1
00:00:00,000 --> 00:00:02,040
Speaker 1: Well, was that a venture partners They invest in what

2
00:00:02,200 --> 00:00:06,080
Speaker 1: is called AI first companies. We're talking about artificial intelligence.

3
00:00:06,080 --> 00:00:08,640
Speaker 1: Those are companies and entrepreneurs that are applying machine learning

4
00:00:08,960 --> 00:00:11,720
Speaker 1: to the real world across the industries and functionality. So

5
00:00:11,800 --> 00:00:15,480
Speaker 1: let's get into it with Ash Fontana, managing director of Zetta.

6
00:00:15,760 --> 00:00:18,800
Speaker 1: Before Zetta, he launched syndicates at angel List, his new

7
00:00:18,840 --> 00:00:21,280
Speaker 1: book The AI First Company, How to Compete and Win

8
00:00:21,640 --> 00:00:25,479
Speaker 1: with Artificial Intelligence, and Ash getting up early joining us

9
00:00:25,480 --> 00:00:27,560
Speaker 1: on the phone from Sydney, Australia, where it's I think

10
00:00:27,640 --> 00:00:29,880
Speaker 1: six thirty in the morning. Nice to have you here,

11
00:00:29,960 --> 00:00:33,800
Speaker 1: Welcome to Bloomberg. Thank you so much for having on

12
00:00:33,840 --> 00:00:36,320
Speaker 1: your side. Well, it's a pleasure. How are you and

13
00:00:36,560 --> 00:00:39,760
Speaker 1: tell us about this past year, um, covid this book.

14
00:00:40,920 --> 00:00:45,159
Speaker 1: What's it been like? Mm hmm, Well, I'm well it

15
00:00:45,240 --> 00:00:47,280
Speaker 1: was it was sort of a funny year, and that

16
00:00:47,400 --> 00:00:49,560
Speaker 1: it was a good year to write a book, um

17
00:00:49,600 --> 00:00:53,640
Speaker 1: to the inside and whatnot, um and finishing all those

18
00:00:53,760 --> 00:00:56,120
Speaker 1: all those loose ends that they're just lying around in

19
00:00:56,160 --> 00:00:59,160
Speaker 1: your house when you see you doing nothing else. So UM,

20
00:00:59,440 --> 00:01:01,080
Speaker 1: it was a good year in that regard. It was

21
00:01:01,120 --> 00:01:02,960
Speaker 1: a very funny year to be in the field of

22
00:01:03,040 --> 00:01:06,720
Speaker 1: artificial intelligence, because uh, you know, we've all heard this

23
00:01:06,800 --> 00:01:09,720
Speaker 1: from a few different people, but the acceleration was incredible,

24
00:01:10,080 --> 00:01:13,960
Speaker 1: and the imperative to automate as the sort of imperative

25
00:01:14,000 --> 00:01:16,360
Speaker 1: to get a better understanding of what's going on in

26
00:01:16,360 --> 00:01:18,640
Speaker 1: our world when we couldn't be out in the world

27
00:01:19,080 --> 00:01:22,760
Speaker 1: UM really increased, and so that the right into adoption

28
00:01:22,800 --> 00:01:25,840
Speaker 1: of these technologies really want through the root, meaning what like,

29
00:01:25,880 --> 00:01:28,200
Speaker 1: give us some examples of things that because I do

30
00:01:28,360 --> 00:01:31,560
Speaker 1: think ash it's fair to say that for a lot

31
00:01:31,640 --> 00:01:34,399
Speaker 1: of people, and I'm not making a judgment call, but

32
00:01:34,480 --> 00:01:37,080
Speaker 1: when we think of AI, it's often guided by Hollywood

33
00:01:37,120 --> 00:01:39,520
Speaker 1: interpretation of it, or we think about you know, controlling

34
00:01:39,520 --> 00:01:44,280
Speaker 1: our minds UM. But it's already in our world and

35
00:01:44,319 --> 00:01:49,160
Speaker 1: it's making decisions about things. M Yeah, indeed it is.

