1
00:00:15,316 --> 00:00:24,036
Speaker 1: Pushkin. About a decade ago, special cameras in the rafters

2
00:00:24,036 --> 00:00:27,196
Speaker 1: of NBA arenas started following all the players around the

3
00:00:27,236 --> 00:00:30,196
Speaker 1: court and tracking the ball everywhere it went. It was

4
00:00:30,236 --> 00:00:34,516
Speaker 1: this huge new trove of data, but nobody really knew

5
00:00:34,516 --> 00:00:37,396
Speaker 1: what to do with it. This is a very familiar

6
00:00:37,436 --> 00:00:40,516
Speaker 1: problem in the modern world, tons of data that just

7
00:00:40,636 --> 00:00:44,596
Speaker 1: kind of sits there. Then a few computer scientists came

8
00:00:44,636 --> 00:00:47,916
Speaker 1: along and had an idea for how to solve the problem.

9
00:00:48,196 --> 00:00:50,676
Speaker 1: They thought they knew how to take all that data

10
00:00:50,756 --> 00:00:55,116
Speaker 1: and turn it into profoundly useful information, and so they

11
00:00:55,156 --> 00:00:58,236
Speaker 1: started a company called Second Spectrum to see if their

12
00:00:58,276 --> 00:01:04,996
Speaker 1: idea would work. I'm Jacob Goldstein, and this is What's

13
00:01:05,036 --> 00:01:07,916
Speaker 1: Your Problem, the show where entrepreneurs and engineers talk about

14
00:01:07,916 --> 00:01:10,116
Speaker 1: how they're going to change the world once they solve

15
00:01:10,156 --> 00:01:14,236
Speaker 1: a few problems. My guest today is Regeev Mahiswaran. He's

16
00:01:14,316 --> 00:01:17,676
Speaker 1: the co founder and president of Second Spectrum. The company

17
00:01:17,796 --> 00:01:20,996
Speaker 1: started out turning video of NBA games into information that

18
00:01:21,076 --> 00:01:24,756
Speaker 1: coaches could use. Today, they work with every NBA team

19
00:01:24,836 --> 00:01:27,596
Speaker 1: and with Major League Soccer teams in the US and

20
00:01:27,796 --> 00:01:31,436
Speaker 1: with Premier League teams in the UK. Regiev's problem is

21
00:01:31,476 --> 00:01:35,156
Speaker 1: this how do you teach a computer to understand sports?

22
00:01:38,236 --> 00:01:41,196
Speaker 1: Regiev and his co founders started Second Spectrum back in

23
00:01:41,276 --> 00:01:44,836
Speaker 1: twenty thirteen. Last year, the company was acquired by a

24
00:01:44,876 --> 00:01:48,676
Speaker 1: sports analytics company called Genius for two hundred million dollars.

25
00:01:49,196 --> 00:01:52,556
Speaker 1: My conversation with Regieve focused on his work with the NBA.

26
00:01:52,636 --> 00:01:55,476
Speaker 1: That's where Second Spectrum has been working the longest. But

27
00:01:55,756 --> 00:01:58,876
Speaker 1: he started out talking about this bigger idea, how do

28
00:01:58,916 --> 00:02:02,476
Speaker 1: you turn data, really any kind of data into something useful.

29
00:02:02,716 --> 00:02:05,076
Speaker 1: So I think this happens everywhere. There's a lot of

30
00:02:05,116 --> 00:02:08,796
Speaker 1: companies that collect data, faster data, more data. But what

31
00:02:09,156 --> 00:02:11,876
Speaker 1: generally happens is the big piles of data. I feel

32
00:02:11,876 --> 00:02:13,636
Speaker 1: like they're like grain. They just if people don't know

33
00:02:13,636 --> 00:02:14,956
Speaker 1: what to do with them, so they just stick them

34
00:02:14,996 --> 00:02:16,836
Speaker 1: in the closet and it's like, yep, I got piles

35
00:02:16,836 --> 00:02:18,996
Speaker 1: of grain in there, and it keeps piling up. And

36
00:02:19,076 --> 00:02:21,636
Speaker 1: I think what we said was, you know, we brought

37
00:02:21,676 --> 00:02:23,396
Speaker 1: a bunch of machine learning, a bunch of other stuff

38
00:02:23,396 --> 00:02:25,836
Speaker 1: to and said, you know, the data, this massive amount

39
00:02:25,876 --> 00:02:28,196
Speaker 1: of coordinate information. No one can work with it. Coaches

40
00:02:28,196 --> 00:02:30,556
Speaker 1: can't work with it. The leagues that have a tough

41
00:02:30,596 --> 00:02:32,196
Speaker 1: time working with the media has a tough time worker

42
00:02:32,236 --> 00:02:34,556
Speaker 1: that we are going to sort of grind that grain

43
00:02:34,676 --> 00:02:38,436
Speaker 1: into story elements that people can actually use. So we

44
00:02:38,516 --> 00:02:41,436
Speaker 1: turn it into oh, that's a pass, or that's a shot,

45
00:02:41,556 --> 00:02:43,236
Speaker 1: or that's a pick and roll, or that's a between

46
00:02:43,236 --> 00:02:45,156
Speaker 1: the lines pass, or that's a blitz, and so we

47
00:02:45,196 --> 00:02:48,756
Speaker 1: start turning them into words by which people can tell stories.

48
00:02:48,756 --> 00:02:50,476
Speaker 1: So I would say, like the grain is not useful,

49
00:02:50,516 --> 00:02:52,676
Speaker 1: you don't want to make like bread and donuts. Is

50
00:02:52,716 --> 00:02:56,356
Speaker 1: there some particular example you can give me from the

51
00:02:56,436 --> 00:03:00,116
Speaker 1: early days of the company of basically of a problem

52
00:03:00,116 --> 00:03:01,756
Speaker 1: you solved, of a thing you set out to do.

53
00:03:01,996 --> 00:03:03,516
Speaker 1: Maybe it was hard to do, maybe it didn't work

54
00:03:03,556 --> 00:03:04,916
Speaker 1: the way you thought it would, but in the end

55
00:03:04,956 --> 00:03:08,956
Speaker 1: you made some useful thing for coaches, teams, whatever. There

56
00:03:09,156 --> 00:03:10,676
Speaker 1: are two things that we did that sort of a

57
00:03:10,716 --> 00:03:13,676
Speaker 1: big we're a big game changer, right. So one was

58
00:03:13,756 --> 00:03:17,556
Speaker 1: the idea that people were taking shots and you couldn't

59
00:03:17,556 --> 00:03:19,116
Speaker 1: tell if they were a good shot or a bad shot.

60
00:03:19,156 --> 00:03:20,556
Speaker 1: You would use some rules. It's like, oh, if you

61
00:03:20,556 --> 00:03:23,476
Speaker 1: took it here, you know, if people weren't guarding you

62
00:03:23,556 --> 00:03:25,276
Speaker 1: too closely, it was good. But you know, if you

63
00:03:25,436 --> 00:03:27,356
Speaker 1: dribbled a lot, and you jumped in somebody's face. It

64
00:03:27,396 --> 00:03:29,516
Speaker 1: was bad. I think we were able to use math

65
00:03:29,556 --> 00:03:32,156
Speaker 1: and say, oh, that shot should go in forty two

66
00:03:32,236 --> 00:03:34,196
Speaker 1: percent of the time. That shot goes in forty nine

67
00:03:34,196 --> 00:03:36,116
Speaker 1: percent of the time for an average player, it goes

68
00:03:36,156 --> 00:03:38,396
Speaker 1: in for this player. So we were able to quantify

69
00:03:38,476 --> 00:03:41,476
Speaker 1: the quality of a shot, which was basically the core

70
00:03:41,596 --> 00:03:43,916
Speaker 1: thing in basketball is you want to get high quality

71
00:03:43,956 --> 00:03:46,236
Speaker 1: shots and prevent high quality shots, and there was no

72
00:03:46,276 --> 00:03:48,676
Speaker 1: way of measuring it. And we basically said, like, no, no,

73
00:03:48,676 --> 00:03:50,996
Speaker 1: we know exactly what the quality of the shot is,

74
00:03:51,196 --> 00:03:53,756
Speaker 1: and then we also know which players add value beyond

75
00:03:53,836 --> 00:03:57,236
Speaker 1: that shot. So players like Steph Curry and Kevin Durant

76
00:03:57,276 --> 00:03:59,196
Speaker 1: can take a shot that it might be a forty

77
00:03:59,196 --> 00:04:00,756
Speaker 1: five for an average player, and for them it goes

78
00:04:00,756 --> 00:04:03,076
Speaker 1: in fifty five. And that was a big deal. That

