Why does the AI swing early on 100mph fastballs?

[nomo17k]

While I think the batter AI has improved gradually over the years, what it still does not do well, in my opinion, is how it does not get fooled as organically (i.e., fooled like humans do).

In terms of swing timing, one of the biggest difference I see between human and CPU is how different each responds to some pitch sequence that sets up to fool by timing.

For example, try throwing only fastballs for a while, and you'd expect human hitters to be set up to anticipate only a fastball on the next pitch. If you suddenly lay an off-speed pitch, then you would expect the batter to swing early on it. Online against human players, this is indeed what often happens (keep throwing high and tight fastballs, and typically skilled players would start trying to pull it for a long ball).

But not so against CPU. More often than not, it actually doesn't get fooled by swinging early on the fist off-speed pitch after a long string of fastball-only sequence.

This probably is due to the AI implementation along the line of what KBLover implied (though I don't exactly fully understand what he meant)... that swing timing may be modulated by AI as offset according to each pitch type/speed and based on player attributes and such, but it doesn't clearly develop a notion of speed differential in terms of pitch sequence and, consequently, AI does not act like it swings early on an off-speed pitch when it anticipates a fastball (it swings late on a fastball when it anticipates an off-speed pitch) after being set up. (As a side effect, it can even swing (somewhat unnaturally) very early on a fastball when the modulation is symmetric around being early/late... that is probably what OP saw, and I do see it sometimes as well.)

In addition, I think CPU really does not have a human-like concept in reaction timing. What I mean is that human batters can get fooled really, really badly with a pitcher with very high speed differential between his fastball and change-up (a frequent complaint among pitch editors). This is the reason why James McDonald used to be one of the best starting pitchers in online and Pat Neshek these days is one of the most popular relievers online. Against those pitchers, one has to sit on one and just try to let another go.

But CPU actually does not get fooled by a wide speed differentials like this. Even when you mix fastball and change-up fairly randomly, CPU does not get fooled in swing timing as humans do. Or more precisely, I feel CPU is (for the lack of better word) faking the state of being fooled by modulating timing offset to each pitch, but it doesn't seem to be really "measuring" the pitch speed, hence it is insensitive to the unusual speed differentials that pitchers like McDonald and Neshek do.

All this I'm just saying from my impression so in detail I'm sure I'm getting things wrong. But without knowing the underlying detail, that is how the difference feels when I compare pitching to CPU and human players online. There are some distinct differences in how the "brain" behind the batters work when I actively try to set them up. I often feel CPU doesn't respond to being set up.

[KBLover]

Quote: Originally Posted by nomo17k

This probably is due to the AI implementation along the line of what KBLover implied (though I don't exactly fully understand what he meant)... that swing timing may be modulated by AI as offset according to each pitch type/speed and based on player attributes and such, but it doesn't clearly develop a notion of speed differential in terms of pitch sequence and, consequently, AI does not act like it swings early on an off-speed pitch when it anticipates a fastball (it swings late on a fastball when it anticipates an off-speed pitch) after being set up.

Basically what I was saying - trying to break it into a step-by-step process that the code might be going through and what variables are potentially in play to create a "he got fooled, make him swing bad" without a nod to how hitters get fooled or how/what actually fooled the CPU and why.

Quote: Originally Posted by nomo17k

In addition, I think CPU really does not have a human-like concept in reaction timing. What I mean is that human batters can get fooled really, really badly with a pitcher with very high speed differential between his fastball and change-up (a frequent complaint among pitch editors). This is the reason why James McDonald used to be one of the best starting pitchers in online and Pat Neshek these days is one of the most popular relievers online. Against those pitchers, one has to sit on one and just try to let another go. But CPU actually does not get fooled by a wide speed differentials like this. Even when you mix fastball and change-up fairly randomly, CPU does not get fooled in swing timing as humans do. Or more precisely, I feel CPU is (for the lack of better word) faking the state of being fooled by modulating timing offset to each pitch, but it doesn't seem to be really "measuring" the pitch speed, hence it is insensitive to the unusual speed differentials that pitchers like McDonald and Neshek do.

And this is pretty much what I was talking about with the "geared speed" that the CPU is looking to hit. A more baseball term is what the CPU is sitting on based on what you've thrown in that PA.

I agree with you - a lot of times the CPU doesn't get fooled because of the other processes above (i.e. ratings, confidence levels, sliders, etc.) dictate that the CPU "read" it without actually having to read the pitch.

Along with the system I've seen in other games with "sitting on" a speed is the difficulty in adjusting from that speed...and how much hitters get fooled being out front or behind pitches based on what they are looking for and what they get - trying to replicate the "organic" timing deception you mention.