You mentioned that AI 2.0 aims for predictability rather than adjusting reaction times. While predictability is valuable, it’s crucial to recognize that reaction time directly impacts fairness. If the AI can react within 16.7ms (frame-perfect at 60fps), and human players average 150–250ms, this disparity can lead to situations where: • Players are punished for correct play, such as being blocked immediately after trying to punish a successful special dex. • The AI can counter mid-animation, making certain strategies ineffective (eg relics somehow being blocked on hit 2/3.)
This isn’t about making the AI easier but ensuring that skillful play is rewarded, not negated by inhuman reaction speeds.
You keep making this statement without foundation or justification, and at this point given that a) I've shown evidence that this is not a problem in other games, b) described how it wouldn't create a problem in any reasonable implementation in this game, and c) stated that this is also irrelevant, because no matter what hypothetical problems it could cause, the whole point of AI 2.0 is to give the devs the tools to eliminate such problems, I believe you should be required to actually address why this should be a concern. You keep saying it is one, despite overwhelming evidence to the contrary. In the absence of something other than "this is obvious" I believe this begins to fall into conspiracy theory territory.
To be absolutely clear:
1. Every fighting game processes information faster than humans do. Every single fighting game.
2. Every fighting game could out-react the player if it chose to do so. Every single fighting game.
3. Every fighting game doesn't out-react the player because it chooses not to. This is an explicit choice built into the AI system. It can, but it is programmed not to, most of the time.
Every. Every. Every. Prove me wrong, find a counter example. I would love to see it. Maybe there is one out there somewhere with some sort of weirdo sophisticated reaction time simulation, but I doubt it. At least one of the very games you cite is most definitely not doing that because in spite of the fact you think it doesn't it demonstrably does have instant reactions built into the AI at highest difficulties.
At this point, this isn't about the AI. This is about you being stuck on an idea, and for some reason willing to die on a hill you are simply insufficiently informed about. At this point I believe there exists no work or reference that will convince you otherwise. I think if the developers of every fighting game you ever played said you're mistaken, you'd simply repeat what you said directly to them as if they did not understand what you were talking about. Including the developers of this one. Which I suppose is a cautionary warning unto itself.
I won’t pretend to know anything about the technical programming aspect, nor about other fighting games.
My take - was the AI being a “frame perfect button masher” a problem before you knew about it?
It existed the entire lifespan of the game; there weren’t any real persistent concern that the AI was responding faster than humans.
If it explains all the problems we are seeing, then sure you have a valid concern. But if its not the root cause of the problems (which Kabam isn’t specifically blaming but is saying we need more sophisticated tools to fix problems when they come up), then I really don’t understand your concern even if it is still present in 2.0.
I won’t pretend to know anything about the technical programming aspect, nor about other fighting games.
My take - was the AI being a “frame perfect button masher” a problem before you knew about it?
It existed the entire lifespan of the game; there weren’t any real persistent concern that the AI was responding faster than humans.
If it explains all the problems we are seeing, then sure you have a valid concern. But if its not the root cause of the problems (which Kabam isn’t specifically blaming but is saying we need more sophisticated tools to fix problems when they come up), then I really don’t understand your concern even if it is still present in 2.0.
To address this point in general, I believe the reason why Kabam even mentioned this at all is not because it is a problem unto itself. The problem is that because the AI is a "frame perfect button masher" it *can* sometimes behave in not just superhuman ways, but ways we didn't even think were possible, because the game wasn't designed to allow such behaviors, but the AI doesn't just do what it was designed to do, it can theoretically do everything possible under the right circumstances.
It is as if the AI is "trying" to figure out ways to do the unexpected, because it is constantly tapping the screen and constantly trying all kinds of things at every possible instant of time. Because of that, if a maneuver should not be possible because it wasn't intended, but because of a quirk of the game there is a single animation frame where it becomes possible, the AI will find it and do it.
Someone might think that the problem, then, is that the AI is too fast - it is capable of finding this momentary windows of opportunity. But speed is irrelevant. If the AI were slower, it would still find these windows of opportunity, it would just find them less often. If there is some momentary weirdo window of time where the AI could do something bizarre, and it tries to do it 60 times per second it will definitely find that moment and do that thing. But if you slow the AI down to only taking actions ten times per second, or once per second, or once per week, it will still find that exact moment in time and do that weird thing, it will just happen much less often.