79
00:04:03,156 --> 00:04:05,956
Speaker 1: was the first big thing, just to step through that

80
00:04:06,076 --> 00:04:09,516
Speaker 1: a little bit more slowly. So before you came along,

81
00:04:10,236 --> 00:04:13,396
Speaker 1: what did people know? What did coaches know about sort

82
00:04:13,396 --> 00:04:17,996
Speaker 1: of percentage probabilities that a shot will go in? They

83
00:04:17,996 --> 00:04:21,116
Speaker 1: would just track it much more coarsely, so they would say, oh,

84
00:04:21,276 --> 00:04:24,396
Speaker 1: from the corner three, or you would shoot this, and

85
00:04:24,476 --> 00:04:26,356
Speaker 1: from above the break you would shoot this, or in

86
00:04:26,396 --> 00:04:28,396
Speaker 1: the lane you would shoot this. But they didn't know

87
00:04:28,476 --> 00:04:32,196
Speaker 1: things like, hey, there's a really tall defender running straight

88
00:04:32,236 --> 00:04:34,716
Speaker 1: at you, or you're moving sideways. You know, and these

89
00:04:34,756 --> 00:04:38,556
Speaker 1: things also matter important. Yeah, and then the raw data

90
00:04:38,596 --> 00:04:40,476
Speaker 1: allowed us to know all that, and then what we

91
00:04:40,516 --> 00:04:44,116
Speaker 1: did was start to build models that basically predicted the

92
00:04:44,116 --> 00:04:46,716
Speaker 1: likelihood of various shots going in. So one of the

93
00:04:46,756 --> 00:04:49,556
Speaker 1: things you did is coming up with a better metric

94
00:04:49,716 --> 00:04:53,476
Speaker 1: for shot quality that has more inputs. The other thing

95
00:04:53,516 --> 00:04:56,356
Speaker 1: you did, is it right, was around the pick and roll.

96
00:04:56,596 --> 00:04:58,796
Speaker 1: So I think there are all these events in basketball,

97
00:04:58,836 --> 00:05:01,076
Speaker 1: and you can get pretty esoteric. There's a pick and roll,

98
00:05:01,156 --> 00:05:03,916
Speaker 1: but there's many types of pick and rolls. There's many

99
00:05:03,916 --> 00:05:06,076
Speaker 1: types of ways to defend pick and rolls. Just to

100
00:05:06,116 --> 00:05:07,636
Speaker 1: be clear, let me let me just say we won't

101
00:05:07,636 --> 00:05:10,436
Speaker 1: go into detail, but pick and roll it's just a play.

102
00:05:10,516 --> 00:05:13,436
Speaker 1: It's an offensive. It's a play. Yeah, it's somewhat complicated.

103
00:05:13,436 --> 00:05:16,236
Speaker 1: It involves two people on offense and two people on defense.

104
00:05:16,316 --> 00:05:18,916
Speaker 1: Let's just say that's that's all you need to know. Okay,

105
00:05:18,956 --> 00:05:20,996
Speaker 1: So what did you do with the pick and roll?

106
00:05:21,516 --> 00:05:23,436
Speaker 1: So I think the first thing we did was we

107
00:05:23,436 --> 00:05:26,876
Speaker 1: had a machine be able to identify it by staring

108
00:05:26,916 --> 00:05:30,316
Speaker 1: at the data. So before us, it was basically humans

109
00:05:30,356 --> 00:05:34,036
Speaker 1: would watch the video and just start counting how many

110
00:05:34,076 --> 00:05:36,716
Speaker 1: happened and counting what the type of defense was, and

111
00:05:36,756 --> 00:05:39,876
Speaker 1: basically they would just collate all that information. But I

112
00:05:39,876 --> 00:05:42,676
Speaker 1: think the big thing that we discovered is when machines

113
00:05:42,756 --> 00:05:45,356
Speaker 1: did it. The big example was there was a year

114
00:05:45,476 --> 00:05:48,796
Speaker 1: where the human collection said Chris Paul led the league

115
00:05:48,836 --> 00:05:51,316
Speaker 1: running eight hundred pick and rolls, and we said, well,

116
00:05:51,396 --> 00:05:54,756
Speaker 1: Chris Paul led the league running four thousand pick and rolls,

117
00:05:55,036 --> 00:05:58,156
Speaker 1: and one of these is not right. But then we

118
00:05:58,156 --> 00:06:01,036
Speaker 1: could just say like, well, here's our four thousand, and

119
00:06:01,076 --> 00:06:03,516
Speaker 1: then you're saying, oh, okay, you're missing sort of you know,

120
00:06:03,556 --> 00:06:06,236
Speaker 1: eighty percent of them. And so I think then I

121
00:06:06,276 --> 00:06:08,876
Speaker 1: think once we did that, we could automate a lot

122
00:06:08,956 --> 00:06:12,356
Speaker 1: more complicated things, like having the machine understand all the

123
00:06:12,396 --> 00:06:14,676
Speaker 1: defenses that you could play against various types of pick

124
00:06:14,716 --> 00:06:17,236
Speaker 1: and rolls, and then once we solve those, we sort

125
00:06:17,236 --> 00:06:20,156
Speaker 1: of built this engine that could generate words accurately, and

126
00:06:20,196 --> 00:06:22,836
Speaker 1: then we started growing teams and each one of them

127
00:06:22,836 --> 00:06:24,996
Speaker 1: would say, hey, can you do these words can you

128
00:06:25,036 --> 00:06:26,956
Speaker 1: do these words? Can you do these words? Like say,

129
00:06:26,956 --> 00:06:28,956
Speaker 1: some of the words they were asking for. So they're

130
00:06:28,996 --> 00:06:31,756
Speaker 1: sort of like you know, flares and loops and zippers

131
00:06:31,756 --> 00:06:34,916
Speaker 1: and jams and trails and whips and you know veer backs,

132
00:06:34,956 --> 00:06:38,036
Speaker 1: and I mean the sort these things are basically what

133
00:06:38,636 --> 00:06:41,836
Speaker 1: plays their their defenses, their moves. That's right. So there's

134
00:06:41,836 --> 00:06:44,716
Speaker 1: like sort of you know, the particular kind of screen

135
00:06:44,796 --> 00:06:47,356
Speaker 1: that they used to get Steph Curry free off in

136
00:06:47,436 --> 00:06:50,196
Speaker 1: the side versus the name of a defense that the

137
00:06:50,316 --> 00:06:53,316
Speaker 1: person running after him might choose to play while he's

138
00:06:53,356 --> 00:06:56,116
Speaker 1: doing that thing, right, so to avoid that screen. Yeah,

139
00:06:57,156 --> 00:06:59,556
Speaker 1: and when they say learn a work, they really mean

140
00:06:59,796 --> 00:07:03,516
Speaker 1: can you teach your machine or get your machine to

141
00:07:03,756 --> 00:07:08,916
Speaker 1: learn to recognize this play this thing? Because they want

142
00:07:08,916 --> 00:07:12,356
Speaker 1: to know every time they do this two questions. Every

143
00:07:12,396 --> 00:07:14,516
Speaker 1: time they do this and we do this, how efficient

144
00:07:14,636 --> 00:07:16,916
Speaker 1: is it? And or every time they do this and

145
00:07:16,956 --> 00:07:19,716
Speaker 1: we do this, show me every time that that a

146
00:07:19,836 --> 00:07:22,116
Speaker 1: video of every time that happened over the last several years.

147
00:07:22,156 --> 00:07:24,036
Speaker 1: Those are the questions they tend to ask. Okay, and

148
00:07:24,116 --> 00:07:26,156
Speaker 1: so it had it has a bunch of quirks, you know,

149
00:07:26,236 --> 00:07:28,836
Speaker 1: working with it, But but it was not easy. But

150
00:07:28,916 --> 00:07:31,316
Speaker 1: we've done it, and now we've created sort of five

151
00:07:31,396 --> 00:07:34,356
Speaker 1: hundred of these words that you know, coaches and you

152
00:07:34,396 --> 00:07:39,156
Speaker 1: know players use on a daily basis. And so I

153
00:07:39,196 --> 00:07:41,316
Speaker 1: can think of two parts of this that would have

154
00:07:41,316 --> 00:07:44,956
Speaker 1: been sort of hard problems to solve, one being the

155
00:07:45,116 --> 00:07:48,996
Speaker 1: kind of nathy computer part, like building the machine, and

156
00:07:49,036 --> 00:07:53,036
Speaker 1: the other part being the getting people to believe you

157
00:07:52,796 --> 00:07:59,156
Speaker 1: and use your insu So tell me a hard thing?

