Tuesday, October 22, 2019

Artificial Intelligence and the NFL Draft

It has to happen soon. 

How many times can teams blow draft picks?  How many times do we have to read that one half of quarterbacks taken in the first round do not make it?  And how many times do we have to watch Todd McShay and Mel Kiper, Jr. wax eloquent on potential draft picks when they turn out to be wrong a lot of the time.  Put differently, if either were an NFL executive, they wouldn't last too long.

There's an old saying in law enforcement that eyewitness accounts are not reliable.  Well, what about scouts and scouting?  And then, what about the drills that the players are compelled to do at their pro days and at the combine and matching them up with film studies of how players perform?  Even with all that homework, NFL teams miss a lot of the time.  Tell me of any other multi-billion dollar industry where a business could misfire more than half the time in recruiting key talent and succeed.  You cannot.  But the NFL is a closed system that puts a floor under a team's poor performance -- it just cannot go out of business. 

So what is my solution?  I think that the math guys need to take over more.  They need to measure the best players in the league against a bunch of variables and set a standard for what is success -- and then measure every draft pick against that.  I'm talking micromovements -- how long does it take JJ Watt to get outside on a tackle and then to turn the corner?  How long does it take in one-on-one blocking and a double-team?  How fast does he go in pursuit of a running back running around the end?  How quickly does the average running back hit a hole?  And on and on and on.  With this information, not only would a team be well-situated to draft players who just get it done faster (even if misused at the college level), they also could game plan based upon getting intelligence on the other team's reaction times and micromovement times.  And when I say micromovement, I mean how getting off a snap say 0.15 seconds faster than another defensive tackle might make the difference between a five-yard gain and a tackle for loss?  Right now, it seems that those who write game plans do it by feel -- with widely varying results.

Teams should invest in a super-simulation software that enables them to "war game" against the upcoming week's opponent.  Sure, lots of variable are involved, because you cannot predict totally how, for instance, the Cowboys might use Ezekiel Elliott or whether Cam Newtown will stay in the pocket or take off.  But even then you can figure in tendencies and figure out how to game plan using your available personnel.  So, for example, suppose you have to start a cornerback you just plucked off another team's practice squad who has taken no snaps in the league and who you have to start because you are depleted.  You know that you will have to have a safety give him help, and, in turn that this help comes at a price -- the safety will have to vacate another part of the field to help out this rookie.  So, you'll have to compensate for that, too -- do you blitz a lot, do you play an extra defensive back?  AI software could help you come up with solutions constantly -- even during the game.

For all of its attempts at sophistication -- and my guess is that teams are doing proprietary stuff that they are not sharing -- game planning can be a disaster, as can picking guys in the draft.  By looking at what I call microdetails -- and what makes players successful -- teams will be able to draft better.  And by looking at microdetails within the context of a game, teams will be able to create strategies to enable them to win with the personnel that they have on the field, even if it means abandoning strategies that they had relied upon when everyone was healthy.  After all, it is not comforting to hear a coach say, "well, our strategy was good, but we just didn't execute."  That's a smokescreen for, "If I had a healthy roster, we could have won with this strategy."  An appropriate writer and fan reaction would be the following:  "Hey, coach, if you created a strategy that gave the group on the field a chance, your team would have fared better."  Because it's true.

Now I realize that there are only so many schemes a team can dial up and that a team can get to a point where its personnel is so depleted than even the best AI cannot help it, because too many players on the team are only marginal.  I get that.  But there is a sufficient supply of players out there that creating strategies to win games shouldn't be that difficult in the abstract when using AI. 

Because right now,there are too many broken plays, broken players and broken teams.

2 comments:

Z`ssx said...

Hello
Thanks for your complete content & thnks to share

..
buy telegram members

Charlie Bell said...

In terms of game planning and in-game strategy adjustments, it would be fascinating to see. I suspect we might see an arms race in which Team A uses AI to game-plan against themselves and uses the resulting output to anticipate how Team B (also using AI) will game plan to beat them. Then Team A may plan counter-strategies -- contrary to their usual tendency -- to counter Team B's anticipated moves, and so on, spiraling for as many iterations as the coder/coaches choose...