And that is actually what happens. Although the AI is super fast and is constantly making decisions sixty times per second, it doesn't try *everything* every animation frame. The AI dev diary makes that clear. The AI only tries certain actions at certain times, which means if there's a way for the AI to pull a fast one on the player, it might only happen when that exact window of opportunity arises *and* the exact action necessary to pull it off happens to be in the AI's current random option list *and* it happens to randomly select that action. So when a player says "hey, the AI tea bagged me in the middle of a combo" and the devs try to reproduce that problem to try to fix it, it doesn't happen to them. It doesn't happen to lots of other players either. And even when they finally do reproduce the issue, they might find there's no way to make it go away, because if they remove the action from the set of possible actions it will not only remove that issue, it will also prevent the AI from doing that in other contexts. It could even make the AI worse, because when you remove an option, that will force the AI to start doing other things more often, and those other things might be even worse in other situations. You could unmask a new issue somewhere else while trying to fix the issue in front of you.
Getting back to speed. Suppose instead of doing things the smart way, we do it the weirdo way. Instead of just making the AI better at producing the behavior we want, we just introduce a delay into the AI. We say whenever you see XYZ and you decide to take an action, delay that action by 100ms to replicate human reaction times. This would be horrible for a number of reasons. To illustrate, imagine the devs want the AI to hold block when it thinks the player is going to attack. The delay would delay going into the block state by approximately the human reaction time. This could randomly turn the intended block into a coincidental well-timed block. Which would be bad. What's more, because this reaction time is hard coded into the AI, there's no way to fix it. That's basically the problem(s) we have with AI 1.0 now.
Make the delay variable so we can tune it to anything from 0 to 250ms? Well, on top of that being more complex and more difficult to both program and troubleshoot this only works if the AI's is fast enough to allow this in the first place. If you're going to allow a programmable delay, the AI must be as fast as the minimum possible delay. Which is going to be what, exactly? 0ms. Because only an insane person would hard code a minimum delay into the AI and on top of that add a programmable delay. If you're going to put in a programmable delay, which is no different from implementing action scripts that fire instantly but include a delay in them, the intrinsic delay in the AI should be zero, to give the designers maximum flexibility in tweaking that delay.
Let me clarify the concern again — not because I’m “stuck on a hill,” but because I think we’re mixing up two separate issues:
1. AI behavior vs. AI reaction time.
Yes — MCOC’s AI behavior needs fixing. Yes — AI 2.0 should give the devs better tools for predictability, tuning, and smarter moves.
But none of that fixes the underlying problem if the AI still reacts at 60fps — 16.7ms per decision window — while human players operate at ~200ms.
When the AI has zero input delay, and players have animation recovery + input lag + human reflex limits, we end up in situations like:
• Dashing in after a dexing special and getting blocked or punished instantly. • Mid-combo actions (MLLLH / medium combo > special) being interrupted, blocked or punished, so having to resort to MLLL because the animation is shorter. • Dash-ins/medium intercepts being unsafe now and being punished in real time even when timed/spaced correctly. • Relics being somehow blocked on the 2nd or 3rd hit when the first hit landed.
That’s not about “how smart the AI is” — it’s about mechanical limits. If you fix behavior but keep the AI’s reaction speed inhuman, you still punish correct player decisions.That’s the issue.
2. Yes, other fighting games absolutely limit AI reaction speed — by design.
You asked for receipts. Here are just a few examples: • Super Smash Bros. Melee: AI researchers working with the Slippi team simulated human-like agents by intentionally capping AI reaction time to ~200ms to keep fairness. Source: https://arxiv.org/abs/1810.07286 • Skullgirls: AI difficulty levels include built-in reaction delays to simulate human reflexes — the devs even made those delay values adjustable during development. • Street Fighter IV & V: Developer notes (via GDC) confirm that AI response patterns are built with delayed inputs at lower difficulties, and reaction capping is standard even at high levels to preserve fairness. • Mortal Kombat X & 11: Same deal — AI difficulty sliders include changes to how fast the AI can block/react, not just how much it combos. (This is directly visible in practice when testing frame traps.)
As further evidence, game developers themselves widely acknowledge this principle. On Hacker News, a respected dev forum, one commenter (alexflint) explains:
“One of the easiest tricks that game designers use to make AI feel more human is to simulate human reaction time… the AI is coded to wait 200ms before reacting.” This reflects exactly what we’re suggesting — not dumbing down the AI, but aligning its reaction timing with human capabilities to preserve fairness. Source: https://news.ycombinator.com/item?id=26289933
The takeaway:
It’s standard practice in fighting games to deliberately “slow down” AI reactions — because developers know raw input speed breaks fairness.
3. “It was never a problem before” — respectfully, it was.
