WEBVTT
1
00:00:01.280 --> 00:00:07.040
Forbes Books presents the Startup Science Podcast
with Gregory Shepherd, brought to you by
2
00:00:07.080 --> 00:00:13.720
Startupscience dot Io. Greg is an
entrepreneur who has built and sold twelve businesses.
3
00:00:14.160 --> 00:00:18.800
He's a recipient of four Private Equity
awards and is a featured ten X
4
00:00:18.800 --> 00:00:24.160
speaker and Forbes author. Here's Gregory
Shepherd. I'm back with David Days,
5
00:00:24.160 --> 00:00:29.039
the professor of Data Science and Philosophy
at UC San Diego. We've been discussing
6
00:00:29.079 --> 00:00:35.679
the intersection of philosophy, cognitive science, and machine learning to tackle complex problems
7
00:00:35.759 --> 00:00:40.560
we see around AI. David,
let's dive into time series models. First,
8
00:00:40.960 --> 00:00:46.119
what is a time series model?
And second, how can entrepreneurs utilize
9
00:00:46.159 --> 00:00:50.359
that for forecasting, planning and deal
me with challenges on the fly? Yeah,
10
00:00:50.359 --> 00:00:56.240
that's a great question. So time
series models are basically models of things
11
00:00:56.240 --> 00:01:00.640
that evolve and unfold over time.
So some dime, I just want to
12
00:01:00.679 --> 00:01:06.040
predict do you have cancer at this
moment based on a radiographic image. That's
13
00:01:06.359 --> 00:01:08.920
not a time series model. I've
got this moment in time, do you
14
00:01:08.959 --> 00:01:14.120
have cancer? But often what I
want to do is predict how might your
15
00:01:14.159 --> 00:01:19.120
cancer unfold over time under various treatment
regimes? So if I give you this
16
00:01:19.200 --> 00:01:23.560
chemotherapy or this surgical intervention or this
radiation treatment, can we figure out how
17
00:01:23.560 --> 00:01:29.879
the cancer is likely to change over
time? That requires us to explicitly model
18
00:01:30.200 --> 00:01:34.640
time. We have to think about
the ways that things change over time,
19
00:01:34.680 --> 00:01:38.000
that the cancer probably isn't going to
change much in one minute, but it
20
00:01:38.040 --> 00:01:42.920
will change over one year, And
so it raises lots of difficult challenges from
21
00:01:42.920 --> 00:01:48.840
a data science perspective in terms of
how we model time. How granular are
22
00:01:48.879 --> 00:01:53.120
we There's a kind of trade off
of if I measure a lot very quickly,
23
00:01:53.359 --> 00:01:57.879
that's great for accuracy, but computationally
inefficient. Measure every once in a
24
00:01:57.879 --> 00:02:01.840
while that'sationally efficient, but I'll get
less accurate. So it's these kinds of
25
00:02:01.879 --> 00:02:07.560
trade offs that we worry about.
Now. For entrepreneurs, the key is
26
00:02:07.640 --> 00:02:13.400
to think about what kinds of sort
of time data you're going to have,
27
00:02:13.680 --> 00:02:17.599
What access are you going to have
to learn about how things unfold over time
28
00:02:19.080 --> 00:02:23.000
in whatever domain or problem you're looking
at, and then figuring out how do
29
00:02:23.039 --> 00:02:28.240
you balance this trade off of you
don't want to remember all of the past
30
00:02:28.599 --> 00:02:32.560
because that's computationally a nightmare, but
you want to remember enough of the past
31
00:02:34.000 --> 00:02:38.560
to inform your future predictions, to
be able to help people plan out and
32
00:02:38.719 --> 00:02:43.639
trace out the possible trajectories of what
lies in front of them. And that
33
00:02:44.240 --> 00:02:47.159
is, you know, there's no
technical answer to that question. That trade
34
00:02:47.199 --> 00:02:52.879
off of how much do I remember? How do I balance accuracy in my
35
00:02:52.960 --> 00:02:58.639
future predictions with the computational cost.
There's no technical solution to that. That's
36
00:02:58.680 --> 00:03:01.639
going to depend on how important is
it to get the future exactly right?
37
00:03:01.680 --> 00:03:07.400
How far out in the future am
I trying to predict? How much do
38
00:03:07.520 --> 00:03:10.919
my customers care about me trying to
hold on to the past data? Do
39
00:03:10.960 --> 00:03:15.599
I have access to really big compute
and storage and all these kinds of things.
40
00:03:15.639 --> 00:03:20.280
So what's important for an entrepreneur,
I think is I mean, you're
41
00:03:20.319 --> 00:03:23.080
going to notice a theme in some
of the things I'm saying. What's important
42
00:03:23.159 --> 00:03:27.080
is that they be intentional, that
they stop and they think to themselves,
43
00:03:27.439 --> 00:03:31.560
Okay, what are we trying to
achieve? What kind of balance is most
44
00:03:31.639 --> 00:03:36.599
important or most valuable? And am
I sure that I'm getting things right from
45
00:03:36.599 --> 00:03:42.400
the perspective of my customers or my
clients. So when you're thinking about these
46
00:03:42.560 --> 00:03:47.599
time series models, how can it
be used to identify capital and emerging market
47
00:03:47.680 --> 00:03:53.280
trends, right, and these customer
behavior patterns, Like it seems like,
48
00:03:53.280 --> 00:04:01.280
you know, the whole buying and
selling of stocks will be incredible disruptive because
49
00:04:01.280 --> 00:04:05.199
of these time series models. Right, So you could predict based on certain
50
00:04:05.240 --> 00:04:11.280
sets of data that are publicly available
outcomes. And then you know, like
51
00:04:11.319 --> 00:04:15.479
I said earlier, like a lot
of the white collar people, I mean,
52
00:04:15.560 --> 00:04:17.279
you know, what do you need
a broker for if the system is
53
00:04:17.519 --> 00:04:21.639
you know, better than an army
of brokers. So what are your thoughts
54
00:04:21.639 --> 00:04:25.639
on that sort of stuff? Yeah, So the first thing to say is
55
00:04:26.040 --> 00:04:30.160
we're already a long way down in
that world. So the vast majority of
56
00:04:30.240 --> 00:04:35.240
trades on stock markets are now done
by automated trading systems that are exactly using
57
00:04:35.279 --> 00:04:43.240
time series models to try to predict
very short run fluctuations in where markets are
58
00:04:43.240 --> 00:04:46.439
going and where particular stocks are going, and to try and take advantage of
59
00:04:46.079 --> 00:04:51.759
very small inefficiencies. Now, what
that highlights is that we're using automation.