158
00:07:59,516 --> 00:08:01,516
Speaker 1: Well from each right, So what's a hard thing from

159
00:08:01,556 --> 00:08:05,196
Speaker 1: the from the MATTHI computer part? I think the mat

160
00:08:05,356 --> 00:08:07,356
Speaker 1: the mat side is just that you don't have a

161
00:08:07,436 --> 00:08:08,916
Speaker 1: lot of data. So it's for example, a lot of

162
00:08:08,956 --> 00:08:11,316
Speaker 1: people who do machine learning will say like, oh, give

163
00:08:11,356 --> 00:08:13,676
Speaker 1: me millions of examples of a thing and I will

164
00:08:13,756 --> 00:08:16,036
Speaker 1: learn it. Where well, there aren't millions of pick and

165
00:08:16,156 --> 00:08:17,796
Speaker 1: rolls that you're going to get this sort of how

166
00:08:17,796 --> 00:08:20,476
Speaker 1: do you solve that? Right? Small sample size? Classic problem,

167
00:08:20,596 --> 00:08:24,036
Speaker 1: small sample size, So you know, how do you solve it? Well,

168
00:08:24,356 --> 00:08:27,676
Speaker 1: there's some mathematical chicanery that we invented. The answer to

169
00:08:27,876 --> 00:08:30,316
Speaker 1: be very clever, like we're really good at bath. Is

170
00:08:30,356 --> 00:08:33,116
Speaker 1: that the unsatisig outs? Yeah, let me let me ask you.

171
00:08:33,196 --> 00:08:36,276
Speaker 1: This was there some like things you tried that didn't work.

172
00:08:36,316 --> 00:08:38,076
Speaker 1: It was just like one equation didn't work and then

173
00:08:38,076 --> 00:08:40,116
Speaker 1: another did. I mean, yeah, yeah, yeah, I know. I

174
00:08:40,156 --> 00:08:42,556
Speaker 1: think we've we we you know, I think we we

175
00:08:42,556 --> 00:08:45,356
Speaker 1: we said over the course of our life, we've used

176
00:08:45,396 --> 00:08:47,756
Speaker 1: everything in the AI textbooks to try and solve the problem,

177
00:08:47,756 --> 00:08:49,556
Speaker 1: and they've evolved over years to try and get them

178
00:08:49,596 --> 00:08:52,396
Speaker 1: get to be better. Um. But you know, I think

179
00:08:52,396 --> 00:08:54,996
Speaker 1: that that's part of the adventure of sort of trying

180
00:08:55,036 --> 00:08:57,196
Speaker 1: to figure that stuff out. It wasn't like we used

181
00:08:57,476 --> 00:09:00,196
Speaker 1: we used method X and the answer popped out, and

182
00:09:00,236 --> 00:09:02,436
Speaker 1: so we've been constantly evolving the ability to do But

183
00:09:02,716 --> 00:09:05,636
Speaker 1: I think I think part of the I mean, whatever

184
00:09:05,676 --> 00:09:07,436
Speaker 1: the secret is, you have to leverage the structure of

185
00:09:07,436 --> 00:09:08,836
Speaker 1: the problem. You have to leverage the fact that you

186
00:09:08,876 --> 00:09:10,956
Speaker 1: know something about the fact that it's a bunch of

187
00:09:10,956 --> 00:09:13,196
Speaker 1: people playing a particular sport. When you don't have enough

188
00:09:13,276 --> 00:09:15,676
Speaker 1: when you don't have enough data, you have to leverage

189
00:09:15,676 --> 00:09:18,636
Speaker 1: structure in some way. Yeah, so you're you're able to

190
00:09:19,356 --> 00:09:22,916
Speaker 1: because it's such a constrained environment, because you know the rules,

191
00:09:22,956 --> 00:09:25,836
Speaker 1: because you know the things you wanted to find. You

192
00:09:25,876 --> 00:09:27,556
Speaker 1: can sort of say this is a pick and roll

193
00:09:27,596 --> 00:09:28,716
Speaker 1: this is a pick and roll. This is a pick

194
00:09:28,716 --> 00:09:29,996
Speaker 1: and roll. This is not a pick and roll. This

195
00:09:30,036 --> 00:09:33,556
Speaker 1: is not a pick and roll that kind of thing. Yeah, good,

196
00:09:33,596 --> 00:09:36,596
Speaker 1: So okay, So that's the math, sort of technical side.

197
00:09:36,676 --> 00:09:39,396
Speaker 1: Then there's the getting people to believe you and you know,

198
00:09:39,676 --> 00:09:43,356
Speaker 1: buy what you're selling ultimately, both metaphorically and literally. Tell

199
00:09:43,396 --> 00:09:47,436
Speaker 1: me about that side, like, were there any interesting I

200
00:09:47,476 --> 00:09:50,036
Speaker 1: don't know, people who didn't want to buy it, you know,

201
00:09:50,116 --> 00:09:53,716
Speaker 1: little stories from that side. Almost every coach that we've

202
00:09:53,756 --> 00:09:55,796
Speaker 1: talked to couldn't believe that we did what we believed.

203
00:09:55,836 --> 00:10:01,036
Speaker 1: So the story is generally involved just sitting there in

204
00:10:01,076 --> 00:10:03,436
Speaker 1: front of these coaches. You go up a couple of levels.

205
00:10:03,796 --> 00:10:07,196
Speaker 1: So I think that you know, this happened with two coaches.

206
00:10:07,236 --> 00:10:10,716
Speaker 1: It's a very similar story. They were very well known coaches,

207
00:10:11,396 --> 00:10:14,436
Speaker 1: you know, long resumes, have been around the NBA for

208
00:10:14,476 --> 00:10:17,676
Speaker 1: a long time, won lots of championships. You know, it

209
00:10:17,676 --> 00:10:20,156
Speaker 1: took several meetings to even get an audience with them.

210
00:10:20,156 --> 00:10:22,276
Speaker 1: You would go through sort of layer one, than layer two,

211
00:10:22,276 --> 00:10:23,796
Speaker 1: then layer three, and then you would get to them.

212
00:10:24,036 --> 00:10:27,196
Speaker 1: It's like knitting the Pope or something exactly, and then

213
00:10:27,196 --> 00:10:29,636
Speaker 1: you would sort of get to them and there they were.

214
00:10:29,876 --> 00:10:32,196
Speaker 1: They would just basically say it like, you have no

215
00:10:32,316 --> 00:10:34,476
Speaker 1: idea what I intend for the players to do. There's

216
00:10:34,516 --> 00:10:37,956
Speaker 1: no way this machine can understand what what you need.

217
00:10:37,996 --> 00:10:40,996
Speaker 1: And we would say try us, and we would say, okay,

218
00:10:41,036 --> 00:10:43,756
Speaker 1: show me all the times that the play started with

219
00:10:43,756 --> 00:10:45,516
Speaker 1: the pick and roll and there was three passes and

220
00:10:45,556 --> 00:10:47,316
Speaker 1: I took a shot in the corners like okay, here

221
00:10:47,316 --> 00:10:48,916
Speaker 1: it is like, show me all the times there was

222
00:10:48,956 --> 00:10:51,676
Speaker 1: a screen. I mean like you have a laptop and

223
00:10:51,716 --> 00:10:54,636
Speaker 1: you have your software running or something. Yep, that's right,

224
00:10:54,676 --> 00:10:56,356
Speaker 1: and they would we would have tables and charts and

225
00:10:56,396 --> 00:10:58,476
Speaker 1: they would we would project onto a screen and they

226
00:10:58,516 --> 00:11:01,596
Speaker 1: would basically just give us the quiz, the sort of

227
00:11:01,596 --> 00:11:03,956
Speaker 1: the the They would put us through the ringer for

228
00:11:03,956 --> 00:11:06,276
Speaker 1: for like two hours, so it's like a it's basically

229
00:11:06,316 --> 00:11:10,796
Speaker 1: a impromptu grilling from the coach and make your machine

230
00:11:10,876 --> 00:11:13,276
Speaker 1: dance for me and show me the correct answers. And

231
00:11:13,356 --> 00:11:14,756
Speaker 1: any time you come up with an answer that I

232
00:11:14,756 --> 00:11:18,196
Speaker 1: don't agree with, you know, I wouldn't think you're an idiot.

