- Feb 1, 2017
- 3,924
- 6,824
It's an interesting conversation. Presumably it could take into account all kinds of historical data, player performance, etc. This could result in the most "efficient" play call given the scenario.
However, the main concern I see is from a Game Theory perspective. If the opponent is aware that this method is being used, they could likely identify the most likely play calls. Basically, you could reverse engineer what is coming. Then it comes down to a Game Theory perspective of trying to program the AI to out-smart your opponent, which would be difficult to program in.
AI isn't exactly an area I'm overly knowledgeable on, so I may be way off. But that would be my $0.02 gut reaction.
Play calling is strategy.
I think your right but how long before we start seeing something like this? Do you think the entire game plan could be devoloped by AI then given to the coach to prepare?Obviously it can't be random and would need source data. You would need for relevant data (ball position, down-yardage, opposing team formation, players that are in the game ... and so on) to be an input.
But on the whole ... I don't like it. I prefer seeing a chess match between coaches/coordinators with the players executing it on the field as they have been coached up to do.
I think UGA would have loved it the last 5 minutes of the game last week. But hell my pet Labrador could have called a better end to that game than Kirby did.There have been many times over the last several years that I would have preferred it.
That's why I was referencing the Game Theory perspective. Essentially, it becomes a repeat game where there would be learning of the opponent involved. It's been quite a few years since I took my game theory course, so I'm sure there are lots of current methods to introduce randomness and ways for the AI to "out-smart" a human in a repeat game process.That is an intreresting take. However it seems to me even if the other team knows your doing it AI would pick up on it and change the play of course this is an infinite loop.
My understanding is pretty much like we do. Repetitive learning of outcomes strengthens the associations while we burn synapse paths the computer does something akin to that. But much much faster.That's why I was referencing the Game Theory perspective. Essentially, it becomes a repeat game where there would be learning of the opponent involved. It's been quite a few years since I took my game theory course, so I'm sure there are lots of current methods to introduce randomness and ways for the AI to "out-smart" a human in a repeat game process.
It would be like teaching a robot how to beat a human at Rock-Paper-Scissors. A human that is purely random would be tough to outsmart which should result in the computer never gaining an advantage. But, if the human has tendencies that the robot can identify and the human does not recognize their own tendencies enough to alter over time, then the robot should be able to win at a higher than expected probability.
The success of this would be very dependent on the programming I would imagine. If it could be programmed correctly with the right data, assumptions, etc, then I could see how it could be possible. Like I said, I'm not very advanced in my understanding of AI though, so I don't know how the learning process progresses for the AI.
Do you think the entire game plan could be devoloped by AI then given to the coach to prepare?
From the defensive coordinator perspective, I think you could use some kind of modeling to determine the odds on what play might be run out of a particular formation. Of course, even a computer couldn't figure out the mysteries of what play might occur when we bring Emory Jones in for the Wildcat formation. There are so many possibilities, it could be a quarterback keeper, or a quarterback keeper, or perhaps even a quarterback keeper.