting on reaction.
What DOES the AI know? Distance - How far apart are the Attacker & Defender
There's a large omission here. How far the A and D are apart and the available options matters greatly. Lots of us noticed the AI suddenly learning to walk up and light or parry. This wasn't a thing in the past. My head canon is that they changed two things.
1. The minimum space that forces a dash Dave used to have a name for this distance where the AI would dash but I forget.
2. The ability of the AI to choose to walk into light range
What they aren't saying here, is whether or not the parameters have been tweaked to make raw intercepts harder and to make backdrafts riskier since the AI doesn't have a problem rolling a light at that reduced distance.
What DOESN’T the AI know? Each Action the AI can take is given a weighted chance of being rolled.
This one is easy. If the weights weren't tweaked, there'd be no reason to bait intercepts. Over 60 frames, it would roll a special attack at least once. Weights are obviously tweaked in different content and baiting specials (or the lack thereof) is clearly "working as intended." Not to mention that these states or actions are not finite and final. They are able to be tweaked or added to. It's obvious that this has happened.
The biggest problem I have with AI 2.0 is that it seems to suggest that nothing can be done about the AI as it stands so wait for AI 2.0. I do not believe that to be true. Most of us don't want a stupidly easy and predictable game, however, how quickly the AI reacts has to be modifiable.
Does it need the average 15-16 frame reaction time? No but should it be able to hit you with a light at the perfect distance because they've reduced the distance that forces a dash while juicing it up to a 60 times per second AI? No. Why was the AI re-parrying and why are we being parried more often than ever (by Shocker for example)? Tweaks to parameters made by Kabam. What about recovery times?
I could go on but it is unreasonable/unacceptable to look at the AI 2.0 post as a solution to the issues plaguing the game right now. And I don't believe that AI doesn't know states (just don't).
Comments
I guess I learnt to play around AI by adapting to their playstyle
Any specific example of AI being absurd?
So my human reaction time will be able to catch up to the AI
I’ll simplify it for the people who aren’t being disingenuous. Programming is hard and we don’t expect perfection but to tell us to wait for AI 2.0 while there are major problems right now is absurd. And no not every single AI change has been inintended. That’s not how programming works unless you’re completely inept (which seems to be your defense of Kabam).
It's the reaction speed that is giving problems
My major problem with the AI is that the way it’s trending, it will be unplayable. For example, the reason we tap block when baiting specials was because if the AI was blocking, it was relatively safe to dash in. Now the AI is dropping block and special intercepting on reaction. Anyone who says this isn’t a new change is plain bad at this game. Watch any of Swedeah’s war vids examples of the new AI shenanigans.
I suppose your PE classes also give you perspective as a professional athlete?
Who actually reads this and agrees with it? I won’t even bother speaking to authority here but if this was how programming works how does Kabam even fix kit bugs? Or any other bugs for that matter? Or is the AI just a special type of program? Maybe as hard as your course… 😂
AI should be reacting to player's actions
Instead it's using the cheap dirty trick of, "looking at someone else's joystick to counter the move before it comes."
Maybe get down from your high horse to see what my point was instead of acting all condescending
2. A beginner class doesn’t give you the basis to make this claim. I’m not saying it isn’t true but neither did I suggest that fixing it was easy. You just happened to miss my entire point and reflexively defend Kabam.
That is just the tip of the iceberg too. Now imagine dealing with 10,000-100,000 lines of code that someone else wrote. If you have any experience with coding, then you know your own code is likely hard enough to read. When you read someone else's code, it is 5 billion times worse, especially if the previous person didn't keep extensive comments or documentation of what the code even does. This means that you're going to be spending a really long time documenting it yourself, or just not touching it which kabam has opted to do. If it ain't broke, don't fix it.
As far as changing the AI goes, one little change can have huge affects on how the entire fight goes. If you decide that the AI is too passive and you want to make it more aggressive, you could tweak that and then the AI with its aggressiveness is now no longer ever holding block which makes champions like hulkling really bad and makes defenders like Peni Parker a nightmare because you can't get rid of her sync-shield. This also can create other knock-on effects where venom might just corner you since his aggressiveness is him just chaining that combo with infinite reach and decides that using specials is for the weak when he can just try hitting you again due to you being in his reach.
- how the double hit champ get parried & how they fixed it ?
- does reducing def recovery time kabam can increase/decrease & does that affect ai behaviour?
Tell me why, as a CS major, that you cannot easily modify how many times the AI makes a decision a second. If you cannot come up with a solution I can suggest a few easy ways to do this.
Then tell me why it’s impossible to force the AI to make a decision other than walk forward at a specific range.
While my programming background is not in the gaming arena (mostly controls/systems), if these minor issues can’t be fixed then QC’d then there’s no way any of the systems with intricate controls would have ever worked.
At this point they are not going to toy with such a fundamental aspect (decision making), which will probably just thought them I to another loop if bugs fixes that lasts months and years.
They are already set on making AI 2.0. Why would spend another year fixing the old thing when they're just going to transition to the new one?
The only time I remember an issue that was truly game breaking was with the original Champion boss nerd update which made every AI just go passive like Arena AI.
Even then, the fix didn't really bring the AI back to the original. And it caused the additional effects which made the AI walk up to you for a parry when you whiff (it doesn't happen now)
I'm not suffering, so I will keep playing with this AI.
At the end of the day, we're not business partners or fellow coders for Kabam. We're customers, and make our own decisions based on our own preferences.
I've answered the second part of the question in the previous answer itself.