60
00:04:51.959 --> 00:04:59.519
We're using AI systems essentially right now
for the very quick decisions. We're not
61
00:04:59.639 --> 00:05:03.279
using as much for the longer term
decisions for where is a sector going what
62
00:05:03.480 --> 00:05:08.240
is a good value stock. We're
still, I think, in a lot
63
00:05:08.279 --> 00:05:14.240
of ways, relying on human expertise, in part because we don't know exactly
64
00:05:14.439 --> 00:05:18.560
which information about the world is most
important, So we're relying on human expertise
65
00:05:18.959 --> 00:05:24.560
to do a lot of that information
integration for us. But you're absolutely right
66
00:05:24.879 --> 00:05:30.720
that it's one of the I would
say holy grails of fintech is how do
67
00:05:30.800 --> 00:05:35.079
we come up with models that are
really good predictors at the let's say,
68
00:05:35.160 --> 00:05:40.279
one week time scale. We've got
models that are very good at the quarterly
69
00:05:40.399 --> 00:05:43.480
predictions. We've got models that are
very good at the one second predictions.
70
00:05:44.079 --> 00:05:46.439
It's actually the middle ground where I
think that there's a lot of uncertainty right
71
00:05:46.480 --> 00:05:50.319
now and a lot of challenges are
arising. One of the issues that comes
72
00:05:50.399 --> 00:05:54.279
up, of course, with the
kinds of time series models you're talking about,
73
00:05:54.399 --> 00:05:59.600
say in financial technologies, is that
you're trying to model the very thing
74
00:05:59.600 --> 00:06:02.120
that you're trying to influence. I
say, there's a sort of self referential
75
00:06:02.199 --> 00:06:06.839
problem here, which is I could
get the model exactly right, but then
76
00:06:06.839 --> 00:06:10.920
I want to do things. I
want to take advantage of inefficiencies in the
77
00:06:10.959 --> 00:06:15.360
market, and that activity the act
of trying to take advantage of the inefficiencies
78
00:06:15.639 --> 00:06:19.959
can actually mean that I'm falsifying my
own model. I'm making my own model
79
00:06:20.000 --> 00:06:25.920
wrong. Oh, inter The other
big challenge is how do we have models
80
00:06:26.000 --> 00:06:30.360
that explicitly include here's the things that
I might do, so that we have
81
00:06:30.399 --> 00:06:35.839
a model that can be a bit
more stable over time. So it seems
82
00:06:35.879 --> 00:06:40.399
that if you're looking at the startup
ecosystem, right, it seems like you
83
00:06:40.399 --> 00:06:44.720
could do you know, you would
have investors in private equity using this right.
84
00:06:44.800 --> 00:06:48.360
So you submit something to the investor
that gives them a bunch of data.
85
00:06:48.439 --> 00:06:54.360
They put it into a series and
then it spits out here's the possible
86
00:06:54.600 --> 00:07:00.560
outcomes or probable outcomes for this particular
business. And I'm not sure if that's
87
00:07:00.600 --> 00:07:05.720
good. You know, because when
you're early on, you know, you're
88
00:07:05.759 --> 00:07:10.160
still trying to figure out your business, and it takes a couple of years
89
00:07:10.199 --> 00:07:15.720
to understand the customer well enough.
And how you should you know, work
90
00:07:15.800 --> 00:07:19.720
through it, because you know,
you can spend years learning about the customer
91
00:07:19.879 --> 00:07:26.680
and build something wrong. So if
an investor is looking at a business and
92
00:07:26.720 --> 00:07:31.040
they're looking at these models, maybe
those models would drastically change if the founder
93
00:07:31.120 --> 00:07:36.079
made a different decision or pivoted or
whatever. So what are the dangers that
94
00:07:36.120 --> 00:07:42.480
you see in how could a founder
utilize it instead of for risk, use
95
00:07:42.519 --> 00:07:46.920
it for opportunity. Yeah, so
I think you're right to raise all of
96
00:07:46.959 --> 00:07:50.920
these kinds of risks, the risk
of sort of the homogenization of startups where
97
00:07:50.920 --> 00:07:57.639
everybody gets directed towards the same thing, the risk of self fulfilling prophecies.
98
00:07:57.680 --> 00:08:01.360
If we do this, then then
we're going to end up in this feedback
99
00:08:01.399 --> 00:08:05.279
loop where we think we're getting it
right, but what we're actually doing is
100
00:08:05.319 --> 00:08:09.360
misreading or misinterpreting what it is that
our customers want. So I think you're
101
00:08:09.480 --> 00:08:15.000
right about that. I think you
also hit on exactly the key idea,
102
00:08:15.040 --> 00:08:20.560
which is you can in some sense
use these models as predictions about everybody else,
103
00:08:20.279 --> 00:08:26.160
and that can help you to see
where are the blank spots in the
104
00:08:26.160 --> 00:08:28.000
map, Where are the parts of
the map that nobody else is going to
105
00:08:28.000 --> 00:08:33.000
be ending up in. So,
and I don't think i'm revealing anything.
106
00:08:33.039 --> 00:08:35.879
Hopefully that isn't too big of a
surprise to some of your listeners, But
107
00:08:37.440 --> 00:08:39.559
a lot of good BC companies are
already doing this. Private equity companies,
108
00:08:39.559 --> 00:08:43.799
they're already doing this. They're trying
to predict where's everybody else going to go,
109
00:08:45.519 --> 00:08:48.639
so that they can figure out where
the marginal advantages might lie, where's
110
00:08:48.679 --> 00:08:54.480
the part of the landscape that is
unexplored, And so I think it's a
111
00:08:54.480 --> 00:09:01.039
great idea. The trick with all
of these things is the quality and quantity
112
00:09:01.080 --> 00:09:05.919
of data that we have in AI. We sort of, you know,
113
00:09:05.919 --> 00:09:09.120
when I look back over the now
over twenty five years that I've been doing
114
00:09:09.120 --> 00:09:16.440
this in AI, we've gone through
cycles where we're constrained by not having good
115
00:09:16.480 --> 00:09:20.360
algorithms. Our algorithms and models aren't
good enough, and then we meet constrained
116
00:09:20.360 --> 00:09:24.240
by the data not being good.
So sometimes we've got algorithms and models that
117
00:09:24.320 --> 00:09:28.039
are really good, but we don't
have good enough data for them. And
118
00:09:28.080 --> 00:09:31.759
then we get good enough data,
and then we realize the shortcomings of the
119
00:09:31.759 --> 00:09:33.360
model, and so we have to
build new models, and we sort of
120
00:09:33.440 --> 00:09:39.080
oscillate back and forth between what's the
sort of rate limiting factor. Is it
121
00:09:39.159 --> 00:09:41.519
the model, the kinds of models
we have, or is it the kinds
122
00:09:41.559 --> 00:09:48.440
of data we have Right now?
