Are we Ready for Some Football?

[SlowGettingUp]

Don't you have to put sensors under the playing surface?

An AI vision system with multiple camera angles wouldn't need sensors. Sometimes it would have to make its best guess if the ball or knees are obscured, but that's no different from refs right now. Heavy rain/snow/fog might be a problem though unless it was lidar-based (much more expensive). Same issue Tesla will have vs Waymo.

 

[XLIX]

If you define forward progress as when the ball stops moving forward, then yes.

Except that would require a major, major rule change.

Not by to mention a computer monitoring the ball isn’t going to know if the player went backwards on his own or if there was contact.

Supposed a QB takes the snap and drops back to pass. He’s pressured so he runs diagonally backwards. He gets sacked. According to your computer, the snap of the ball is the forward progress spot.

Knee down is harder, but not beyond what an AI vision system could do with multiple camera angles.

Oh wonderful. Let’s take a ten minute break between all plays so the ref can get the ruling from the AI vision system.

 

[XLIX]

An AI vision system with multiple camera angles wouldn't need sensors. Sometimes it would have to make its best guess if the ball or knees are obscured, but that's no different from refs right now.

EXACTLY.

It’s no different from what the refs are already doing right now. So it’s not some sort of infallible solution.

 

[State]

No, not with Stubby and Go Down to IR Moses as our starting OTs.

 

[SlowGettingUp]

Supposed a QB takes the snap and drops back to pass. He’s pressured so he runs diagonally backwards. He gets sacked. According to your computer, the snap of the ball is the forward progress spot.

I'm not talking about a dumb vision system like they use in tennis (or the coming ball/strike system in baseball). An AI could absolutely be trained on all the different cases - once you have a good model it can see whether someone was pushed back or moved back on their own. With a smart system you could keep the rules exactly as they are now. And it would be pretty much instantaneous. No bias, and with multiple camera angles it could absolutely do a more accurate job than can refs.

But don't hold your breath.

 

[XLIX]

I'm not talking about a dumb vision system like they use in tennis (or the coming ball/strike system in baseball). An AI could absolutely be trained on all the different cases - once you have a good model it can see whether someone was pushed back or moved back on their own. With a smart system you could keep the rules exactly as they are now. And it would be pretty much instantaneous. No bias, and with multiple camera angles it could absolutely do a more accurate job than can refs.

 

[SlowGettingUp]

If you keep the current system but allow AI challenges then there would be zero waiting - by the time a challenge was made the AI would already have the answer.

The key is they have huge amounts of video data for training.

 

[Real fan 02]

I was hoping Mapu would develop into a Matt Milano type. It looks like Vrabel likes the torpedo type LBs.

Agreed. He has talent. Just don't know what the issue is with him. Injuries? Fit? If anything I felt he could be a nice chess piece to spy Running QBs and could add depth. Seems like the top guys are Landry, Spillane and Chaisson at the top of the depth chart... Jennings seems to be falling down down the depth chart also. If we see Jennings playing into the 4th quarter this week then he's likely auditioning for another team.

 

[XLIX]

If you keep the current system but allow AI challenges then there would be zero waiting - by the time a challenge was made the AI would already have the answer.

The key is they have huge amounts of video data for training.

aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaand how do you "train AI" when you have countless videos that people look at and still can't agree on what the right call is?

 

[One-If-By-Sea]

Agreed. He has talent. Just don't know what the issue is with him. Injuries? Fit? If anything I felt he could be a nice chess piece to spy Running QBs and could add depth. Seems like the top guys are Landry, Spillane and Chaisson at the top of the depth chart... Jennings seems to be falling down down the depth chart also. If we see Jennings playing into the 4th quarter this week then he's likely auditioning for another team.

I think you are right to add Jennings to the Mapu conversation. I think both are more thoughtful, not as violent, and more in the BB prevent defense mold. The Raiders tried to sign Elliss after they lost Spillane because both are "missile" type LBs, not like Mapu or Jennings. That is the only explanation I know. Both Jennings and Mapu have to think less and be more violent.

 

[Real fan 02]

I think you are right to add Jennings to the Mapu conversation. I think both are more thoughtful, not as violent, and more in the BB prevent defense mold. The Raiders tried to sign Elliss after they lost Spillane because both are "missile" type LBs, not like Mapu or Jennings. That is the only explanation I know. Both Jennings and Mapu have to think less and be more violent.

I believe your right. The defense is an attacking style more of a modern type of defense to attack passing offenses. I really wish we'd had been this style of defense before. But to your points both of those guys seem like long shots.

 

[SlowGettingUp]

aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaand how do you "train AI" when you have countless videos that people look at and still can't agree on what the right call is?

99.9% of calls are clear if you get to review them in slow motion. The edge cases are going to be problematic in any system, and worst case you mark them as "unclear." If you have a built-in physics model that knows about velocity, acceleration and momentum of players and the ball you can do a lot more than you can with just a video. For example you can figure out (to some degree of precision) when a knee will hit the ground even if it is obscured.

Just idle talk anyhow - such a system would cost a lot and there is no momentum to ever develop it.

 

[One-If-By-Sea]

I believe your right. The defense is an attacking style more of a modern type of defense to attack passing offenses. I really wish we'd had been this style of defense before. But to your points both of those guys seem like long shots.

I think they are both good players but I doubt there is any trade value.

 

[XLIX]

99.9% of calls are clear if you get to review them in slow motion.

That's right - but those aren't the plays we need help with. No one argues over a 3 yard gain on 1st and 10.

The edge cases are going to be problematic in any system,

EXACTLY. There's nothing AI can do to solve the problem. There are just far, far too many variables football.

If you have a built-in physics model that knows about velocity, acceleration and momentum of players and the ball you can do a lot more than you can with just a video. For example you can figure out (to some degree of precision) when a knee will hit the ground even if it is obscured.

Fans will not tolerate a system that determines a knee is down even though it is obscured.

 

[SlowGettingUp]

That's right - but those aren't the plays we need help with. No one argues over a 3 yard gain on 1st and 10.

EXACTLY. There's nothing AI can do to solve the problem. There are just far, far too many variables football.

Fans will not tolerate a system that determines a knee is down even though it is obscured.

Refs make calls like that all the time when they spot the ball or decide on a fumble where the knee may be down. An AI would at least be totally impartial.

 

[XLIX]

Refs make calls like that all the time when they spot the ball or decide on a fumble where the knee may be down. An AI would at least be totally impartial.

Impartiality is not the problem. Accuracy is.

 

[jmt57]

If you keep the current system but allow AI challenges then there would be zero waiting - by the time a challenge was made the AI would already have the answer.

The key is they have huge amounts of video data for training.

100% true.

My concern is with gambling becoming so prevalent and acceptable in sports,
that software could be compromised by those in search of a big payday.