233
00:11:18,196 --> 00:11:20,556
Speaker 1: But we held up. In fact, sometimes these coaches would say, wait,

234
00:11:20,596 --> 00:11:22,836
Speaker 1: show me this list, and if these two guys aren't

235
00:11:22,916 --> 00:11:25,236
Speaker 1: number one, and number two, I think your thing is wrong,

236
00:11:25,276 --> 00:11:27,076
Speaker 1: and then we would do it, and like thank goodness,

237
00:11:27,276 --> 00:11:29,156
Speaker 1: the right players would end up number one and number

238
00:11:29,196 --> 00:11:31,876
Speaker 1: two and and then normally after sort of two hours

239
00:11:32,076 --> 00:11:35,636
Speaker 1: of grilling, they would be pretty good and they would

240
00:11:35,636 --> 00:11:37,036
Speaker 1: just leave the room, and then we knew we would

241
00:11:37,036 --> 00:11:45,876
Speaker 1: have a contract. And is the output like the thing

242
00:11:45,956 --> 00:11:48,396
Speaker 1: that actually you see on the screen when you're running

243
00:11:48,396 --> 00:11:51,276
Speaker 1: your software. Is it lists? Is it little videos where

244
00:11:51,276 --> 00:11:53,596
Speaker 1: all the players are dots? Is it both of those things?

245
00:11:53,916 --> 00:11:56,476
Speaker 1: It's everything. So we have sort of ranking stables that

246
00:11:56,516 --> 00:11:58,316
Speaker 1: can answer who's the best at X, Y and Z.

247
00:11:58,476 --> 00:12:01,796
Speaker 1: We have sort of a variety of visualizations that show

248
00:12:02,596 --> 00:12:05,836
Speaker 1: breakdowns of various actions. You can always click on anything

249
00:12:05,916 --> 00:12:08,356
Speaker 1: and show all the video of every moment you know

250
00:12:08,516 --> 00:12:11,196
Speaker 1: anything that you asked it. So we've built lots of

251
00:12:11,236 --> 00:12:14,516
Speaker 1: visualizations and data formats to make it easy for various

252
00:12:14,556 --> 00:12:17,196
Speaker 1: people in the organizations to use them. There are instances

253
00:12:17,196 --> 00:12:19,956
Speaker 1: in the world in different domains where like a data

254
00:12:20,036 --> 00:12:23,156
Speaker 1: driven approach suggests doing one thing, but that thing is

255
00:12:23,276 --> 00:12:26,756
Speaker 1: contrary to conventional wisdom, right, But like in football, it

256
00:12:26,756 --> 00:12:30,876
Speaker 1: seems pretty clear that the data suggests coaches should go

257
00:12:30,916 --> 00:12:32,756
Speaker 1: for it on fourth down more often than they do,

258
00:12:32,876 --> 00:12:35,116
Speaker 1: and it seems like as the result, coaches have started

259
00:12:35,156 --> 00:12:36,996
Speaker 1: going for it more on fourth down. But there is

260
00:12:37,036 --> 00:12:38,836
Speaker 1: this thing where if a coach goes for it on

261
00:12:38,876 --> 00:12:41,796
Speaker 1: fourth down and doesn't make it, the team doesn't get

262
00:12:41,836 --> 00:12:45,076
Speaker 1: the first down, then the coach gets pillaried. Right. So

263
00:12:45,556 --> 00:12:48,396
Speaker 1: even though the data says it makes sense to go

264
00:12:48,516 --> 00:12:51,436
Speaker 1: for it in a kind of a personal incentives way,

265
00:12:51,556 --> 00:12:53,996
Speaker 1: it might not make sense right in a social way,

266
00:12:54,036 --> 00:12:56,476
Speaker 1: which is real and important, it might not make sense.

267
00:12:57,196 --> 00:12:59,396
Speaker 1: Is there a basketball version of that? Is there a

268
00:12:59,476 --> 00:13:02,156
Speaker 1: version of that you have observed where your data suggests

269
00:13:02,156 --> 00:13:04,436
Speaker 1: that coaches should be doing something, but they're reluctant to

270
00:13:04,476 --> 00:13:08,636
Speaker 1: do it because it's contrary to sort of conventional wisdom. No,

271
00:13:08,716 --> 00:13:10,596
Speaker 1: I think there's a lot more nuance and what we do.

272
00:13:10,636 --> 00:13:14,596
Speaker 1: I think there's the public basically is aware that they say, oh,

273
00:13:14,596 --> 00:13:16,956
Speaker 1: people should take more threes. But I think the reality

274
00:13:17,076 --> 00:13:19,476
Speaker 1: is much more complex than that, because I think it's

275
00:13:19,556 --> 00:13:22,116
Speaker 1: not about just simply take a three. What you want

276
00:13:22,116 --> 00:13:24,476
Speaker 1: to do is get the best possible shot for yourself.

277
00:13:24,516 --> 00:13:27,116
Speaker 1: Sometimes that's a three, sometimes that's something else. Different kinds

278
00:13:27,156 --> 00:13:29,476
Speaker 1: of threes are not the same, and your talent is

279
00:13:29,476 --> 00:13:30,916
Speaker 1: not the same. In fact, if you look at all

280
00:13:30,956 --> 00:13:33,196
Speaker 1: the you know that you know, I'm just gonna sort

281
00:13:33,236 --> 00:13:35,676
Speaker 1: of Lebron James and Kevin Durant and Steph Curry and

282
00:13:35,756 --> 00:13:38,156
Speaker 1: Janis and Yo Kitchen, all these players, and you know

283
00:13:38,516 --> 00:13:41,876
Speaker 1: they are very different players, and teams are built around

284
00:13:41,916 --> 00:13:44,116
Speaker 1: these players, and what you want to do is optimize

285
00:13:44,156 --> 00:13:46,596
Speaker 1: the team built around these very unique players. And so

286
00:13:46,956 --> 00:13:48,556
Speaker 1: a lot of it is in the details of how

287
00:13:48,596 --> 00:13:51,356
Speaker 1: you structure your offense and your defense to sort of

288
00:13:51,476 --> 00:13:54,116
Speaker 1: use them to get your team the best possible shot.

289
00:13:54,316 --> 00:13:55,916
Speaker 1: So a lot of it is that these coaches do

290
00:13:55,956 --> 00:13:58,636
Speaker 1: a lot of work on basically on the micro level

291
00:13:58,676 --> 00:14:00,876
Speaker 1: to sort of create these plays. That it's not just

292
00:14:00,876 --> 00:14:03,596
Speaker 1: sort of walk up the court and take a three.

293
00:14:03,676 --> 00:14:05,076
Speaker 1: They do a lot of work to try and put

294
00:14:05,116 --> 00:14:07,996
Speaker 1: their position their players in positions to have a lot

295
00:14:08,036 --> 00:14:10,716
Speaker 1: of good options to get the best possible shot. So

296
00:14:10,916 --> 00:14:13,316
Speaker 1: I think that it's it's really a lot more nuanced

297
00:14:13,356 --> 00:14:21,196
Speaker 1: in the actual execution of it. Fans would be really

298
00:14:21,236 --> 00:14:26,996
Speaker 1: surprised at the level of sophistication entirely across the league

299
00:14:27,036 --> 00:14:29,676
Speaker 1: and many leagues in sports, especially the top ones. There's

300
00:14:29,716 --> 00:14:32,236
Speaker 1: there's a degree of sophistication and how they use data

301
00:14:32,276 --> 00:14:37,796
Speaker 1: and video that I think would surprise almost everyone. You

302
00:14:37,916 --> 00:14:41,036
Speaker 1: think that the sort of play calling side of coaching

303
00:14:41,076 --> 00:14:43,796
Speaker 1: will become more and more delegated to the machine. I mean,

304
00:14:43,836 --> 00:14:49,116
Speaker 1: I could see coaches as essentially psychologists, being persistently useful.

305
00:14:49,156 --> 00:14:51,796
Speaker 1: But do you see a time when the sort of

306
00:14:51,836 --> 00:14:55,796
Speaker 1: core strategic decision making will just be be done better

307
00:14:55,796 --> 00:14:58,156
Speaker 1: by a machine than by coach. I don't. I don't

308
00:14:58,156 --> 00:15:01,556
Speaker 1: think so. Our thesis has always been we build iron

309
00:15:01,556 --> 00:15:04,996
Speaker 1: Man suits and and different people will want iron you know,

310
00:15:05,116 --> 00:15:07,556
Speaker 1: different iron Man suits, and you will want because you

311
00:15:07,596 --> 00:15:09,276
Speaker 1: want you need somebody in the middle of the iron

312
00:15:09,356 --> 00:15:12,196
Speaker 1: Man suit directing it, using everything they know about the world.