We are on the we need better data
123
00:09:48.759 --> 00:09:52.519
part of that cycle. I think
Yeah, the quality is horrific right now,
124
00:09:52.559 --> 00:09:54.799
exactly. And so in fact,
we you know, we know from
125
00:09:54.840 --> 00:10:01.360
public comments that some of the big
tech company their major advances in the last
126
00:10:01.480 --> 00:10:07.000
six to twelve months have not come
because they've gotten better algorithms or because they've
127
00:10:07.000 --> 00:10:11.559
gotten more compute. They've come because
they've gotten much better data, much cleaner
128
00:10:11.639 --> 00:10:15.559
data. They've gone through and had
people label the data and make more sense
129
00:10:15.600 --> 00:10:18.440
of the data. So we really
are at a place right now that if
130
00:10:18.480 --> 00:10:22.600
you want to be able to get
ahead, the best advice I would give
131
00:10:22.600 --> 00:10:28.000
an entrepreneur is figure out if you
have access to data that nobody else has
132
00:10:28.000 --> 00:10:31.399
figured out how to use yet.
That's so, let me ask you about
133
00:10:31.399 --> 00:10:37.440
this causal discovery and what sort of
techniques could founders use to find opportunities,
134
00:10:37.519 --> 00:10:41.879
just like we were talking about the
investors, you know, finding those opportunities.
135
00:10:43.840 --> 00:10:48.320
Yeah, So the basic idea of
causal discovery is a kind of one
136
00:10:48.399 --> 00:10:50.879
that we're all familiar with, which
is we can look out of the world
137
00:10:50.879 --> 00:10:54.799
and we can see things that are
correlated with one another, but that doesn't
138
00:10:54.799 --> 00:10:58.080
mean that they're causal. So if
I want to predict or figure out whether
139
00:10:58.080 --> 00:11:03.039
it's raining outside and Grandam and San
Diego doesn't rain very much. But if
140
00:11:03.039 --> 00:11:05.519
I want to figure that out,
I can look to see if people are
141
00:11:05.519 --> 00:11:09.679
carrying umbrellas. But if I want
to make it rain, it does no
142
00:11:09.799 --> 00:11:13.200
good to tell people to start carrying
umbrellas. That's not an effective strategy to
143
00:11:13.279 --> 00:11:16.519
make it rain, to say,
hey, everybody start carrying an umbrella around.
144
00:11:16.399 --> 00:11:22.480
And so things that are good predictors
are often not good causes. They're
145
00:11:22.480 --> 00:11:26.000
not the things that we can use
to control or shape our world. But
146
00:11:26.120 --> 00:11:31.120
in fact, when we're doing policy, when we're creating new systems, we
147
00:11:31.159 --> 00:11:35.279
often want to control our world.
I don't just want to diagnose that somebody
148
00:11:35.279 --> 00:11:37.440
has cancer. I want to be
able to do something to fix the cancer.
149
00:11:37.440 --> 00:11:41.720
But that means I need causal knowledge. So over the last thirty years,
150
00:11:41.799 --> 00:11:46.080
there's been a whole lot of folks, myself included, who have developed
151
00:11:46.159 --> 00:11:52.519
new algorithms to figure out something about
the way the world works causally, for
152
00:11:52.639 --> 00:11:56.519
all data that we collect, where
we don't intervene on the world, we
153
00:11:56.559 --> 00:12:00.440
don't mess with the world, but
nonetheless we can figure out how the world
154
00:12:00.519 --> 00:12:05.759
works. So the advantage that I
think this holds for entrepreneurs is that many
155
00:12:05.919 --> 00:12:11.759
businesses depend on having some kinds of
causal knowledge. They depend on being able
156
00:12:11.799 --> 00:12:16.799
to do things in the world to
make people's lives easier, to help people
157
00:12:18.080 --> 00:12:24.600
find new kinds of information, new
opportunities, and so we often need causal
158
00:12:24.679 --> 00:12:28.480
knowledge. But most AI systems are
only going to give you good predictions.
159
00:12:28.840 --> 00:12:33.320
They're going to tell you well,
you might try having everyone carry around an
160
00:12:33.399 --> 00:12:37.519
umbrella and see if that creates rain. So there's a real opportunity I think
161
00:12:37.519 --> 00:12:43.879
for entrepreneurs to think carefully about where
and when they might need causal knowledge.
162
00:12:43.000 --> 00:12:46.000
To think about you know, what
does it matter that they're going to be
163
00:12:46.000 --> 00:12:50.120
trying to wiggle something in the world
to bring about a good outcome for their
164
00:12:50.159 --> 00:12:54.279
customer that you know this new ad
campaign, is it going to raise sales?
165
00:12:54.440 --> 00:12:58.080
Well, that requires causal knowledge,
and so I actually think a lot
166
00:12:58.080 --> 00:13:03.600
of entrepreneurs should be taking on board
these cause of discovery algorithms and using them
167
00:13:05.000 --> 00:13:09.879
to be able to better predict what's
going to happen when they make real changes.
168
00:13:09.240 --> 00:13:13.840
And it's not just in terms of
using them for your customers, it's
169
00:13:13.879 --> 00:13:18.000
also using them for yourself. If
we bring in this new market how is
170
00:13:18.039 --> 00:13:22.279
that going to impact our business?
That requires doing causal time series modeling.
171
00:13:22.320 --> 00:13:24.840
I mean, it's putting together the
two different kinds of techniques that we've been
172
00:13:24.879 --> 00:13:31.759
talking about so far. It's hard, but there are some really good introductions
173
00:13:31.840 --> 00:13:35.639
available online and a lot of open
source code, so there's a lot of
174
00:13:37.080 --> 00:13:39.879
code that's out there that makes it
very easy to build these kinds of causal
175
00:13:39.919 --> 00:13:45.960
models and to learn that the data
they might have, so you could use
176
00:13:46.000 --> 00:13:48.440
that then for on a broader basis, you could do that. I did
177
00:13:48.480 --> 00:13:52.360
this huge research on starting in twenty
sixteen, on why, when and how
178
00:13:52.399 --> 00:13:58.200
founders fail, which is what I
wrote the book about, and it was
179
00:13:58.240 --> 00:14:01.720
so manual, right, you know, you're sitting there looking at the transcriptions
180
00:14:01.720 --> 00:14:05.519
from the interviews and looking at case
use cases and all these papers that people
181
00:14:05.600 --> 00:14:11.600
drafted and all these different reports.