It just wasn’t clearly understood. For years, players have described: • Specials being blocked mid-combo. • Intercepts being impossible/unsafe. • Being light intercepted/backdraft intercepted far more often. • Relics being blocked on the 2nd or 3rd hit. • AI being able to parry us even without masteries (eg story content,) simply because it is triggering block “frame perfectly.” • “Parry not working” when it was actually mistimed due to AI speed.
Once Kabam admitted the AI was frame-perfect at 60fps, it explained a lot of those moments. We’re not inventing a new problem — we’re just finally naming it.
And we’re bringing it up now, before AI 2.0 locks in, because that’s when it can still be fixed without breaking everything else.
4. One last technical point: programmable delays ≠ hardcoded reaction lag.
No one is asking Kabam to hard-code permanent 200ms delays across the board. We’re asking for a tunable delay layer between recognition and input — something that simulates human reflex thresholds (like 100–200ms) where appropriate.
This is industry-standard, not weird. Fighting game devs use it because it works.
TL;DR — My Position, Again: • Yes, fix AI behavior. • Yes, make it smarter. • But also, give it human-like mechanical limits, not robotic freedom. • That’s how you make fights fair, skill-based, and still challenging — not oppressive.
This isn’t a conspiracy theory. It’s a simple, proven concept used across the fighting game industry, and MCOC deserves that same standard of fairness - especially in high-end content and Battlegrounds.
Happy to keep the discussion going — but wanted to put this part to rest properly. Thanks again for engaging.
If you keep referencing the fact that you can’t punish every special or — even dumber — that you can’t string together a MLLLH combo, you just continue to prove you don’t know what you’re talking about.
3. “It was never a problem before” — respectfully, it was.
It just wasn’t clearly understood. For years, players have described: • Specials being blocked mid-combo. • Intercepts being impossible/unsafe. • Being light intercepted/backdraft intercepted far more often. • Relics being blocked on the 2nd or 3rd hit. • AI being able to parry us even without masteries (eg story content,) simply because it is triggering block “frame perfectly.” • “Parry not working” when it was actually mistimed due to AI speed.
Okay, I'm tapping out now. I'll just say that this AI speed thing is not the only thing you're wrong about. There have been several instances where Parry had issues, and in fact the two biggest were related to Unity timing issues, not AI speed. The most recent was due to a bug in Unity that was actually *fixed* in the Unity build, which threw off the way MCOC managed time. This was discussed at the time. This was the most recent instance of Unity changes affecting the timing of the game in odd ways: going all the way back to 12.0 when people thought the 12.0 changes broke Parry, what actually happened was that at around the same time MCOC's Unity engine was upgraded for supportability reasons, and that caused many intrinsic behaviors in the game to change, Parry timing among them.
None of this has anything to do with AI speed per se. Nor are these guesses: the Unity iOS time window bug was well documented at the time, and Kabam confirmed that specific bug threw off Parry timing within the game causing it to sometimes fall outside the well-timed block window.
At this point, the technical level of the conversation has absolutely zero chance of being elevated, so it would be pointless to continue it. You're just reiterating the same points over and over without acknowledging anything that would contradict them; not even disagreeing with them simply ignoring them completely.
@DNA3000 Appreciate the discussion, genuinely. Just to clarify — I never said Unity timing bugs weren’t real. I’m aware of them, and they’ve absolutely affected Parry before. I also never said that the Ai speed was the sole reason for parry inputs appearing to fail. My point was that some Parry complaints over the years were likely due to AI speed, not all of them — and that now, with Kabam confirming the AI’s frame-perfect input speed at 60fps, we can finally explain some of those edge cases that felt off but weren’t formally acknowledged, along with a whole host of other AI issues (as listed in my previous responses).
That aside, I’ve provided multiple sources, examples, and developer-backed principles that support the core argument. If we still disagree, fair enough — but I think I’ve made the case clearly and constructively enough for it to warrant internal escalation.
In all honesty, throughout this thread I’ve seen you attribute various issues in the game to the AI’s reaction speed. I’m not saying those issues don’t exist, but the AI’s reaction speed is not the cause of those issues. Most of those issues only exist because the AI has the opportunity to take advantage of those issues, which is something that we, as the players, can also do. The AI’s reaction speed doesn’t need to be changed, the original issues need to be fixed.
Comments
To be absolutely clear:
1. Every fighting game processes information faster than humans do. Every single fighting game.
2. Every fighting game could out-react the player if it chose to do so. Every single fighting game.
3. Every fighting game doesn't out-react the player because it chooses not to. This is an explicit choice built into the AI system. It can, but it is programmed not to, most of the time.
