00:00:00
Dan Loney: Well, certainly the world of
sports has changed quite a bit in the last couple of decades
with the advent of big data and
00:00:07
the massive sums of money being
brought in. Now, we are obviously in the advent of AI, which is
already having an impact. How
00:00:15
much and how much might sports
continue to develop? We bring in Cade Massey, who's a Practice
Professor in the Operations,
00:00:23
Information and Decisions
department at the Wharton School. You also hear him as
host of the Wharton Moneyball
00:00:29
podcast. Hi Cade, how are you, sir?
Cade Massey: Good, Dan. Good to see you. Thanks for having me. - You know,
you and I are obviously big
00:00:36
sports lovers, and so
you've watched it a lot, this side of it, a lot closer than I,
but it has been just amazing to
00:00:43
see how sports, in general, the
operation, the playing on the field, has changed in the last
couple of decades because of
00:00:50
big data, and now, in part,
because of AI. - Yeah, I mean,
something's obvious. People
00:00:57
complain even about how much it's
changed baseball, how much it's changed basketball. It's not
obvious that it's always changed
00:01:04
it for the better. It's just
accelerating optimization. And sometimes, when that happens,
you realize maybe the rules need
00:01:10
to be tweaked, maybe
optimization for these particular rules isn't what we
actually want. And so, you know,
00:01:15
basketball is thinking about
changing. Baseball is always experimenting with changes. But
big data has definitely
00:01:21
accelerated that in multiple
sports over recent years. - Is there a moment in
time that you can think of that
00:01:28
really kind of started to make
the turn occur?You know, whatever the sport was.
But the involvement, I guess
00:01:35
maybe the moneyball kind of
philosophy kind of coming into baseball kind of started this
process.
00:01:40
- Yeah, I mean, you know, all
sports, they say this about the NFL, but it's really true about
all leagues, are copycat
00:01:46
leagues. And so when somebody
sees something that works, really kind of whether it's by
chance or not, it gets copied.
00:01:53
And then in some sports, that
doesn't last very long, you know? People got a little
curious about going for two
00:01:59
until Belichick went for two on
his own 28, or whatever it was, and didn't get it, and all of a
sudden set the whole thing back
00:02:05
for years. But with baseball,
the moment I think about is with the shifts. You know,
people had shifted for decades,
00:02:12
but whenever
advanced analytics started suggesting, well, we can really
figure out player tendencies,
00:02:19
and we can position our
defensive players in a way that substantively changes their
success getting on base, and one
00:02:25
team adopts it, they
adopt it kind of extremely, right? So you see people do
things they hadn't done in
00:02:30
decades. They have some success,
and it just runs through the league. And I don't know what
year that was. IYou know,
00:02:37
2017 or something like that,
but you saw this was an analytics driven innovation, and
as soon as it was proven to be
00:02:44
advantageous, it was contagious
in the league and other teams adopted it. - So I guess is it any surprise to
you then that, with the advent of
00:02:53
a lot of this data and how it's
being used in terms of coaching and playing the game, that
obviously, front offices
00:03:02
are using it as kind of a
marker in terms of how they're paying out contracts and the
length of contracts, and how,
00:03:09
you know, these business
decisions are being made by front offices around the world
of sports?
00:03:16
- Well, you know,
probably the single biggest example, we just had a game last
night, Dan, that was a real high
00:03:23
watermark for the analytics
world with Florida under Todd Golden winning the NCAA Men's
Championship. We haven't had
00:03:31
that many moments like that
around sports. So obviously, Theo Epstein with the Red Sox and the
Cubs in baseball. There's not
00:03:38
been one in NHL. Arguably, there
really hasn't been one in the NBA. I mean, Brad Stevens is not
really an analytics guy, per se,
00:03:47
with the Celtics. Daryl Morey,
he's famously never won. In football, it's kind of an
open question, because Howie
00:03:55
Roseman is not strictly
analytics. He's a salary cap guy, right? But they've really
been, obviously, successful now
00:04:03
with almost two different teams.
And Howie's approach to salary cap is what you're talking
about, is much more analytics-y,
00:04:10
Moneyball approach to salary
cap, and that has really run through the NFL, and that has
made big changes across the NFL.
00:04:17
And it's not quite AI, but it's
a predecessor to AI, and it's in a different part of the
building.
00:04:24
- So how is AI having an impact in
the world of sports right now? - Well, it's still an open
question. That's the big thing,
00:04:31
Dan. It reminds me of when
motion tracking hit football, you know, probably six, seven
years ago, and you'd talk to teams
00:04:38
and they'd say, "What's going to happen?"