313
00:15:12,356 --> 00:15:13,956
Speaker 1: But you want to be a lot more powerful. And

314
00:15:14,036 --> 00:15:16,676
Speaker 1: I think that that's the model we've always used, is

315
00:15:17,036 --> 00:15:19,356
Speaker 1: we build Iron Man suits for everybody, and then different

316
00:15:19,396 --> 00:15:21,356
Speaker 1: coaches and different assistant coaches are going to use them,

317
00:15:21,636 --> 00:15:23,596
Speaker 1: you know, to be a lot more powerful. That's what

318
00:15:23,636 --> 00:15:25,156
Speaker 1: we want to do. We want to build Iron Man

319
00:15:25,196 --> 00:15:30,076
Speaker 1: suits for everybody. Does basketball look different? Is it played

320
00:15:30,116 --> 00:15:35,636
Speaker 1: differently because of your work? I certainly know that there

321
00:15:35,676 --> 00:15:37,876
Speaker 1: have been examples where you know, in a couple of

322
00:15:37,996 --> 00:15:41,716
Speaker 1: NBA Finals strategies were changed because of the data that

323
00:15:41,796 --> 00:15:45,036
Speaker 1: we provided. You know, close ones are where one team

324
00:15:45,076 --> 00:15:48,596
Speaker 1: basically discovered a pick and roll defense strategy that was

325
00:15:48,636 --> 00:15:51,996
Speaker 1: effective and employed that and it to you know, very

326
00:15:51,996 --> 00:15:53,796
Speaker 1: good effect. Another one it was one of these sort

327
00:15:53,796 --> 00:15:57,756
Speaker 1: of off ball screen defense strategies where they found the

328
00:15:57,836 --> 00:16:00,476
Speaker 1: data show that there were particular strategies because that would

329
00:16:00,476 --> 00:16:02,316
Speaker 1: shave off a couple of points of efficiency from the

330
00:16:02,316 --> 00:16:04,956
Speaker 1: other team, and they employed that very heavily and they

331
00:16:04,996 --> 00:16:06,996
Speaker 1: won a very close series. So I think, can you

332
00:16:07,036 --> 00:16:10,956
Speaker 1: say what teams it was? I should Okay, I get

333
00:16:10,996 --> 00:16:13,996
Speaker 1: people looking at they could probably figure it out. The

334
00:16:14,076 --> 00:16:15,476
Speaker 1: point that I want to realize, like there are a

335
00:16:15,476 --> 00:16:16,916
Speaker 1: lot of people who come out and think like, oh,

336
00:16:16,956 --> 00:16:19,676
Speaker 1: these these teams are sort of the geniuses and these

337
00:16:19,676 --> 00:16:21,916
Speaker 1: teams are the Luddites, and that's not the case. Like

338
00:16:22,196 --> 00:16:25,196
Speaker 1: almost every team is. Like people would be surprised that

339
00:16:25,396 --> 00:16:27,556
Speaker 1: at the minimum, that's fine. I mean, everybody in the

340
00:16:27,676 --> 00:16:30,596
Speaker 1: NBA is every player in the NBA is good at

341
00:16:30,636 --> 00:16:33,596
Speaker 1: playing basketball, but some are better than others. And so similarly,

342
00:16:33,596 --> 00:16:36,276
Speaker 1: you might imagine that every sort of quant team in

343
00:16:36,316 --> 00:16:38,956
Speaker 1: the NBA is good at at being you know, doing

344
00:16:39,036 --> 00:16:42,196
Speaker 1: quantitative analysis, but some are probably better than others. So

345
00:16:42,236 --> 00:16:43,716
Speaker 1: I would say, you're right, there are some who are

346
00:16:43,756 --> 00:16:45,716
Speaker 1: better than others, but the minimum level is a lot

347
00:16:45,796 --> 00:16:48,876
Speaker 1: higher than when everyone would sure, yeah, yeah, And can

348
00:16:48,916 --> 00:16:53,276
Speaker 1: you say who's better than others? I cannot. I I

349
00:16:53,476 --> 00:16:56,236
Speaker 1: this is our lifeblood of being the reason that all

350
00:16:56,276 --> 00:16:59,476
Speaker 1: these coaches trust us and is that we don't. We

351
00:16:59,476 --> 00:17:02,156
Speaker 1: don't leak. So I'm sorry. It would be fun one day,

352
00:17:02,276 --> 00:17:04,516
Speaker 1: One day, I'll call all the stories. It's not okay,

353
00:17:04,516 --> 00:17:10,476
Speaker 1: I'm sorry. Regiev is not going to name names today Alas,

354
00:17:10,916 --> 00:17:13,236
Speaker 1: but he is going to talk about a big, interesting

355
00:17:13,276 --> 00:17:16,436
Speaker 1: problem that second spectrum is close to solving but has

356
00:17:16,516 --> 00:17:27,356
Speaker 1: not quite nailed yet. That's in a minute. Now back

357
00:17:27,356 --> 00:17:30,876
Speaker 1: to the show. I want to talk about things you

358
00:17:30,916 --> 00:17:33,676
Speaker 1: haven't figured out yet, like what are the next problems

359
00:17:33,676 --> 00:17:36,156
Speaker 1: you're trying to solve, Like what is the frontier? So

360
00:17:36,236 --> 00:17:38,836
Speaker 1: I think the frontier where we're trying to basically basically

361
00:17:38,876 --> 00:17:40,836
Speaker 1: get full body pose of the human, not just a

362
00:17:40,876 --> 00:17:43,436
Speaker 1: little dot. We're trying to get their entire skeleton. So

363
00:17:43,516 --> 00:17:46,236
Speaker 1: the basic sort of product you have, the basic thing

364
00:17:46,236 --> 00:17:48,716
Speaker 1: you do turns each player into a dot, just like

365
00:17:48,756 --> 00:17:51,156
Speaker 1: in the classic like x's and o's diagram or whatever.

366
00:17:51,236 --> 00:17:52,876
Speaker 1: And so you can watch little dots moving around and

367
00:17:52,876 --> 00:17:54,636
Speaker 1: you can see how far the dots are from the other.

368
00:17:55,076 --> 00:17:57,476
Speaker 1: But there's a lot going on that you don't see

369
00:17:57,476 --> 00:17:59,196
Speaker 1: in the dot, right you might want to know, like

370
00:17:59,276 --> 00:18:01,156
Speaker 1: what are the things that are important that you don't

371
00:18:01,156 --> 00:18:04,396
Speaker 1: see in the dot? That's exactly right. So the challenge

372
00:18:04,396 --> 00:18:06,396
Speaker 1: we're working out right now is turning the dot into

373
00:18:06,436 --> 00:18:10,036
Speaker 1: a human skeleton and then having that skeleton and generate

374
00:18:10,116 --> 00:18:12,596
Speaker 1: data in you know, one hundred to two hundred milliseconds.

375
00:18:12,756 --> 00:18:14,436
Speaker 1: So that's the challenge that we're working on right now,

376
00:18:14,436 --> 00:18:16,356
Speaker 1: and we've actually made a lot of progress on that,

377
00:18:16,436 --> 00:18:19,236
Speaker 1: and that's right now. And so what's an example of

378
00:18:19,516 --> 00:18:21,236
Speaker 1: why that would be useful? What are some of the

379
00:18:21,276 --> 00:18:24,716
Speaker 1: ways you could learn from seeing each player as a

380
00:18:24,756 --> 00:18:28,036
Speaker 1: whole body rather than as a dot. So I think

381
00:18:28,036 --> 00:18:29,316
Speaker 1: there are a lot of things you can do. There's

382
00:18:29,356 --> 00:18:32,836
Speaker 1: one is, obviously, you know, with health and fitness, where

383
00:18:32,836 --> 00:18:34,716
Speaker 1: you can figure out did they land on one foot,

384
00:18:34,716 --> 00:18:36,356
Speaker 1: did they land on two feet? You know, did they

385
00:18:36,436 --> 00:18:38,316
Speaker 1: you know they take off from this port of that foot.