Every. Every. Every. Prove me wrong, find a counter example. I would love to see it. Maybe there is one out there somewhere with some sort of weirdo sophisticated reaction time simulation, but I doubt it. At least one of the very games you cite is most definitely not doing that because in spite of the fact you think it doesn't it demonstrably does have instant reactions built into the AI at highest difficulties.
At this point, this isn't about the AI. This is about you being stuck on an idea, and for some reason willing to die on a hill you are simply insufficiently informed about. At this point I believe there exists no work or reference that will convince you otherwise. I think if the developers of every fighting game you ever played said you're mistaken, you'd simply repeat what you said directly to them as if they did not understand what you were talking about. Including the developers of this one. Which I suppose is a cautionary warning unto itself.
My take - was the AI being a “frame perfect button masher” a problem before you knew about it?
It existed the entire lifespan of the game; there weren’t any real persistent concern that the AI was responding faster than humans.
If it explains all the problems we are seeing, then sure you have a valid concern. But if its not the root cause of the problems (which Kabam isn’t specifically blaming but is saying we need more sophisticated tools to fix problems when they come up), then I really don’t understand your concern even if it is still present in 2.0.
It is as if the AI is "trying" to figure out ways to do the unexpected, because it is constantly tapping the screen and constantly trying all kinds of things at every possible instant of time. Because of that, if a maneuver should not be possible because it wasn't intended, but because of a quirk of the game there is a single animation frame where it becomes possible, the AI will find it and do it.
Someone might think that the problem, then, is that the AI is too fast - it is capable of finding this momentary windows of opportunity. But speed is irrelevant. If the AI were slower, it would still find these windows of opportunity, it would just find them less often. If there is some momentary weirdo window of time where the AI could do something bizarre, and it tries to do it 60 times per second it will definitely find that moment and do that thing. But if you slow the AI down to only taking actions ten times per second, or once per second, or once per week, it will still find that exact moment in time and do that weird thing, it will just happen much less often.
And that is actually what happens. Although the AI is super fast and is constantly making decisions sixty times per second, it doesn't try *everything* every animation frame. The AI dev diary makes that clear. The AI only tries certain actions at certain times, which means if there's a way for the AI to pull a fast one on the player, it might only happen when that exact window of opportunity arises *and* the exact action necessary to pull it off happens to be in the AI's current random option list *and* it happens to randomly select that action. So when a player says "hey, the AI tea bagged me in the middle of a combo" and the devs try to reproduce that problem to try to fix it, it doesn't happen to them. It doesn't happen to lots of other players either. And even when they finally do reproduce the issue, they might find there's no way to make it go away, because if they remove the action from the set of possible actions it will not only remove that issue, it will also prevent the AI from doing that in other contexts. It could even make the AI worse, because when you remove an option, that will force the AI to start doing other things more often, and those other things might be even worse in other situations. You could unmask a new issue somewhere else while trying to fix the issue in front of you.
Getting back to speed. Suppose instead of doing things the smart way, we do it the weirdo way. Instead of just making the AI better at producing the behavior we want, we just introduce a delay into the AI. We say whenever you see XYZ and you decide to take an action, delay that action by 100ms to replicate human reaction times. This would be horrible for a number of reasons. To illustrate, imagine the devs want the AI to hold block when it thinks the player is going to attack. The delay would delay going into the block state by approximately the human reaction time. This could randomly turn the intended block into a coincidental well-timed block. Which would be bad. What's more, because this reaction time is hard coded into the AI, there's no way to fix it. That's basically the problem(s) we have with AI 1.0 now.
Make the delay variable so we can tune it to anything from 0 to 250ms? Well, on top of that being more complex and more difficult to both program and troubleshoot this only works if the AI's is fast enough to allow this in the first place. If you're going to allow a programmable delay, the AI must be as fast as the minimum possible delay. Which is going to be what, exactly? 0ms. Because only an insane person would hard code a minimum delay into the AI and on top of that add a programmable delay. If you're going to put in a programmable delay, which is no different from implementing action scripts that fire instantly but include a delay in them, the intrinsic delay in the AI should be zero, to give the designers maximum flexibility in tweaking that delay.
Let me clarify the concern again — not because I’m “stuck on a hill,” but because I think we’re mixing up two separate issues:
1. AI behavior vs. AI reaction time.
Yes — MCOC’s AI behavior needs fixing.
Yes — AI 2.0 should give the devs better tools for predictability, tuning, and smarter moves.
But none of that fixes the underlying problem if the AI still reacts at 60fps — 16.7ms per decision window — while human players operate at ~200ms.
When the AI has zero input delay, and players have animation recovery + input lag + human reflex limits, we end up in situations like:
• Dashing in after a dexing special and getting blocked or punished instantly.