It sounds like you could use that exact
182
00:14:11.679 --> 00:14:16.879
system to do that in a fraction
of the time. Just point at a
183
00:14:16.000 --> 00:14:22.159
collect all the data to make sure
it's good data, and then stem the
184
00:14:22.200 --> 00:14:24.639
machine in there to go look at
the data and come out and say,
185
00:14:24.679 --> 00:14:28.120
well, you know, if a
founder does this, they have a ten
186
00:14:28.159 --> 00:14:31.799
percent more likely chance of failing versus
if they do this, they have a
187
00:14:31.799 --> 00:14:33.679
twenty percent chance of being successful.
I mean that sounds like it would that
188
00:14:33.720 --> 00:14:37.799
be a good use of this.
I think that'd be a fantastic use of
189
00:14:37.879 --> 00:14:41.559
this these methods, and I'd be
happy to talk about it with you offline.
190
00:14:41.919 --> 00:14:45.039
Yeah, I mean, I will
say so we you know, one
191
00:14:45.080 --> 00:14:46.440
of the projects that I'm working on
right now, one of the things we're
192
00:14:46.480 --> 00:14:52.200
looking at is trying to understand the
causes of blood pressure crash and surgery.
193
00:14:52.320 --> 00:14:56.559
So drop some blood pressure are a
massive predicted, a very important predictor of
194
00:14:56.759 --> 00:15:03.720
post surgical complications. And so what
we're trying to do is figure out why
195
00:15:03.759 --> 00:15:07.919
do some people have these rapid,
unexpected drops of blood pressure and others don't.
196
00:15:09.399 --> 00:15:11.840
But we don't just want to predict
it, because even if we can
197
00:15:11.879 --> 00:15:16.159
predict it, that leads the anesthesiologists
wondering, Okay, so what do I
198
00:15:16.200 --> 00:15:18.879
do now? Right? They have
a limited set of tools. Which of
199
00:15:18.919 --> 00:15:22.120
their tools should they use? But
it's going to depend on why the blood
200
00:15:22.120 --> 00:15:26.080
pressure drop happens, what the cause
of it is, And so that's what
201
00:15:26.120 --> 00:15:28.720
we're trying to figure out. And
of course we're doing this in a context
202
00:15:28.720 --> 00:15:31.639
where We can't just go in and
do AB testing, right, We can't
203
00:15:31.679 --> 00:15:37.200
just randomly start doing things to people
when they're in surgery, and so these
204
00:15:37.279 --> 00:15:39.840
kinds of methods are invaluable for those. Similarly, you want to be able
205
00:15:39.879 --> 00:15:46.399
to tell founders here are ways to
increase your chances of success that actually are
206
00:15:46.480 --> 00:15:50.159
the causes and drivers of success.
But you don't want to just do a
207
00:15:50.200 --> 00:15:52.440
bunch of AB tests on entrepreneurs.
I think people would be annoyed if you'd
208
00:15:52.440 --> 00:15:58.159
said, here, just try this
and see what happens. Yeah, the
209
00:15:58.159 --> 00:16:03.360
investors wouldn't be happy either. So
let me ask you this too. So
210
00:16:03.480 --> 00:16:07.840
when you look at this, you
know, AI MLO, the whole set,
211
00:16:07.919 --> 00:16:11.799
right, the whole thing that we
categorize as AI, do you feel
212
00:16:11.840 --> 00:16:18.240
like there are some more practical applications
that you could point to that founders that
213
00:16:18.279 --> 00:16:22.519
are listening or investors that are listening
could say, this is a direction we
214
00:16:22.559 --> 00:16:26.440
should look at specifically in AI,
because you know, right now, you
215
00:16:26.440 --> 00:16:30.360
go to webinars, you read papers, all the emails that you get,
216
00:16:30.399 --> 00:16:33.879
everything, and everybody's saying everything is
AI, you know, even though they're
217
00:16:33.879 --> 00:16:37.919
not really AI. And so you
have these founders that are sort of scrambling
218
00:16:37.960 --> 00:16:41.519
saying I need to get into this
AI thing. Where do I point myself?
219
00:16:41.879 --> 00:16:47.240
Do you have any recommendations for what
categories or verticals that they should be
220
00:16:47.279 --> 00:16:52.120
looking at specifically? Well, I
think the most important thing is to think
221
00:16:52.120 --> 00:17:00.600
about what are the tasks that are
often repeated and where you you know what
222
00:17:00.720 --> 00:17:06.680
counts as success, so very know
when you've been successful. It's the kind
223
00:17:06.720 --> 00:17:10.759
of thing that you can build AI
to do. So that's one of the
224
00:17:10.759 --> 00:17:15.839
sort of hallmarks of AI systems is
they really perform best when we know whether
225
00:17:15.880 --> 00:17:22.559
we've succeeded, And so thinking about
the overall goal and workflow and structure of
226
00:17:22.599 --> 00:17:27.839
your organization and asking where are the
pieces that we keep doing this thing over
227
00:17:27.880 --> 00:17:32.680
and over and we know what success
looks like, so we'll know that we've
228
00:17:32.720 --> 00:17:36.319
succeeded when we get there. Those
are the pieces that you want to look
229
00:17:36.359 --> 00:17:40.160
at and say, that's where we
should be using AI. The parts that
230
00:17:40.200 --> 00:17:45.839
require judgment or that are really contact
sensitive or that requires somebody to go you
231
00:17:45.880 --> 00:17:48.400
know, this case is just a
little bit different than that other one.