386
00:18:38,916 --> 00:18:40,316
Speaker 1: So there's a lot of health and fitness stuff that

387
00:18:40,316 --> 00:18:43,636
Speaker 1: you can figure out health and fitness, meaning learning to

388
00:18:44,436 --> 00:18:47,636
Speaker 1: reduce the risk of injury, learning that certain kinds of

389
00:18:47,876 --> 00:18:52,476
Speaker 1: moves or subtle distinctions in moves make a player more

390
00:18:52,636 --> 00:18:54,956
Speaker 1: or less likely to be injured. That's right, And so

391
00:18:54,956 --> 00:18:57,156
Speaker 1: that's one class of things we can do. Another class

392
00:18:57,196 --> 00:18:59,716
Speaker 1: is sort of just understanding moves better. So if a

393
00:18:59,796 --> 00:19:03,636
Speaker 1: player made, you know, dribble moves between their legs and

394
00:19:03,676 --> 00:19:06,636
Speaker 1: crossovers and step backs and did all all kinds of

395
00:19:06,636 --> 00:19:08,916
Speaker 1: things with their hands and feet to get open, now

396
00:19:08,916 --> 00:19:12,356
Speaker 1: we can categorize those things. There's that, there's health, there's strategy,

397
00:19:12,636 --> 00:19:15,716
Speaker 1: there's media, there's officiating. There's you can help with officiating

398
00:19:15,756 --> 00:19:17,556
Speaker 1: if you start knowing where people are. So there's a

399
00:19:17,596 --> 00:19:19,876
Speaker 1: lot of things that you can do once you understand

400
00:19:19,916 --> 00:19:22,196
Speaker 1: more of the human skeleton. And let me ask you this,

401
00:19:22,996 --> 00:19:25,556
Speaker 1: why is that hard? It sounds hard, but like tell

402
00:19:25,596 --> 00:19:28,676
Speaker 1: me about that being hard. So basically, you have a

403
00:19:28,676 --> 00:19:30,916
Speaker 1: bunch of cameras in a stadium and it sees these

404
00:19:30,916 --> 00:19:34,196
Speaker 1: sort of blotches of that you tell them it's a

405
00:19:34,276 --> 00:19:36,796
Speaker 1: human and and you're to say like, oh, that's a knee,

406
00:19:36,796 --> 00:19:39,156
Speaker 1: and that's an ankle, and that's a that's a waste,

407
00:19:39,156 --> 00:19:41,556
Speaker 1: and so it's just sort of you're you're teaching a

408
00:19:41,596 --> 00:19:44,316
Speaker 1: machine to see a human from scratch, and it's not

409
00:19:44,396 --> 00:19:46,356
Speaker 1: that it's it's it's you know, there's been a lot

410
00:19:46,396 --> 00:19:48,396
Speaker 1: of progress in computer vision over the years that as

411
00:19:48,436 --> 00:19:51,356
Speaker 1: making this making this problem, you know, easier and easier.

412
00:19:51,556 --> 00:19:53,556
Speaker 1: A lot of it is also just doing it extremely

413
00:19:53,636 --> 00:19:55,396
Speaker 1: quickly like you want to. You want this machine to

414
00:19:55,396 --> 00:19:56,396
Speaker 1: be able to do it and on the order one

415
00:19:56,436 --> 00:19:58,876
Speaker 1: hundred two hundred milliseconds, to be able to enable a

416
00:19:58,876 --> 00:20:00,676
Speaker 1: lot of things to happen, and happen in real time

417
00:20:00,716 --> 00:20:03,476
Speaker 1: in sports one hundreds or two hundred milliseconds, but you

418
00:20:03,556 --> 00:20:06,436
Speaker 1: might want it to be only delayed from reality by

419
00:20:06,556 --> 00:20:09,476
Speaker 1: a tenth of a second. Let me ask a dumb question,

420
00:20:10,236 --> 00:20:12,436
Speaker 1: why can't you just let it take its time to

421
00:20:12,476 --> 00:20:14,276
Speaker 1: figure it out. You're not trying to do it in

422
00:20:14,316 --> 00:20:15,836
Speaker 1: the middle of the game, right, or are you trying

423
00:20:15,836 --> 00:20:17,076
Speaker 1: to do it in the middle. So, because I think

424
00:20:17,156 --> 00:20:19,116
Speaker 1: sports is one of these interesting this is part of

425
00:20:19,116 --> 00:20:21,876
Speaker 1: the reason why sports machines. Understanding of sports is so hard.

426
00:20:22,556 --> 00:20:24,396
Speaker 1: In sports, it's not like you want to analyze a

427
00:20:24,516 --> 00:20:27,276
Speaker 1: video and then know what happened a long time later

428
00:20:27,276 --> 00:20:28,956
Speaker 1: that might be the case for coaches, but for a

429
00:20:28,956 --> 00:20:31,716
Speaker 1: lot of media applications or refereeing applications, you want to

430
00:20:31,716 --> 00:20:35,156
Speaker 1: be able to do it basically instantaneously. Yeah, right, And

431
00:20:35,196 --> 00:20:37,076
Speaker 1: so it's almost like the self driving car, like you

432
00:20:37,116 --> 00:20:39,156
Speaker 1: need it to be able to do its thing very

433
00:20:39,276 --> 00:20:41,996
Speaker 1: very quickly to be able to react to it. You know,

434
00:20:42,236 --> 00:20:44,476
Speaker 1: is self driving car research helpful to you? Are you

435
00:20:44,516 --> 00:20:47,876
Speaker 1: borrowing from that? Well? Yes, because I think that's generally

436
00:20:47,916 --> 00:20:51,156
Speaker 1: pushing the field of computer computer vision forward because it's

437
00:20:51,156 --> 00:20:54,556
Speaker 1: basically the same problem. You need to be identify people

438
00:20:54,636 --> 00:20:56,996
Speaker 1: and what they're doing very very quickly, So one is

439
00:20:56,996 --> 00:20:58,556
Speaker 1: to sort of avoid running into them, the other one

440
00:20:58,636 --> 00:21:01,836
Speaker 1: is sort of figuring what's going on on a sporting event.

441
00:21:01,876 --> 00:21:04,956
Speaker 1: And so I think the general trend of computer vision

442
00:21:05,036 --> 00:21:10,436
Speaker 1: research moving to solve these problems quickly is helpful. Have

443
00:21:10,556 --> 00:21:13,396
Speaker 1: you figured out anything yet about injury, Like you have

444
00:21:13,436 --> 00:21:16,676
Speaker 1: any useful injury insights? I mean I've always been, you know,

445
00:21:16,716 --> 00:21:18,876
Speaker 1: to be honest, wary of injury because it's such a

446
00:21:18,956 --> 00:21:20,636
Speaker 1: it's it's a hard thing to do. But I think

447
00:21:20,676 --> 00:21:23,596
Speaker 1: that well we have why wary of studying it, wary

448
00:21:23,596 --> 00:21:27,236
Speaker 1: of trying to to be honest that the teams use

449
00:21:27,316 --> 00:21:29,356
Speaker 1: the data we give them in that way, even though

450
00:21:29,396 --> 00:21:31,756
Speaker 1: I don't want it to happen, because because they just

451
00:21:31,836 --> 00:21:34,796
Speaker 1: it's just it's any information is better than no information.

452
00:21:34,836 --> 00:21:37,276
Speaker 1: And so I've always been, you know, the whole nine years,

453
00:21:37,276 --> 00:21:38,876
Speaker 1: I've said, we'll give you the information, but I'd be

454
00:21:39,076 --> 00:21:41,756
Speaker 1: I'd be careful. But you know, it's such a such

455
00:21:41,756 --> 00:21:44,356
Speaker 1: an important thing about keeping athletes. You know, why are

456
00:21:44,356 --> 00:21:47,156
Speaker 1: you so wary of trying to figure out injury risk?

457
00:21:47,476 --> 00:21:50,116
Speaker 1: I just I don't know that at this point. I

458
00:21:50,156 --> 00:21:54,636
Speaker 1: think that I've seen stuff that is is uh, super

459
00:21:54,636 --> 00:21:56,836
Speaker 1: predictive that the scientists that I would sort of laid

460
00:21:56,876 --> 00:21:57,916
Speaker 1: down on. I mean, I know there are a bunch

461
00:21:57,916 --> 00:22:00,236
Speaker 1: of other companies who have so because it's such a

462
00:22:00,236 --> 00:22:02,196
Speaker 1: hard problem, because you don't feel like you can you

463
00:22:02,236 --> 00:22:05,236
Speaker 1: can reliably predict it. Yet, Yeah, what is it about

464
00:22:05,316 --> 00:22:08,996
Speaker 1: injury risk that makes it so much harder to understand

465
00:22:09,196 --> 00:22:11,636
Speaker 1: than sort of things you know that are more kind

466
00:22:11,636 --> 00:22:13,516
Speaker 1: of within the game itself. I mean, I just think

467
00:22:13,516 --> 00:22:16,436
Speaker 1: that you know, it's it's it's almost a sample sized thing.