36
00:01:49,640 --> 00:01:53,200
Speaker 1: And I think there's two aspects of this UM to

37
00:01:53,280 --> 00:01:56,200
Speaker 1: bring it down to sort of really examples, you know.

38
00:01:56,280 --> 00:01:59,280
Speaker 1: One is what do you do when there aren't people

39
00:01:59,280 --> 00:02:02,880
Speaker 1: around the aren't have people in close contact with each

40
00:02:02,880 --> 00:02:05,720
Speaker 1: other in a place like a warehouse, Well, you have

41
00:02:05,840 --> 00:02:09,600
Speaker 1: to use the machine. So firstly, understand what people would

42
00:02:09,639 --> 00:02:12,440
Speaker 1: be doing that warehouse or the usual process let's observe

43
00:02:12,480 --> 00:02:15,080
Speaker 1: it with some cameras and try to break it down

44
00:02:15,160 --> 00:02:18,440
Speaker 1: with some sort of system that analyzes the feed from

45
00:02:18,440 --> 00:02:21,839
Speaker 1: those cameras. Um. And then too, how do you move

46
00:02:22,000 --> 00:02:24,200
Speaker 1: things around when there are people there to do that?

47
00:02:24,760 --> 00:02:26,880
Speaker 1: People can't be there to do that, it's not safe

48
00:02:26,880 --> 00:02:29,320
Speaker 1: for them to do that, and to use robots, and

49
00:02:29,680 --> 00:02:33,240
Speaker 1: robots are funny. That's sort of like a system that

50
00:02:33,320 --> 00:02:36,320
Speaker 1: has lots of different bits of artificial intelligence in it.

51
00:02:36,760 --> 00:02:39,640
Speaker 1: So you know, that's one sort of pretty tangible example.

52
00:02:39,680 --> 00:02:42,120
Speaker 1: But you know what else happened last year was just

53
00:02:42,160 --> 00:02:45,840
Speaker 1: seven these systems were used. Whether they were energy systems

54
00:02:45,919 --> 00:02:49,200
Speaker 1: like electricity good, whether they were economic systems like what

55
00:02:49,280 --> 00:02:53,840
Speaker 1: happened in markets, they just started, um doing things you

56
00:02:53,880 --> 00:02:56,680
Speaker 1: haven't seen them do before. And you know what artificial

57
00:02:56,720 --> 00:02:59,760
Speaker 1: intelligence really is, I know people think of it a

58
00:02:59,800 --> 00:03:02,720
Speaker 1: lot on the time. Is something that you know, is

59
00:03:03,360 --> 00:03:06,120
Speaker 1: like a weird sort of mind. Um, It's really just

60
00:03:06,280 --> 00:03:08,880
Speaker 1: like a bunch of physical models strung together. And so

61
00:03:09,200 --> 00:03:12,240
Speaker 1: when all these systems started going awry, and I really

62
00:03:12,240 --> 00:03:16,679
Speaker 1: helped us understand what's going on by doing statistical analysis

63
00:03:16,880 --> 00:03:20,519
Speaker 1: on the flow of power through degrees when no one's

64
00:03:20,520 --> 00:03:23,880
Speaker 1: striving to work and everyone's staying at home. Why what

65
00:03:23,960 --> 00:03:27,480
Speaker 1: was happening there? How do we actually manage better how

66
00:03:27,560 --> 00:03:30,360
Speaker 1: we move energy around the grid in that situation that

67
00:03:30,400 --> 00:03:33,160
Speaker 1: we've mada sine before while and AI can sort of

68
00:03:33,520 --> 00:03:37,360
Speaker 1: very quickly make a new model that helps you generate

69
00:03:37,400 --> 00:03:41,880
Speaker 1: a prediction in that new world rather than you have

70
00:03:42,040 --> 00:03:44,680
Speaker 1: to rely on your old models that were used to

71
00:03:44,720 --> 00:03:49,680
Speaker 1: generate predictions under normal uses usage patterns of electricity. UM.

72
00:03:49,840 --> 00:03:52,280
Speaker 1: So it's really good at doing things like that, and

73
00:03:52,320 --> 00:03:56,080
Speaker 1: the advantage to it is that it's constantly able to

74
00:03:56,160 --> 00:03:59,800
Speaker 1: take in and make those decisions using real time data.