232
00:17:48.799 --> 00:17:51.720
Those are the places that you want
to make sure to reserve the judgment for
233
00:17:51.759 --> 00:17:56.079
the human in your organization because those
are the parts that AI is still struggling
234
00:17:56.119 --> 00:18:00.000
with. And so you know,
kind of around way, I can come
235
00:18:00.039 --> 00:18:04.680
back to your question, which is
when people ask, well, what vertical
236
00:18:06.000 --> 00:18:10.720
is you know, most right for
disruption by AI or these kinds of things,
237
00:18:10.960 --> 00:18:14.839
my response is all of them and
none of them, because in every
238
00:18:14.920 --> 00:18:19.039
vertical, in every place that people
have looked, some parts of the jobs,
239
00:18:19.200 --> 00:18:26.000
some parts of that sector are clear
repeated, easy to turn relatively speaking,
240
00:18:26.039 --> 00:18:30.240
easy to turn into AI, and
other parts are not. I mean,
241
00:18:30.279 --> 00:18:33.359
we're already seeing this, for example, with generative AI systems, where
242
00:18:33.920 --> 00:18:37.200
there was sort of this idea that
Dolly and chat gpt we're going to put
243
00:18:37.240 --> 00:18:44.720
all advertising workers out of jobs because
we'll just have advertisements created by the generative
244
00:18:44.759 --> 00:18:49.559
AI systems. And what people have
very quickly discovered is generative AI systems can
245
00:18:49.599 --> 00:18:55.559
do a lot of the tasks that
somebody does in in advertising office, but
246
00:18:55.680 --> 00:18:59.200
not all of them. And there
have been multiple studies that have found exactly
247
00:18:59.240 --> 00:19:03.920
this, that almost every job has
some parts of it that AI should not
248
00:19:03.960 --> 00:19:07.640
be trying to do because AI is
not going to do it well. And
249
00:19:07.720 --> 00:19:11.839
almost every job has some parts that
AI should be doing, and can do
250
00:19:12.000 --> 00:19:15.680
very well. So to an entrepreneur, to a founder, I guess my
251
00:19:15.759 --> 00:19:23.480
advice is, think about the parts
of people's lives that are boring, repetitive,
252
00:19:25.799 --> 00:19:30.000
easy to turn into just a you
know, the system takes care of
253
00:19:30.000 --> 00:19:33.160
that for me, and that's going
to be the sweet spot in terms of
254
00:19:33.279 --> 00:19:37.839
rapid adoption and happy customers. Some
more like a practical versus like a sentiment
255
00:19:38.160 --> 00:19:42.759
right, because it's exactly does a
terrible job with sentiment. It doesn't know
256
00:19:42.799 --> 00:19:45.720
what you're actually trying to do.
You have to be extreme. You know,
257
00:19:45.799 --> 00:19:48.799
it's a machine, right, so
you're saying like they're practical, like
258
00:19:49.039 --> 00:19:55.160
hey, go find this or tell
me what these numbers mean, or tell
259
00:19:55.200 --> 00:19:57.480
me if my numbers are off compared
to these numbers or that sort of stuff.
260
00:19:57.559 --> 00:20:03.400
Is that it exactly? So you
know, plan five itineraries for an
261
00:20:03.440 --> 00:20:07.319
upcoming vacation. Actually, it would
be really good at that picking which of
262
00:20:07.400 --> 00:20:10.200
those itineraries we want to do.
I don't want to delegate that to an
263
00:20:10.200 --> 00:20:15.160
AI system because that's going to depend
on the context and knowing the details of
264
00:20:15.240 --> 00:20:17.640
you know, what it is that
I like to do on vacation, but
265
00:20:17.680 --> 00:20:21.880
also what my wife likes to do
on vacation. That's a pretty important consideration
266
00:20:22.039 --> 00:20:26.319
as well. And so I think
that that's really the key is to understand
267
00:20:26.400 --> 00:20:33.839
that almost nothing can be completely turned
over to AI, but almost everything can
268
00:20:33.880 --> 00:20:40.319
be partially assisted by or augmented by
AI. And finding those those niches,
269
00:20:40.400 --> 00:20:47.160
finding those places for augmentation both within
an organization and in dealing with customers is
270
00:20:47.279 --> 00:20:49.920
really I think, you know,
whoever figures that part out first is going
271
00:20:49.960 --> 00:20:55.039
to be a rapid winner in this
whole AI race. Wow, you know,
272
00:20:55.200 --> 00:20:57.640
I really appreciate your time. I
could talk to you for hours.
273
00:20:59.079 --> 00:21:02.079
You're just fascinating. I really appreciate
your time. I appreciate you joining the
274
00:21:02.119 --> 00:21:04.799
podcast. Thank you so much,
David. Thanks so much for having me
275
00:21:04.839 --> 00:21:08.880
on. Greg And that's it for
another episode of the Startup Science Podcast.
276
00:21:10.480 --> 00:21:12.960
If you can please leave a five
star review on Apple or Spotify, I'd
277
00:21:14.000 --> 00:21:17.640
really appreciate it. And Founders,
as a thank you for leading the changes
278
00:21:17.680 --> 00:21:22.920
in our world, I'm offering free
access to the Startup Science platform Visionaries by
279
00:21:22.960 --> 00:21:26.240
Gregory Shepherd. It's a one stop
shop for founders. It has everything that
280
00:21:26.599 --> 00:21:32.880
startups need, from academy to investors, to grant finders and pitch Finders,
281
00:21:32.920 --> 00:21:37.160
and a whole lot more. Go
to Gregory Shepherd dot com to get visionaries
282
00:21:37.240 --> 00:21:41.240
access, use the code Founders and
all capitals for free access to the platform
283
00:21:41.359 --> 00:21:45.200
and check it out for yourself.
I'll see you next time, and I'm
284
00:21:45.240 --> 00:21:51.359
as stay my friends. This Startup
Science Podcast with Gregory Shepherd is brought to
285
00:21:51.400 --> 00:21:57.200
you by startupscience dot Io. Founders
can log on to startupscience dot io and
286
00:21:57.319 --> 00:22:03.359
use code thorbes to get free access. To connect and find out more about
287
00:22:03.359 --> 00:22:08.839
greg go to Gregory Sheppard dot com. The Startup Science Podcast is a production
288
00:22:10.160 --> 00:22:11.200
of Forbes Books.
1
00:00:01.280 --> 00:00:07.040
Forbes Books presents the Startup Science Podcast
with Gregory Shepherd, brought to you by
2
00:00:07.080 --> 00:00:13.720
Startupscience dot Io. Greg is an
entrepreneur who has built and sold twelve businesses.
3
00:00:14.160 --> 00:00:18.800
He's a recipient of four Private Equity
awards and is a featured ten X
4
00:00:18.800 --> 00:00:24.160
speaker and Forbes author. Here's Gregory
Shepherd. I'm back with David Days,
5
00:00:24.160 --> 00:00:29.039
the professor of Data Science and Philosophy
at UC San Diego. We've been discussing
6
00:00:29.079 --> 00:00:35.679
the intersection of philosophy, cognitive science, and machine learning to tackle complex problems
7
00:00:35.759 --> 00:00:40.560
we see around AI. David,
let's dive into time series models. First,
8
00:00:40.960 --> 00:00:46.119
what is a time series model?
And second, how can entrepreneurs utilize
9
00:00:46.159 --> 00:00:50.359
that for forecasting, planning and deal
me with challenges on the fly? Yeah,
10
00:00:50.359 --> 00:00:56.240
that's a great question. So time
series models are basically models of things
11
00:00:56.240 --> 00:01:00.640
that evolve and unfold over time.