• Mid-combo actions (MLLLH / medium combo > special) being interrupted, blocked or punished, so having to resort to MLLL because the animation is shorter.
• Dash-ins/medium intercepts being unsafe now and being punished in real time even when timed/spaced correctly.
• Relics being somehow blocked on the 2nd or 3rd hit when the first hit landed.
That’s not about “how smart the AI is” — it’s about mechanical limits. If you fix behavior but keep the AI’s reaction speed inhuman, you still punish correct player decisions. That’s the issue.
2. Yes, other fighting games absolutely limit AI reaction speed — by design.
You asked for receipts. Here are just a few examples:
• Super Smash Bros. Melee:
AI researchers working with the Slippi team simulated human-like agents by intentionally capping AI reaction time to ~200ms to keep fairness.
Source: https://arxiv.org/abs/1810.07286
• Skullgirls:
AI difficulty levels include built-in reaction delays to simulate human reflexes — the devs even made those delay values adjustable during development.
• Street Fighter IV & V:
Developer notes (via GDC) confirm that AI response patterns are built with delayed inputs at lower difficulties, and reaction capping is standard even at high levels to preserve fairness.
• Mortal Kombat X & 11:
Same deal — AI difficulty sliders include changes to how fast the AI can block/react, not just how much it combos.
(This is directly visible in practice when testing frame traps.)
As further evidence, game developers themselves widely acknowledge this principle. On Hacker News, a respected dev forum, one commenter (alexflint) explains:
“One of the easiest tricks that game designers use to make AI feel more human is to simulate human reaction time… the AI is coded to wait 200ms before reacting.”
This reflects exactly what we’re suggesting — not dumbing down the AI, but aligning its reaction timing with human capabilities to preserve fairness.
Source: https://news.ycombinator.com/item?id=26289933
The takeaway:
It’s standard practice in fighting games to deliberately “slow down” AI reactions — because developers know raw input speed breaks fairness.
3. “It was never a problem before” — respectfully, it was.
It just wasn’t clearly understood. For years, players have described:
• Specials being blocked mid-combo.
• Intercepts being impossible/unsafe.
• Being light intercepted/backdraft intercepted far more often.
• Relics being blocked on the 2nd or 3rd hit.
• AI being able to parry us even without masteries (eg story content,) simply because it is triggering block “frame perfectly.”
• “Parry not working” when it was actually mistimed due to AI speed.
Once Kabam admitted the AI was frame-perfect at 60fps, it explained a lot of those moments. We’re not inventing a new problem — we’re just finally naming it.
And we’re bringing it up now, before AI 2.0 locks in, because that’s when it can still be fixed without breaking everything else.
4. One last technical point: programmable delays ≠ hardcoded reaction lag.
No one is asking Kabam to hard-code permanent 200ms delays across the board.
We’re asking for a tunable delay layer between recognition and input — something that simulates human reflex thresholds (like 100–200ms) where appropriate.
This is industry-standard, not weird. Fighting game devs use it because it works.
TL;DR — My Position, Again:
• Yes, fix AI behavior.
• Yes, make it smarter.
• But also, give it human-like mechanical limits, not robotic freedom.
• That’s how you make fights fair, skill-based, and still challenging — not oppressive.
This isn’t a conspiracy theory. It’s a simple, proven concept used across the fighting game industry, and MCOC deserves that same standard of fairness - especially in high-end content and Battlegrounds.
Happy to keep the discussion going — but wanted to put this part to rest properly. Thanks again for engaging.
None of this has anything to do with AI speed per se. Nor are these guesses: the Unity iOS time window bug was well documented at the time, and Kabam confirmed that specific bug threw off Parry timing within the game causing it to sometimes fall outside the well-timed block window.
At this point, the technical level of the conversation has absolutely zero chance of being elevated, so it would be pointless to continue it. You're just reiterating the same points over and over without acknowledging anything that would contradict them; not even disagreeing with them simply ignoring them completely.
Appreciate the discussion, genuinely. Just to clarify — I never said Unity timing bugs weren’t real. I’m aware of them, and they’ve absolutely affected Parry before.
I also never said that the Ai speed was the sole reason for parry inputs appearing to fail.
My point was that some Parry complaints over the years were likely due to AI speed, not all of them — and that now, with Kabam confirming the AI’s frame-perfect input speed at 60fps, we can finally explain some of those edge cases that felt off but weren’t formally acknowledged, along with a whole host of other AI issues (as listed in my previous responses).
That aside, I’ve provided multiple sources, examples, and developer-backed principles that support the core argument. If we still disagree, fair enough — but I think I’ve made the case clearly and constructively enough for it to warrant internal escalation.