OT- AI program can predict opposition moves to your plays

Submitted by iawolve on October 4th, 2019 at 10:29 AM

This actually a pretty cool use case for AI. It is one thing to do film study of other opponents or your game last year, it is another to have the ability to move your players around executing a play and watch the other team respond.

Obviously, this is built on probabilities and not exact, but I find it pretty fascinating since now you have the option to simulate new tactics or tweaks to your base plays with feedback on the potential for success.

https://www.sciencemag.org/news/2019/09/watch-ai-help-basketball-coaches-outmaneuver-opposing-team

UP to LA

October 4th, 2019 at 12:58 PM ^

Eh, the immediate dividends to something like this seem like they'd be pretty modest -- it's not like experienced coaches have a ton of blind-spots with respect to how different sets get defended.

Where more sophisticated spatially-aware machine learning approaches *do* seem to offer the potential for competitive advantage are in evaluating player/roster combination contributions, especially on defense. But that's more of a roster management issue, and probably more impactful in NBA front offices.

bronxblue

October 4th, 2019 at 2:14 PM ^

That's my thinking as well.  If you know what the play is going to be it's relatively simple to concoct a defense to stop it.  But what will be interesting is when they can insert actual player "statistics" or characteristics in for those little dots and then figure out which best pairings can defend/execute said plays.  So if you know James Harden can run at x speed and Draymond can get from the baseline in y, that could be entered in and make a better defensive system and roster composition.

Bo248

October 4th, 2019 at 10:55 AM ^

Maybe the players could wear Google glasses to know wear to attack/defend?

Threadjack: whats your super power or what would you want if you were playing?

   Me- brain wave disruption (kinda like AI)

 

Bodogblog

October 4th, 2019 at 11:03 AM ^

Same with the Borg: set phasers to random pattern.  But that will only hold them off for a moment before they adapt.  Boarding party to retrieve Locutus and have Lt. Commander Data put it to sleep is the only option. 

Carpetbagger

October 4th, 2019 at 11:06 AM ^

I realize this isn't true AI, but until google maps can figure out it's quicker to get off the highway one exit past where it has told me to get off for the last 2 years straight I'm not too worried about computers predicting anything but math equations and the past.

If the computers ever tried to take over the world they'd be the most predictable opponent ever and humans would mop the floor with them in about 2 days.

NittanyFan

October 4th, 2019 at 2:11 PM ^

Human beings aren't truly random when we start mashing the rock/scissors/paper buttons on games like that.  For example, after hitting scissors, our next choice may be rock 45% of the time, scissors 30%, and paper 25%.  The computer is simply picking up our patterns. 

It's kind of neat, but neither is the math overly sophisticated.  Could build an "AI" like that in Excel VBA in 2 hours.

Carpetbagger

October 4th, 2019 at 12:58 PM ^

Exactly NOLA. Computers can do one thing better than humans. Math. They can do math millions of times faster than even the best human being.

It may look like "AI" but it's really just really complicated math done really, really fast. GIGO still exists because humans still exist. And humans still build the equations computers run.

OldManUfer

October 4th, 2019 at 6:00 PM ^

That is both mostly inaccurate and misleading where it is accurate. Quite an impressive feat.

There are many things at which computers are better than humans that few people would consider math done very quickly. Humans, on the other hand, are still the only intelligence of which we're aware that can do what could reasonably be considered novel math.