So some dime, I just want to
12
00:01:00.679 --> 00:01:06.040
predict do you have cancer at this
moment based on a radiographic image. That's
13
00:01:06.359 --> 00:01:08.920
not a time series model. I've
got this moment in time, do you
14
00:01:08.959 --> 00:01:14.120
have cancer? But often what I
want to do is predict how might your
15
00:01:14.159 --> 00:01:19.120
cancer unfold over time under various treatment
regimes? So if I give you this
16
00:01:19.200 --> 00:01:23.560
chemotherapy or this surgical intervention or this
radiation treatment, can we figure out how
17
00:01:23.560 --> 00:01:29.879
the cancer is likely to change over
time? That requires us to explicitly model
18
00:01:30.200 --> 00:01:34.640
time. We have to think about
the ways that things change over time,
19
00:01:34.680 --> 00:01:38.000
that the cancer probably isn't going to
change much in one minute, but it
20
00:01:38.040 --> 00:01:42.920
will change over one year, And
so it raises lots of difficult challenges from
21
00:01:42.920 --> 00:01:48.840
a data science perspective in terms of
how we model time. How granular are
22
00:01:48.879 --> 00:01:53.120
we There's a kind of trade off
of if I measure a lot very quickly,
23
00:01:53.359 --> 00:01:57.879
that's great for accuracy, but computationally
inefficient. Measure every once in a
24
00:01:57.879 --> 00:02:01.840
while that'sationally efficient, but I'll get
less accurate. So it's these kinds of
25
00:02:01.879 --> 00:02:07.560
trade offs that we worry about.
Now. For entrepreneurs, the key is
26
00:02:07.640 --> 00:02:13.400
to think about what kinds of sort
of time data you're going to have,
27
00:02:13.680 --> 00:02:17.599
What access are you going to have
to learn about how things unfold over time
28
00:02:19.080 --> 00:02:23.000
in whatever domain or problem you're looking
at, and then figuring out how do
29
00:02:23.039 --> 00:02:28.240
you balance this trade off of you
don't want to remember all of the past
30
00:02:28.599 --> 00:02:32.560
because that's computationally a nightmare, but
you want to remember enough of the past
31
00:02:34.000 --> 00:02:38.560
to inform your future predictions, to
be able to help people plan out and
32
00:02:38.719 --> 00:02:43.639
trace out the possible trajectories of what
lies in front of them. And that
33
00:02:44.240 --> 00:02:47.159
is, you know, there's no
technical answer to that question. That trade
34
00:02:47.199 --> 00:02:52.879
off of how much do I remember? How do I balance accuracy in my
35
00:02:52.960 --> 00:02:58.639
future predictions with the computational cost.
There's no technical solution to that. That's
36
00:02:58.680 --> 00:03:01.639
going to depend on how important is
it to get the future exactly right?
37
00:03:01.680 --> 00:03:07.400
How far out in the future am
I trying to predict? How much do
38
00:03:07.520 --> 00:03:10.919
my customers care about me trying to
hold on to the past data? Do
39
00:03:10.960 --> 00:03:15.599
I have access to really big compute
and storage and all these kinds of things.
40
00:03:15.639 --> 00:03:20.280
So what's important for an entrepreneur,
I think is I mean, you're
41
00:03:20.319 --> 00:03:23.080
going to notice a theme in some
of the things I'm saying. What's important
42
00:03:23.159 --> 00:03:27.080
is that they be intentional, that
they stop and they think to themselves,
43
00:03:27.439 --> 00:03:31.560
Okay, what are we trying to
achieve? What kind of balance is most
44
00:03:31.639 --> 00:03:36.599
important or most valuable? And am
I sure that I'm getting things right from
45
00:03:36.599 --> 00:03:42.400
the perspective of my customers or my
clients. So when you're thinking about these
46
00:03:42.560 --> 00:03:47.599
time series models, how can it
be used to identify capital and emerging market
47
00:03:47.680 --> 00:03:53.280
trends, right, and these customer
behavior patterns, Like it seems like,
48
00:03:53.280 --> 00:04:01.280
you know, the whole buying and
selling of stocks will be incredible disruptive because
49
00:04:01.280 --> 00:04:05.199
of these time series models. Right, So you could predict based on certain
50
00:04:05.240 --> 00:04:11.280
sets of data that are publicly available
outcomes. And then you know, like
51
00:04:11.319 --> 00:04:15.479
I said earlier, like a lot
of the white collar people, I mean,
52
00:04:15.560 --> 00:04:17.279
you know, what do you need
a broker for if the system is
53
00:04:17.519 --> 00:04:21.639
you know, better than an army
of brokers. So what are your thoughts
54
00:04:21.639 --> 00:04:25.639
on that sort of stuff? Yeah, So the first thing to say is
55
00:04:26.040 --> 00:04:30.160
we're already a long way down in
that world. So the vast majority of
56
00:04:30.240 --> 00:04:35.240
trades on stock markets are now done
by automated trading systems that are exactly using
57
00:04:35.279 --> 00:04:43.240
time series models to try to predict
very short run fluctuations in where markets are
58
00:04:43.240 --> 00:04:46.439
going and where particular stocks are going, and to try and take advantage of
59
00:04:46.079 --> 00:04:51.759
very small inefficiencies. Now, what
that highlights is that we're using automation.
60
00:04:51.959 --> 00:04:59.519
We're using AI systems essentially right now
for the very quick decisions. We're not
61
00:04:59.639 --> 00:05:03.279
using as much for the longer term
decisions for where is a sector going what
62
00:05:03.480 --> 00:05:08.240
is a good value stock. We're
still, I think, in a lot
63
00:05:08.279 --> 00:05:14.240
of ways, relying on human expertise, in part because we don't know exactly
64
00:05:14.439 --> 00:05:18.560
which information about the world is most
important, So we're relying on human expertise
65
00:05:18.959 --> 00:05:24.560
to do a lot of that information
integration for us. But you're absolutely right
66
00:05:24.879 --> 00:05:30.720
that it's one of the I would
say holy grails of fintech is how do
67
00:05:30.800 --> 00:05:35.079
we come up with models that are
really good predictors at the let's say,
68
00:05:35.160 --> 00:05:40.279
one week time scale. We've got
models that are very good at the quarterly
69
00:05:40.399 --> 00:05:43.480
predictions. We've got models that are
very good at the one second predictions.