468
00:22:16,476 --> 00:22:19,916
Speaker 1: You know, people get injured in all sorts of you know,

469
00:22:20,036 --> 00:22:23,196
Speaker 1: unique ways, you know, not super often, right, and so

470
00:22:23,436 --> 00:22:25,236
Speaker 1: you know, I said, there aren't that many pick and rolls,

471
00:22:25,236 --> 00:22:26,596
Speaker 1: but they're way more pick and rolls than there are

472
00:22:26,716 --> 00:22:30,396
Speaker 1: you know, yeah, whatever ACL tears or whatever these and

473
00:22:30,556 --> 00:22:33,676
Speaker 1: presumably injury is more complex. Yeah, it's a more complex problem,

474
00:22:33,756 --> 00:22:36,436
Speaker 1: and it happens way less often, that's right, which makes

475
00:22:36,436 --> 00:22:38,796
Speaker 1: it harder on both sides. That's right. I want to

476
00:22:38,796 --> 00:22:42,876
Speaker 1: talk a little bit about the applications or potential applications

477
00:22:42,916 --> 00:22:47,596
Speaker 1: of your work beyond basketball, the extensions of your work

478
00:22:47,676 --> 00:22:52,516
Speaker 1: beyond beyond sports. Um. I mean sports and games more

479
00:22:52,556 --> 00:22:58,036
Speaker 1: generally seem interesting as a sort of testing ground. Obviously

480
00:22:58,076 --> 00:23:00,836
Speaker 1: they're useful in and of themselves, but also as a

481
00:23:00,916 --> 00:23:04,396
Speaker 1: testing ground for I guess for AI in particular. Right,

482
00:23:04,396 --> 00:23:07,076
Speaker 1: I mean if you think of of chests famously and

483
00:23:07,076 --> 00:23:09,516
Speaker 1: then go, you know, you had deep mind figure out

484
00:23:09,596 --> 00:23:11,676
Speaker 1: chess and then go and then they solve the protein

485
00:23:11,756 --> 00:23:17,756
Speaker 1: folding problem, this profound problem in in biochemistry basically, right, Um,

486
00:23:18,636 --> 00:23:25,356
Speaker 1: tell me about that. I mean, well, why are games

487
00:23:25,476 --> 00:23:29,196
Speaker 1: and sports a good place to start? Yeah? So I

488
00:23:29,196 --> 00:23:30,916
Speaker 1: think that, you know, I think one of the things

489
00:23:30,916 --> 00:23:33,636
Speaker 1: that we're doing in sports is the general problem of

490
00:23:33,756 --> 00:23:36,316
Speaker 1: human activity recognition. Right, So that's really what we're doing.

491
00:23:36,316 --> 00:23:37,796
Speaker 1: But people do a bunch of stuff in a space,

492
00:23:37,796 --> 00:23:39,396
Speaker 1: and we want to figure out what they're doing. And

493
00:23:39,636 --> 00:23:41,956
Speaker 1: you know, we didn't actually put you know, sports in

494
00:23:41,996 --> 00:23:43,996
Speaker 1: the name of the company because we didn't we thought

495
00:23:43,996 --> 00:23:45,796
Speaker 1: we might go beyond sports, but it did not. There

496
00:23:45,836 --> 00:23:47,156
Speaker 1: were so much stuff to do in sports we sort

497
00:23:47,156 --> 00:23:50,076
Speaker 1: have stayed here. Sports is interesting because it's just it

498
00:23:50,156 --> 00:23:53,436
Speaker 1: has the ability to create a lot of data capture

499
00:23:54,036 --> 00:23:58,036
Speaker 1: and the activities are are sort of bounded and well known,

500
00:23:58,196 --> 00:24:00,756
Speaker 1: so you can just have this, you know, this intense

501
00:24:00,876 --> 00:24:05,196
Speaker 1: capture of this rectangle, various rectangles based on sports, and

502
00:24:05,236 --> 00:24:08,436
Speaker 1: then it's very clear what some of the activity recognition

503
00:24:08,916 --> 00:24:10,396
Speaker 1: is and so it's a it's a great place to

504
00:24:10,436 --> 00:24:13,516
Speaker 1: sort of start with human activity recognition. So what is

505
00:24:14,156 --> 00:24:19,516
Speaker 1: what is the like obvious adjacent thing in the world

506
00:24:19,836 --> 00:24:22,596
Speaker 1: where what you do might be useful. I think that

507
00:24:22,636 --> 00:24:26,516
Speaker 1: if you look around the world, this can be applied everywhere.

508
00:24:26,556 --> 00:24:29,156
Speaker 1: So for example, you know, in your house, right if

509
00:24:29,156 --> 00:24:31,116
Speaker 1: you have cameras in your house, and you'll be able

510
00:24:31,116 --> 00:24:32,876
Speaker 1: to figure it out like oh, you know, someone has

511
00:24:32,876 --> 00:24:35,516
Speaker 1: fell down or some you know, the kids are fighting

512
00:24:35,636 --> 00:24:38,116
Speaker 1: right my house. It feels a little creepy to me,

513
00:24:38,316 --> 00:24:40,836
Speaker 1: like I don't particularly want it in my house. I

514
00:24:40,876 --> 00:24:43,636
Speaker 1: will say that's fine, Yeah I'm not I'm not necessarily

515
00:24:43,676 --> 00:24:45,716
Speaker 1: against it, but like it doesn't feel creepy at all

516
00:24:45,796 --> 00:24:50,556
Speaker 1: in the NBA to know that. So a lot of

517
00:24:50,596 --> 00:24:52,156
Speaker 1: the other things are sort of you know, I've seen

518
00:24:52,196 --> 00:24:55,516
Speaker 1: there are other companies that basically watch how people move

519
00:24:55,556 --> 00:24:57,556
Speaker 1: around stores so you can say, oh, this is where

520
00:24:57,716 --> 00:25:00,036
Speaker 1: we should put food, or this is where people are congregating,

521
00:25:00,116 --> 00:25:03,036
Speaker 1: or this is where we get blockages, or this is

522
00:25:03,076 --> 00:25:05,356
Speaker 1: the travel patterns inside a store. So I know a

523
00:25:05,396 --> 00:25:08,116
Speaker 1: bunch of companies doing that. So I know that's sort

524
00:25:08,156 --> 00:25:12,116
Speaker 1: of security and stores. There's a bunch of companies out there.

525
00:25:12,156 --> 00:25:13,636
Speaker 1: You know. We were approached by people who are like

526
00:25:13,676 --> 00:25:15,916
Speaker 1: to go to concert venues and say, okay, can you

527
00:25:15,996 --> 00:25:17,476
Speaker 1: can you put a bunch of cameras and figure out

528
00:25:17,516 --> 00:25:20,396
Speaker 1: how people move in and out of concerts and figure

529
00:25:20,436 --> 00:25:23,836
Speaker 1: out where the bottleneck exactly, how does the foot traffic flow.

530
00:25:23,916 --> 00:25:25,916
Speaker 1: So we would have approached by lots of different industries

531
00:25:25,956 --> 00:25:28,116
Speaker 1: to say, can you apply what you're doing to us?

532
00:25:28,316 --> 00:25:30,596
Speaker 1: And what have you said? When those people have approached you,

533
00:25:30,876 --> 00:25:32,556
Speaker 1: I always like I would like to get to it.

534
00:25:32,596 --> 00:25:34,756
Speaker 1: But I think the sports has kept us so busy

535
00:25:34,996 --> 00:25:37,476
Speaker 1: over the years that we've just had plenty of work

536
00:25:37,476 --> 00:25:41,636
Speaker 1: to do in sports. In a minute, the lightning round

537
00:25:41,956 --> 00:25:44,636
Speaker 1: with lots of questions about basketball and a little bit

538
00:25:44,676 --> 00:25:52,956
Speaker 1: about soccer. That's the end of the ads. Now we're

539
00:25:52,996 --> 00:25:58,596
Speaker 1: going back to the show. Um, still a lightning round,

540
00:25:58,916 --> 00:26:01,996
Speaker 1: Lightning Round. Who is your favorite NBA player of all time?

541
00:26:02,876 --> 00:26:05,916
Speaker 1: Larry Bird? What does the data say about Larry Bird

542
00:26:05,956 --> 00:26:10,396
Speaker 1: as a player? He was very good? That lines up.

543
00:26:11,716 --> 00:26:14,636
Speaker 1: So it seems like still, at least to some extent,

544
00:26:14,756 --> 00:26:18,596
Speaker 1: there is there is this cultural divide in the NBA

545
00:26:18,676 --> 00:26:20,836
Speaker 1: and in other sports between the data people and the

546
00:26:20,876 --> 00:26:23,876
Speaker 1: sort of old school sports people. And I'm curious, what

547
00:26:24,036 --> 00:26:27,356
Speaker 1: do the data people not get about the sort of

548
00:26:27,396 --> 00:26:32,116
Speaker 1: traditional sports people. I think that what happened was, you know,

549
00:26:32,236 --> 00:26:36,316
Speaker 1: the traditional sport sports people spoken words and the data

550
00:26:36,316 --> 00:26:39,996
Speaker 1: people spoken numbers, and I think that that's what needed

551
00:26:40,036 --> 00:26:42,316
Speaker 1: to bridge, Like basically both people had to speak words

552
00:26:42,316 --> 00:26:44,356
Speaker 1: in numbers, and I think that's happened so that you

553
00:26:44,396 --> 00:26:48,196
Speaker 1: think that divide is done now it no longer exists. Yeah,

554
00:26:48,236 --> 00:26:49,956
Speaker 1: Is it still fun for you to watch basketball? Or

555
00:26:49,996 --> 00:26:52,316
Speaker 1: does it just feel like work. Oh no, it's fun.