75
00:03:59,840 --> 00:04:02,000
Speaker 1: So as the data can change, as we saw a

76
00:04:02,080 --> 00:04:04,560
Speaker 1: year ago or a little bit more than a year ago,

77
00:04:04,800 --> 00:04:10,000
Speaker 1: the data points changed dramatically. Nobody would have predicted it initially,

78
00:04:10,200 --> 00:04:13,160
Speaker 1: right until, of course we started to see how significant

79
00:04:13,200 --> 00:04:16,520
Speaker 1: and how severe the pandemic was. But that's the advantage

80
00:04:16,560 --> 00:04:22,440
Speaker 1: of AI. Yeah, that's exactly right. UM. You know a

81
00:04:22,480 --> 00:04:26,000
Speaker 1: lot of these systems are trained in an environment that

82
00:04:26,279 --> 00:04:29,240
Speaker 1: represents a real world environment. But every time they get

83
00:04:29,279 --> 00:04:32,960
Speaker 1: a new observation, every time they take a new photo

84
00:04:33,120 --> 00:04:37,240
Speaker 1: of what's on a shelf and supermarket or what's how

85
00:04:37,320 --> 00:04:39,960
Speaker 1: things are moving around the warehouse. Will they make an

86
00:04:39,960 --> 00:04:43,279
Speaker 1: observation of how power is flowing through the grid, you know,

87
00:04:43,360 --> 00:04:46,279
Speaker 1: it's flowing to this part of the grid and not

88
00:04:46,400 --> 00:04:49,520
Speaker 1: that part of the grid. It learns and it updates

89
00:04:49,320 --> 00:04:51,920
Speaker 1: the view of the world. Now we do that too

90
00:04:52,000 --> 00:04:54,640
Speaker 1: as humans. Of course, we're always updating out view of

91
00:04:54,640 --> 00:04:57,560
Speaker 1: the world. But you know, we're actually very hesitant to

92
00:04:57,680 --> 00:05:00,280
Speaker 1: update out of view of the world because, um, we

93
00:05:00,400 --> 00:05:03,680
Speaker 1: like being able to make decisions quickly based on instinct,

94
00:05:03,720 --> 00:05:06,400
Speaker 1: and that requires having very solid models of the world.

95
00:05:07,200 --> 00:05:10,880
Speaker 1: Whereas AIS don't sort of has that bias. They don't

96
00:05:10,880 --> 00:05:14,159
Speaker 1: have that recency bias, so to speak, and they don't

97
00:05:14,160 --> 00:05:18,200
Speaker 1: have that need to be super intuitive about things. So

98
00:05:18,440 --> 00:05:20,560
Speaker 1: they're updating the view of the world very very quickly,

99
00:05:20,600 --> 00:05:22,680
Speaker 1: and that can be really helpful. And in the situation

100
00:05:23,880 --> 00:05:26,920
Speaker 1: so as what is an AI first company because you

101
00:05:26,960 --> 00:05:31,200
Speaker 1: invest in them, Yeah, and and our first company is

102
00:05:31,200 --> 00:05:35,799
Speaker 1: a company that genuinely puts AI artificial intelligence to start

103
00:05:35,880 --> 00:05:38,840
Speaker 1: every conversation, in the conversations about who they're going to

104
00:05:38,960 --> 00:05:41,240
Speaker 1: hire next, you go to hire people that you know

105
00:05:41,360 --> 00:05:44,760
Speaker 1: can build these models, build these these machine learning models

106
00:05:44,800 --> 00:05:48,640
Speaker 1: that eventually start looking like AIS. You know what Are

107
00:05:48,640 --> 00:05:50,280
Speaker 1: you going to put your money until you're gonna put

108
00:05:50,320 --> 00:05:54,080
Speaker 1: your money into acquiring valuable data? What policy concerns do

109
00:05:54,160 --> 00:05:56,479
Speaker 1: you have? Are you really going to play a role

110
00:05:56,480 --> 00:06:00,279
Speaker 1: in the data own data privacy? Are first companies really

111
00:06:00,560 --> 00:06:04,799
Speaker 1: put these conversations first so that they can genuinely build

112
00:06:04,800 --> 00:06:07,880
Speaker 1: this stuff rather than try to sprinkle it on top later.