70
00:05:44.079 --> 00:05:46.439
It's actually the middle ground where I
think that there's a lot of uncertainty right
71
00:05:46.480 --> 00:05:50.319
now and a lot of challenges are
arising. One of the issues that comes
72
00:05:50.399 --> 00:05:54.279
up, of course, with the
kinds of time series models you're talking about,
73
00:05:54.399 --> 00:05:59.600
say in financial technologies, is that
you're trying to model the very thing
74
00:05:59.600 --> 00:06:02.120
that you're trying to influence. I
say, there's a sort of self referential
75
00:06:02.199 --> 00:06:06.839
problem here, which is I could
get the model exactly right, but then
76
00:06:06.839 --> 00:06:10.920
I want to do things. I
want to take advantage of inefficiencies in the
77
00:06:10.959 --> 00:06:15.360
market, and that activity the act
of trying to take advantage of the inefficiencies
78
00:06:15.639 --> 00:06:19.959
can actually mean that I'm falsifying my
own model. I'm making my own model
79
00:06:20.000 --> 00:06:25.920
wrong. Oh, inter The other
big challenge is how do we have models
80
00:06:26.000 --> 00:06:30.360
that explicitly include here's the things that
I might do, so that we have
81
00:06:30.399 --> 00:06:35.839
a model that can be a bit
more stable over time. So it seems
82
00:06:35.879 --> 00:06:40.399
that if you're looking at the startup
ecosystem, right, it seems like you
83
00:06:40.399 --> 00:06:44.720
could do you know, you would
have investors in private equity using this right.
84
00:06:44.800 --> 00:06:48.360
So you submit something to the investor
that gives them a bunch of data.
85
00:06:48.439 --> 00:06:54.360
They put it into a series and
then it spits out here's the possible
86
00:06:54.600 --> 00:07:00.560
outcomes or probable outcomes for this particular
business. And I'm not sure if that's
87
00:07:00.600 --> 00:07:05.720
good. You know, because when
you're early on, you know, you're
88
00:07:05.759 --> 00:07:10.160
still trying to figure out your business, and it takes a couple of years
89
00:07:10.199 --> 00:07:15.720
to understand the customer well enough.
And how you should you know, work
90
00:07:15.800 --> 00:07:19.720
through it, because you know,
you can spend years learning about the customer
91
00:07:19.879 --> 00:07:26.680
and build something wrong. So if
an investor is looking at a business and
92
00:07:26.720 --> 00:07:31.040
they're looking at these models, maybe
those models would drastically change if the founder
93
00:07:31.120 --> 00:07:36.079
made a different decision or pivoted or
whatever. So what are the dangers that
94
00:07:36.120 --> 00:07:42.480
you see in how could a founder
utilize it instead of for risk, use
95
00:07:42.519 --> 00:07:46.920
it for opportunity. Yeah, so
I think you're right to raise all of
96
00:07:46.959 --> 00:07:50.920
these kinds of risks, the risk
of sort of the homogenization of startups where
97
00:07:50.920 --> 00:07:57.639
everybody gets directed towards the same thing, the risk of self fulfilling prophecies.
98
00:07:57.680 --> 00:08:01.360
If we do this, then then
we're going to end up in this feedback
99
00:08:01.399 --> 00:08:05.279
loop where we think we're getting it
right, but what we're actually doing is
100
00:08:05.319 --> 00:08:09.360
misreading or misinterpreting what it is that
our customers want. So I think you're
101
00:08:09.480 --> 00:08:15.000
right about that. I think you
also hit on exactly the key idea,
102
00:08:15.040 --> 00:08:20.560
which is you can in some sense
use these models as predictions about everybody else,
103
00:08:20.279 --> 00:08:26.160
and that can help you to see
where are the blank spots in the
104
00:08:26.160 --> 00:08:28.000
map, Where are the parts of
the map that nobody else is going to
105
00:08:28.000 --> 00:08:33.000
be ending up in. So,
and I don't think i'm revealing anything.
106
00:08:33.039 --> 00:08:35.879
Hopefully that isn't too big of a
surprise to some of your listeners, But
107
00:08:37.440 --> 00:08:39.559
a lot of good BC companies are
already doing this. Private equity companies,
108
00:08:39.559 --> 00:08:43.799
they're already doing this. They're trying
to predict where's everybody else going to go,
109
00:08:45.519 --> 00:08:48.639
so that they can figure out where
the marginal advantages might lie, where's
110
00:08:48.679 --> 00:08:54.480
the part of the landscape that is
unexplored, And so I think it's a
111
00:08:54.480 --> 00:09:01.039
great idea. The trick with all
of these things is the quality and quantity
112
00:09:01.080 --> 00:09:05.919
of data that we have in AI. We sort of, you know,
113
00:09:05.919 --> 00:09:09.120
when I look back over the now
over twenty five years that I've been doing
114
00:09:09.120 --> 00:09:16.440
this in AI, we've gone through
cycles where we're constrained by not having good
115
00:09:16.480 --> 00:09:20.360
algorithms. Our algorithms and models aren't
good enough, and then we meet constrained
116
00:09:20.360 --> 00:09:24.240
by the data not being good.
So sometimes we've got algorithms and models that
117
00:09:24.320 --> 00:09:28.039
are really good, but we don't
have good enough data for them. And
118
00:09:28.080 --> 00:09:31.759
then we get good enough data,
and then we realize the shortcomings of the
119
00:09:31.759 --> 00:09:33.360
model, and so we have to
build new models, and we sort of
120
00:09:33.440 --> 00:09:39.080
oscillate back and forth between what's the
sort of rate limiting factor. Is it
121
00:09:39.159 --> 00:09:41.519
the model, the kinds of models
we have, or is it the kinds
122
00:09:41.559 --> 00:09:48.440
of data we have Right now?
We are on the we need better data
123
00:09:48.759 --> 00:09:52.519
part of that cycle. I think
Yeah, the quality is horrific right now,
124
00:09:52.559 --> 00:09:54.799
exactly. And so in fact,
we you know, we know from
125
00:09:54.840 --> 00:10:01.360
public comments that some of the big
tech company their major advances in the last
126
00:10:01.480 --> 00:10:07.000
six to twelve months have not come
because they've gotten better algorithms or because they've
127
00:10:07.000 --> 00:10:11.559
gotten more compute. They've come because
they've gotten much better data, much cleaner
128
00:10:11.639 --> 00:10:15.559
data. They've gone through and had
people label the data and make more sense
129
00:10:15.600 --> 00:10:18.440
of the data. So we really
are at a place right now that if
130
00:10:18.480 --> 00:10:22.600
you want to be able to get
ahead, the best advice I would give
131
00:10:22.600 --> 00:10:28.000
an entrepreneur is figure out if you
have access to data that nobody else has
132
00:10:28.000 --> 00:10:31.399
figured out how to use yet.
That's so, let me ask you about
133
00:10:31.399 --> 00:10:37.440
this causal discovery and what sort of
techniques could founders use to find opportunities,
134
00:10:37.519 --> 00:10:41.879
just like we were talking about the
investors, you know, finding those opportunities.