Like, I don't know, but you need
00:04:41
to be involved, because things
are going to change as a result. So get the data, start getting
used to metabolizing it and
00:04:47
using it, because it's going to
matter. It feels the same way with AI right now. We don't know
exactly what the consequences
00:04:52
are going to be, but
they will be there. So if you think about sports broadly, we
divide it from an analytics
00:04:58
world, we divide it into a game
day, in game strategy. And then there's the personnel side, all
the scouting. And then there's
00:05:06
high performance, like
individual player and injury prevention, that kind of thing.
And if I had to put my chips on
00:05:12
one of those three areas, Dan, I
would put it on the latter. I'd put it on high performance. And
the thinking, my thinking is
00:05:18
the in game strategy, it's not
quite optimized yet, but it's pretty close. I mean, these
teams have been playing under
00:05:24
these rules for a long time. I
don't think there's any huge gains that AI is going to
generate there. On personnel,
00:05:29
there's a different kind of
problem. Personnel, yeah, we're going to get some gains out of
it, for sure, and we could dive
00:05:35
into that, but you're never
going to forecast personnel perfectly. It's just, we
can't expect perfect forecast.
00:05:43
And so there's a limit to the
advantage of this kind of data. But on performance -- go ahead. - That's
the dynamic, because when you
00:05:50
talk about personnel, there are
so many other components that could come into play that can
change the path of a player in
00:05:57
terms of how he performs, he or
she performs on the field, not just the data itself.
00:06:01
- That's right. I mean,
there's a lot of path dependence there from his life or her
life, but also teammates, from
00:06:09
chance, from coaching staff. I
mean, there's irreducible uncertainty. This is
the fancy phrase we use in my
00:06:16
world, irreducible uncertainty.
And as forecasters, as human beings, we hate that, we're loath
to accept it. But on the
00:06:23
personnel side, you're never
going to perfectly forecast, we're just not. I mean, maybe in
a thousand years, I don't know,
00:06:29
whenever we think -- what, right now,
we think is random, we figure out what's not random about it.
Right now, we just lump all the
00:06:34
things we don't understand into
randomness. That's going to be that way for a while. The domain
that is more susceptible, I'd
00:06:41
say, is the performance side.
Just take injury prevention, for example. It's probably the
biggest frontier, the biggest
00:06:48
margin that we can make
improvements on in sports, and it's still pretty wide open. I
mean, people are still figuring
00:06:55
out, how can we anticipate
injuries better? How can we prevent them better? And the
data that can go into that are
00:07:02
almost unlimited, and the models
we need to pulling signal out of those data are yet to be
invented. And I think that's
00:07:09
probably the biggest frontier,
biggest margin. - One final thing, because I also
saw an article that talked about
00:07:14
how AI is going to have an
impact on the refereeing or the judging of sports as well, which
I think is very interesting,
00:07:22
because there's obviously,
depending on the sport, there's a lot of conversation about, you
know, are the referees good? Are
00:07:28
the umpires good, or are they
not? Should they be replaced by technology? You know, I
mean, obviously that's kind of
00:07:33
a next area to watch as well.
- Well, you can think about the -- I think
that's interesting, Dan. There's
00:07:38
a few different things that
strike me straight away. One is like the baseball umpires, of
course. I mean, those guys could
00:07:43
be almost perfectly replaced.
Obviously, there's some limits, but calling balls and strikes is
pretty straightforward. What if
00:07:49
we had models, though, that
evaluated referees in soccer, for example, and had a sense of,
here's what this ref is
00:07:55
attending to, here's what this
ref is missing, here's what this ref is good at, this is what the
ref needs to work on. Could we
00:08:02
evaluate refs in that way? Could
it be a tool to make them better? In a way that we use -- right now,
we're using, you know, data on
00:08:09
pitches to improve umpires over
time. Could we do something with AI to improve referees? But
then, obviously, the easy one to
00:08:16
think about is, all these sports
that have judges, like Olympic sports that have judges, right?
That dive. How do we know who's
00:08:24
really saying which dive is a
better dive in the Olympics? And could a model help us do that a
little more precisely? Even if
00:08:31
it's just an assistant, you
know, kind of an advisor to an actual judge. - Cade, great insight. Thanks very
much. - Thank you, Dan. Any time.
00:08:40
- You got it. Cade Massey, who's
Practice Professor in the Operations, Information and
Decision Department here at the
00:08:46
Wharton School, and you also
hear him as host of the Moneyball podcast.