113
00:06:08,839 --> 00:06:11,280
Speaker 1: So are these the company? So help me out here?

114
00:06:11,320 --> 00:06:15,920
Speaker 1: Are these the companies who are creating the AI infrastructure,

115
00:06:16,080 --> 00:06:19,680
Speaker 1: the AI algorithms? Because it sounds like we are moving

116
00:06:19,720 --> 00:06:24,200
Speaker 1: increasingly to a world where you know, every company, just

117
00:06:24,240 --> 00:06:27,279
Speaker 1: like retail became kind of a digital e commerce company,

118
00:06:27,279 --> 00:06:29,840
Speaker 1: had to adopt some kind of digital strategy. We saw

119
00:06:29,880 --> 00:06:32,160
Speaker 1: that a lot in the pandemic or companies in general.

120
00:06:32,720 --> 00:06:35,760
Speaker 1: Is it Are we moving towards a world where every

121
00:06:35,800 --> 00:06:40,040
Speaker 1: company is going to have some type of AI strategy

122
00:06:40,200 --> 00:06:45,560
Speaker 1: or AI first strategy? Yeah? I really think we are.

123
00:06:45,800 --> 00:06:47,719
Speaker 1: You know, the way you put it there, I think

124
00:06:48,120 --> 00:06:50,520
Speaker 1: it's how I would put it um and that is

125
00:06:51,200 --> 00:06:55,520
Speaker 1: all software will eventually become intelligent. Every bit of technology

126
00:06:55,560 --> 00:06:58,240
Speaker 1: that we use will have a bit of a conicutve

127
00:06:58,360 --> 00:07:01,279
Speaker 1: element to it. And in court we've seen so far

128
00:07:01,360 --> 00:07:04,840
Speaker 1: as a couple of really nice companies get started, like

129
00:07:05,560 --> 00:07:09,200
Speaker 1: UI Path and Palanteer and clubd Error and a couple

130
00:07:09,240 --> 00:07:11,560
Speaker 1: of these companies that have gone public. Now you know

131
00:07:11,640 --> 00:07:15,840
Speaker 1: they're really making the tools that other people build their eyes. Um,

132
00:07:15,920 --> 00:07:19,640
Speaker 1: they're not really a access companies themselves. They're not building

133
00:07:19,680 --> 00:07:22,240
Speaker 1: the AI s. But you know, now that they're really

134
00:07:22,280 --> 00:07:25,400
Speaker 1: well established and their products are mature, people can use

135
00:07:25,440 --> 00:07:30,600
Speaker 1: them in retail, in manufacturing, in healthcare to build their

136
00:07:30,640 --> 00:07:35,440
Speaker 1: own aies in those fields, combining their own experience of

137
00:07:35,480 --> 00:07:40,280
Speaker 1: how things work. You know, how diseases progress, how consumers

138
00:07:40,360 --> 00:07:44,520
Speaker 1: behave with the power of these models to get a

139
00:07:44,520 --> 00:07:47,520
Speaker 1: predictive system underway. But what to give me an idea

140
00:07:47,560 --> 00:07:49,720
Speaker 1: because you do have some investments and I've been looking

141
00:07:49,720 --> 00:07:53,200
Speaker 1: at your website earlier today, tell me what is an

142
00:07:53,200 --> 00:07:55,440
Speaker 1: AI or what are some of the AI first companies

143
00:07:55,480 --> 00:07:57,200
Speaker 1: that you have invested in, and just give us an

144
00:07:57,200 --> 00:07:58,960
Speaker 1: idea of what they're doing, just so that our our

145
00:07:59,000 --> 00:08:01,120
Speaker 1: audience and the ssters that are out there, because this

146
00:08:01,200 --> 00:08:05,720
Speaker 1: does sound like something that will become more popular potentially

147
00:08:05,720 --> 00:08:10,120
Speaker 1: going forward. Yeah, for sure. I think A really good

