AI Play Calling
- Thread starter PCGatorAlum
- Start date
- Sep 12, 2015
- 154
- 110
Isn’t this what those game simulators are doing? Taking all of the data on players, games, etc and simulating the matchups? Granted AI is supposed to learn over time, but this is used for repetitive functions that can be measured. Players in a football game is too random once the ball is snapped.
Besides the difficulties I hate the idea that this could happen. We would never get to see coaching debacles like Kirby and Chimp trying to screw themselves out of a win.
Besides the difficulties I hate the idea that this could happen. We would never get to see coaching debacles like Kirby and Chimp trying to screw themselves out of a win.
- Dec 31, 2018
- 14,867
- 28,101
- Jun 9, 2014
- 78,386
- 110,660
Founding Member
Kind of surprised that Fooley sprung for that color Mac.
Here’s the AI units for Mac, Nuss, and Nord.
- Feb 1, 2017
- 3,921
- 6,821
In case you haven't noticed nerds rule the world.
Last edited:
- Jun 13, 2014
- 1,036
- 1,330
Founding Member
- Oct 8, 2017
- 2,162
- 5,568
Here’s the AI units for Mac, Nuss, and Nord.
Well done. Very well done. I love this place.
- Jun 12, 2014
- 5,395
- 9,394
Founding Member
The way most nascent AI works is that you start with a bunch of 'truth' data that has examples of inputs and outputs, and the computer 'learns' what input leads to what output. For example you could start with truth data that includes thousands of pictures of cars and thousands of pictures of not cars and using machine learning the computer can figure out whether a new picture is a car or not. That's without any human programmer providing rules for how to determine whether it's a car.
The success of the AI is highly dependent on the quality and quantity of the truth data. That's where I think the challenge would be for calling football plays. You don't have a bunch of truth data of what happened when you ran your plays in real game conditions against a specific opponent. And if you are installing new plays and wrinkles, you don't have any truth data outside of a few weeks of practice.
Anyway, I'd guess that we are pretty far away from anything much more than looking at statistics and tendencies.
The success of the AI is highly dependent on the quality and quantity of the truth data. That's where I think the challenge would be for calling football plays. You don't have a bunch of truth data of what happened when you ran your plays in real game conditions against a specific opponent. And if you are installing new plays and wrinkles, you don't have any truth data outside of a few weeks of practice.
Anyway, I'd guess that we are pretty far away from anything much more than looking at statistics and tendencies.
- Jun 11, 2014
- 21,077
- 33,799
Founding Member
- Aug 28, 2014
- 16,541
- 26,096
Not AI, It's an Expert System. AI not really a thing.
So yeah an Expert System (like the ones that play chess) would be interesting against a human.
However if they play each other they'll just blow up like in the movie War Games, because when a computer thinks too hard it overheats. Everybody knows that.
So yeah an Expert System (like the ones that play chess) would be interesting against a human.
However if they play each other they'll just blow up like in the movie War Games, because when a computer thinks too hard it overheats. Everybody knows that.
- Jun 12, 2014
- 2,094
- 5,807
Founding Member
So, for example, let's say a team has a tight end that can't block, can't run routes, and can't catch. And let's say this tight end can sort-of throw the ball. That's the 'truth data.' We'll call this hypothetical guy some computery name like "Krull." So when the team 'inputs' Krull in the game, the computer has 'learned' that the 'output' is likely that he'll be handed the ball, then execute a throw-back pass to the quarterback who is hobbling around on an injured knee. Correct?
- Jun 11, 2014
- 34,945
- 33,041
Founding Member
- Jun 22, 2014
- 7,485
- 14,474
Founding Member
- Jun 11, 2014
- 34,945
- 33,041
Founding Member
- Jun 22, 2014
- 7,485
- 14,474
Founding Member
- Nov 16, 2017
- 5,547
- 13,084
Tennessee just did an experiment with this actually and reached an interesting conclusion. Some footage got leaked:
- Jun 9, 2014
- 78,386
- 110,660
Founding Member