556
00:26:53,756 --> 00:26:55,996
Speaker 1: It's fun. According to the data, who is the greatest

557
00:26:55,996 --> 00:26:59,436
Speaker 1: basketball player of all time? We're unable to do that

558
00:26:59,516 --> 00:27:02,236
Speaker 1: because the data capture only goes back five years. Who's

559
00:27:02,276 --> 00:27:04,596
Speaker 1: the greatest player of the last five years? The last

560
00:27:04,636 --> 00:27:09,236
Speaker 1: five years? Steph Carry? Who's the greatest soccer player the

561
00:27:09,316 --> 00:27:12,876
Speaker 1: last five years? MESSI? Can you compare Steph Curry and

562
00:27:12,916 --> 00:27:16,956
Speaker 1: Messy in some quantitative way? Yeah? I think that all

563
00:27:16,996 --> 00:27:20,436
Speaker 1: the great players have some number where they perform something

564
00:27:20,876 --> 00:27:25,316
Speaker 1: far above expectation. I mean, can you is it a

565
00:27:25,396 --> 00:27:27,916
Speaker 1: dumb question to say who's better Steph Curry or MESSI?

566
00:27:28,716 --> 00:27:31,676
Speaker 1: It's not, but it's not a question that coaches tend

567
00:27:31,756 --> 00:27:34,276
Speaker 1: to worry about. Well. Sure, but but if it's not

568
00:27:34,316 --> 00:27:36,396
Speaker 1: a dumb question, I'm going to ask you who's better

569
00:27:36,436 --> 00:27:39,556
Speaker 1: Steph Curry or Messy? The reason is that we haven't

570
00:27:39,596 --> 00:27:41,876
Speaker 1: built it. So it's almost every question requires building a

571
00:27:41,916 --> 00:27:44,156
Speaker 1: set of tools to answer them. We could, but as

572
00:27:44,196 --> 00:27:45,796
Speaker 1: no one has asked to build those tools do. I

573
00:27:45,796 --> 00:27:47,756
Speaker 1: think what you would want to do is say you

574
00:27:47,756 --> 00:27:49,676
Speaker 1: could answer that question, but I'd have to pay you

575
00:27:49,756 --> 00:27:51,836
Speaker 1: to do it. Basically, because I think, like I think,

576
00:27:51,876 --> 00:27:53,436
Speaker 1: there is a way to say it because like how

577
00:27:53,516 --> 00:27:55,836
Speaker 1: much of it? Basically, these questions you're asking is how

578
00:27:55,916 --> 00:28:00,036
Speaker 1: much was an outlier? Was this particular person compared to everyone? Yeah?

579
00:28:00,076 --> 00:28:02,956
Speaker 1: How much value did he add? Would be a clumsy

580
00:28:02,996 --> 00:28:04,956
Speaker 1: way to say it. Yeah, yeah, And I think that

581
00:28:04,996 --> 00:28:06,876
Speaker 1: there are definitely ways to answer that. That's sort of

582
00:28:06,916 --> 00:28:09,316
Speaker 1: not that is not where we have spent our time.

583
00:28:09,316 --> 00:28:10,916
Speaker 1: But I don't think it's unanswerable. I think that there

584
00:28:10,956 --> 00:28:13,076
Speaker 1: are interesting ways you can go about answering that question.

585
00:28:13,196 --> 00:28:15,436
Speaker 1: This one goes to sort of data versus kind of

586
00:28:15,476 --> 00:28:21,396
Speaker 1: public acclaim among NBA players, Like based on the data

587
00:28:21,556 --> 00:28:26,716
Speaker 1: versus what people think in general. Who's the most underrated

588
00:28:26,756 --> 00:28:30,996
Speaker 1: player right now? Say Chris Paul. I mean he's not

589
00:28:31,036 --> 00:28:33,836
Speaker 1: really underrated, but he's not quite at the pantheon. You

590
00:28:33,916 --> 00:28:36,676
Speaker 1: can be highly rated but still underrated. Underrated does not

591
00:28:36,756 --> 00:28:39,516
Speaker 1: mean low rated. It means not rated high enough. That's right.

592
00:28:40,116 --> 00:28:43,396
Speaker 1: Who do you think the most overrated player? Oh, that's

593
00:28:43,396 --> 00:28:46,196
Speaker 1: a good question. It's it's it's tough. I don't know.

594
00:28:46,316 --> 00:28:49,316
Speaker 1: I'm saying no. One just jumps to mind because I think, like, well, also,

595
00:28:49,356 --> 00:28:51,116
Speaker 1: you get it. You could get in trouble, right, you're

596
00:28:51,156 --> 00:28:54,036
Speaker 1: gonna you're gonna call out one are your clients stars? Well,

597
00:28:54,076 --> 00:28:55,716
Speaker 1: there's there's two questions when it's like, hey, I thought

598
00:28:55,716 --> 00:28:57,196
Speaker 1: of someone, but I don't know who to say. I

599
00:28:57,236 --> 00:28:58,956
Speaker 1: don't want to say it, but right now I can't

600
00:28:58,956 --> 00:29:01,716
Speaker 1: actually think it because because I I it's a lot

601
00:29:01,756 --> 00:29:05,076
Speaker 1: of people. People are much better at rating nowadays. I

602
00:29:05,116 --> 00:29:06,996
Speaker 1: think that's that's a lot of what has changed over

603
00:29:07,036 --> 00:29:09,716
Speaker 1: the years is that because there's so many more numbers

604
00:29:09,796 --> 00:29:14,236
Speaker 1: like the error bar, and ratings have gotten a lot narrower.

605
00:29:14,276 --> 00:29:16,476
Speaker 1: So the people have gotten more, players have gotten more

606
00:29:16,476 --> 00:29:20,556
Speaker 1: appropriately rated because fans are savvier about exactly. So what

607
00:29:20,636 --> 00:29:23,916
Speaker 1: has happened in the place is big centers are not

608
00:29:23,996 --> 00:29:27,076
Speaker 1: as valuable in basketball. But you actually see that, like

609
00:29:27,116 --> 00:29:28,756
Speaker 1: they're not rated as highly, you know, and so a

610
00:29:28,796 --> 00:29:30,716
Speaker 1: lot of the a lot of big centers who would

611
00:29:30,716 --> 00:29:33,116
Speaker 1: have had you know, massive contracts ten plus years ago

612
00:29:33,276 --> 00:29:35,356
Speaker 1: aren't getting those now. But that's because you know, the

613
00:29:35,436 --> 00:29:38,636
Speaker 1: ratings have adapted to value them less. The data shows

614
00:29:38,636 --> 00:29:40,956
Speaker 1: that big centers aren't as valuable as people thought. That's right.

615
00:29:41,196 --> 00:29:43,996
Speaker 1: You think you're going to work at second spectrum for forever,

616
00:29:44,036 --> 00:29:46,636
Speaker 1: for as long as you're working. It's our baby, it's

617
00:29:46,636 --> 00:29:48,516
Speaker 1: been our the baby of many, many people. I mean,

618
00:29:48,516 --> 00:29:52,356
Speaker 1: babies grow up. I think that there are problems I

619
00:29:52,396 --> 00:29:54,036
Speaker 1: want to solve, and sure I would love to be

620
00:29:54,036 --> 00:29:55,916
Speaker 1: able to solve all those problems, and then once they

621
00:29:55,956 --> 00:29:57,796
Speaker 1: I'll say what will I do next? But I think

622
00:29:58,076 --> 00:30:04,276
Speaker 1: there's certainly plenty of problems to be solved. Rejieve mehs

623
00:30:04,356 --> 00:30:08,636
Speaker 1: Run is the co founder and president of Second Spectrum.

624
00:30:08,676 --> 00:30:12,116
Speaker 1: Today's show was produced by Edith Russolo, engineered by Amanda

625
00:30:12,196 --> 00:30:16,076
Speaker 1: kay Wong, and edited by Robert Smith. I'm Jacob Goldstein,

626
00:30:16,156 --> 00:30:18,196
Speaker 1: and we'll be back next week with another episode of

627
00:30:18,236 --> 00:30:25,076
Speaker 1: What's Your Problem