148
00:08:10,120 --> 00:08:13,640
Speaker 1: example is a company called Tractable, and what they do

149
00:08:13,760 --> 00:08:16,080
Speaker 1: is they basically help people get back on their feet

150
00:08:16,080 --> 00:08:19,280
Speaker 1: after being a disaster, you know, whether that's a weather

151
00:08:19,360 --> 00:08:22,200
Speaker 1: disaster or having a car crash. And the way they

152
00:08:22,280 --> 00:08:26,280
Speaker 1: do that they analyze images. So you have a little

153
00:08:26,360 --> 00:08:28,840
Speaker 1: fender vendor on the way to work, you take a

154
00:08:28,920 --> 00:08:32,920
Speaker 1: photo and they can analyze that photo a naked decision

155
00:08:33,440 --> 00:08:36,679
Speaker 1: almost on the spot about whether your car is going

156
00:08:36,720 --> 00:08:40,280
Speaker 1: to be repaired or whether it's a write off. And

157
00:08:40,600 --> 00:08:42,920
Speaker 1: that just gets the money in your account more quickly.

158
00:08:43,200 --> 00:08:45,640
Speaker 1: It gets you back on the road more quickly, and

159
00:08:45,840 --> 00:08:49,000
Speaker 1: it for the insurance company makes the cost of process

160
00:08:49,040 --> 00:08:51,960
Speaker 1: in that claim a lot lower. And you know how

161
00:08:52,080 --> 00:08:54,640
Speaker 1: is that an AI first company, They had to for

162
00:08:54,760 --> 00:08:58,200
Speaker 1: two years before they got any customers, really go and

163
00:08:58,280 --> 00:09:02,920
Speaker 1: collect so much datas, so many different images, different ways

164
00:09:02,960 --> 00:09:06,520
Speaker 1: in which cars go through crashes, and how different panels

165
00:09:06,880 --> 00:09:10,120
Speaker 1: deform and whatnot, so that they could train their models

166
00:09:10,160 --> 00:09:12,719
Speaker 1: to the point where they can develop a really accurate

167
00:09:13,120 --> 00:09:18,600
Speaker 1: readoubt of the damage really quickly, just purely using computer

168
00:09:18,720 --> 00:09:23,000
Speaker 1: vision and not using getting a loss adjuster or having

169
00:09:23,040 --> 00:09:24,640
Speaker 1: someone come out and have a look at the car.

170
00:09:25,280 --> 00:09:27,760
Speaker 1: How much is that being used by the insurance industry,

171
00:09:27,760 --> 00:09:31,920
Speaker 1: the auto industry at this point. Yeah, it's being used

172
00:09:31,920 --> 00:09:34,640
Speaker 1: today by major insurers all around the world to make

173
00:09:34,679 --> 00:09:37,600
Speaker 1: decisions every day. Um, and then you know, you can

174
00:09:37,640 --> 00:09:41,199
Speaker 1: see how it could soon be used for other other purposes.

175
00:09:41,280 --> 00:09:43,080
Speaker 1: You know, if you're in a hail storm and your

176
00:09:43,160 --> 00:09:45,880
Speaker 1: riot gets damaged, well, they can take a photo from

177
00:09:45,960 --> 00:09:48,800
Speaker 1: space or from a drone and then analyze that photo

178
00:09:48,880 --> 00:09:50,960
Speaker 1: and quickly give you an assessment of the damage and

179
00:09:51,320 --> 00:09:53,959
Speaker 1: processes you're playing. You can see how it could be

180
00:09:54,040 --> 00:09:56,680
Speaker 1: used in all sorts of the types of situations. But today,

181
00:09:57,280 --> 00:10:00,200
Speaker 1: dozens and dozens of the world's largest insurance company these

182
00:10:01,280 --> 00:10:04,080
Speaker 1: working with them to process claims every single day, and

183
00:10:04,120 --> 00:10:06,360
Speaker 1: it's in the hands of people in their apps that

184
00:10:06,400 --> 00:10:10,559
Speaker 1: they've got on their phone and their their insurance company apps.

185
00:10:10,720 --> 00:10:12,960
Speaker 1: One thing I do want to ask you, because when

186
00:10:12,960 --> 00:10:16,480
Speaker 1: we're taking images pictures up above, I mean, I'm thinking

187
00:10:16,520 --> 00:10:19,480
Speaker 1: that there are some of our listeners or watchers on

188
00:10:19,520 --> 00:10:21,480
Speaker 1: YouTube that are just saying, well, wait a minute, Okay,

189
00:10:21,480 --> 00:10:23,600
Speaker 1: So now I start to get a little thinky about

190
00:10:23,640 --> 00:10:27,040
Speaker 1: AI and then the privacy concerns. What are the boundaries

191
00:10:27,080 --> 00:10:31,840
Speaker 1: that need to be set with these AI first companies. Yeah,

192
00:10:31,880 --> 00:10:34,800
Speaker 1: that's a great question, and you know, I'm all for

193
00:10:35,240 --> 00:10:38,640
Speaker 1: thinking about the role of government here as being a

194
00:10:38,640 --> 00:10:42,840
Speaker 1: really important one because in a sense, as as individuals

195
00:10:43,679 --> 00:10:47,880
Speaker 1: our data, my data, your data in itself is not

196
00:10:48,040 --> 00:10:51,080
Speaker 1: necessarily worth much, but as a collective data is worth

197
00:10:51,080 --> 00:10:53,600
Speaker 1: a lot to these companies. And you know, we can

198
00:10:53,640 --> 00:10:57,000
Speaker 1: work with government to make sure these companies respect our

199
00:10:57,000 --> 00:11:00,400
Speaker 1: privacy where they need to and properly compensate equal for

200
00:11:00,440 --> 00:11:04,640
Speaker 1: their data, um when they when they use it. I

201
00:11:04,640 --> 00:11:07,920
Speaker 1: think the EU is a real leading light here. Um.

202
00:11:08,000 --> 00:11:11,680
Speaker 1: You know, recently they released some legislation where they just

203
00:11:11,800 --> 00:11:16,959
Speaker 1: outright bands and applications I around facial recognition and whatnot,

204
00:11:17,360 --> 00:11:20,920
Speaker 1: and I think there are categorically some really scary uses

205
00:11:20,960 --> 00:11:23,440
Speaker 1: of it. And you know, we do have the power

206
00:11:23,920 --> 00:11:27,439
Speaker 1: through the legislature the branded stuff, um, but you know,

207
00:11:27,600 --> 00:11:31,520
Speaker 1: besides a couple of sort of very clear cases, it's

208
00:11:31,800 --> 00:11:35,040
Speaker 1: a more marginal consideration. And you know, I think having

209
00:11:35,080 --> 00:11:39,680
Speaker 1: better standards for auditing, being more upfront about what data

210
00:11:39,679 --> 00:11:41,880
Speaker 1: we're using, you know, with the cookie stuff that's hit

211
00:11:41,920 --> 00:11:45,600
Speaker 1: the news today, I think that's a really good approach.

212
00:11:45,640 --> 00:11:47,880
Speaker 1: But you know, looking to the EU and what they're doing,

213
00:11:47,960 --> 00:11:51,560
Speaker 1: I think an indication of what government do. Well look

214
00:11:51,600 --> 00:11:53,280
Speaker 1: forward to having you back and maybe at a time

215
00:11:53,280 --> 00:11:54,840
Speaker 1: where you don't have to get up so early, because

216
00:11:55,160 --> 00:11:57,880
Speaker 1: I do think this is an interesting area, uh and

217
00:11:58,000 --> 00:12:01,000
Speaker 1: certainly one that continues to development and really be a

218
00:12:01,000 --> 00:12:03,439
Speaker 1: part of our world. More broadly, Ash, thank you so much.

219
00:12:03,520 --> 00:12:07,760
Speaker 1: Ash Fontana, managing director of Zetta the his book The

220
00:12:07,840 --> 00:12:11,280
Speaker 1: AI First Company, How to Compete and Win with Artificial Intelligence,

221
00:12:11,559 --> 00:12:14,000
Speaker 1: joining us on the phone from Sydney, Australia,