135
00:10:43.840 --> 00:10:48.320
Yeah, So the basic idea of
causal discovery is a kind of one
136
00:10:48.399 --> 00:10:50.879
that we're all familiar with, which
is we can look out of the world
137
00:10:50.879 --> 00:10:54.799
and we can see things that are
correlated with one another, but that doesn't
138
00:10:54.799 --> 00:10:58.080
mean that they're causal. So if
I want to predict or figure out whether
139
00:10:58.080 --> 00:11:03.039
it's raining outside and Grandam and San
Diego doesn't rain very much. But if
140
00:11:03.039 --> 00:11:05.519
I want to figure that out,
I can look to see if people are
141
00:11:05.519 --> 00:11:09.679
carrying umbrellas. But if I want
to make it rain, it does no
142
00:11:09.799 --> 00:11:13.200
good to tell people to start carrying
umbrellas. That's not an effective strategy to
143
00:11:13.279 --> 00:11:16.519
make it rain, to say,
hey, everybody start carrying an umbrella around.
144
00:11:16.399 --> 00:11:22.480
And so things that are good predictors
are often not good causes. They're
145
00:11:22.480 --> 00:11:26.000
not the things that we can use
to control or shape our world. But
146
00:11:26.120 --> 00:11:31.120
in fact, when we're doing policy, when we're creating new systems, we
147
00:11:31.159 --> 00:11:35.279
often want to control our world.
I don't just want to diagnose that somebody
148
00:11:35.279 --> 00:11:37.440
has cancer. I want to be
able to do something to fix the cancer.
149
00:11:37.440 --> 00:11:41.720
But that means I need causal knowledge. So over the last thirty years,
150
00:11:41.799 --> 00:11:46.080
there's been a whole lot of folks, myself included, who have developed
151
00:11:46.159 --> 00:11:52.519
new algorithms to figure out something about
the way the world works causally, for
152
00:11:52.639 --> 00:11:56.519
all data that we collect, where
we don't intervene on the world, we
153
00:11:56.559 --> 00:12:00.440
don't mess with the world, but
nonetheless we can figure out how the world
154
00:12:00.519 --> 00:12:05.759
works. So the advantage that I
think this holds for entrepreneurs is that many
155
00:12:05.919 --> 00:12:11.759
businesses depend on having some kinds of
causal knowledge. They depend on being able
156
00:12:11.799 --> 00:12:16.799
to do things in the world to
make people's lives easier, to help people
157
00:12:18.080 --> 00:12:24.600
find new kinds of information, new
opportunities, and so we often need causal
158
00:12:24.679 --> 00:12:28.480
knowledge. But most AI systems are
only going to give you good predictions.
159
00:12:28.840 --> 00:12:33.320
They're going to tell you well,
you might try having everyone carry around an
160
00:12:33.399 --> 00:12:37.519
umbrella and see if that creates rain. So there's a real opportunity I think
161
00:12:37.519 --> 00:12:43.879
for entrepreneurs to think carefully about where
and when they might need causal knowledge.
162
00:12:43.000 --> 00:12:46.000
To think about you know, what
does it matter that they're going to be
163
00:12:46.000 --> 00:12:50.120
trying to wiggle something in the world
to bring about a good outcome for their
164
00:12:50.159 --> 00:12:54.279
customer that you know this new ad
campaign, is it going to raise sales?
165
00:12:54.440 --> 00:12:58.080
Well, that requires causal knowledge,
and so I actually think a lot
166
00:12:58.080 --> 00:13:03.600
of entrepreneurs should be taking on board
these cause of discovery algorithms and using them
167
00:13:05.000 --> 00:13:09.879
to be able to better predict what's
going to happen when they make real changes.
168
00:13:09.240 --> 00:13:13.840
And it's not just in terms of
using them for your customers, it's
169
00:13:13.879 --> 00:13:18.000
also using them for yourself. If
we bring in this new market how is
170
00:13:18.039 --> 00:13:22.279
that going to impact our business?
That requires doing causal time series modeling.
171
00:13:22.320 --> 00:13:24.840
I mean, it's putting together the
two different kinds of techniques that we've been
172
00:13:24.879 --> 00:13:31.759
talking about so far. It's hard, but there are some really good introductions
173
00:13:31.840 --> 00:13:35.639
available online and a lot of open
source code, so there's a lot of
174
00:13:37.080 --> 00:13:39.879
code that's out there that makes it
very easy to build these kinds of causal
175
00:13:39.919 --> 00:13:45.960
models and to learn that the data
they might have, so you could use
176
00:13:46.000 --> 00:13:48.440
that then for on a broader basis, you could do that. I did
177
00:13:48.480 --> 00:13:52.360
this huge research on starting in twenty
sixteen, on why, when and how
178
00:13:52.399 --> 00:13:58.200
founders fail, which is what I
wrote the book about, and it was
179
00:13:58.240 --> 00:14:01.720
so manual, right, you know, you're sitting there looking at the transcriptions
180
00:14:01.720 --> 00:14:05.519
from the interviews and looking at case
use cases and all these papers that people
181
00:14:05.600 --> 00:14:11.600
drafted and all these different reports.
It sounds like you could use that exact
182
00:14:11.679 --> 00:14:16.879
system to do that in a fraction
of the time. Just point at a
183
00:14:16.000 --> 00:14:22.159
collect all the data to make sure
it's good data, and then stem the
184
00:14:22.200 --> 00:14:24.639
machine in there to go look at
the data and come out and say,
185
00:14:24.679 --> 00:14:28.120
well, you know, if a
founder does this, they have a ten
186
00:14:28.159 --> 00:14:31.799
percent more likely chance of failing versus
if they do this, they have a
187
00:14:31.799 --> 00:14:33.679
twenty percent chance of being successful.
I mean that sounds like it would that
188
00:14:33.720 --> 00:14:37.799
be a good use of this.
I think that'd be a fantastic use of
189
00:14:37.879 --> 00:14:41.559
this these methods, and I'd be
happy to talk about it with you offline.
190
00:14:41.919 --> 00:14:45.039
Yeah, I mean, I will
say so we you know, one
191
00:14:45.080 --> 00:14:46.440
of the projects that I'm working on
right now, one of the things we're
192
00:14:46.480 --> 00:14:52.200
looking at is trying to understand the
causes of blood pressure crash and surgery.
193
00:14:52.320 --> 00:14:56.559
So drop some blood pressure are a
massive predicted, a very important predictor of
194
00:14:56.759 --> 00:15:03.720
post surgical complications. And so what
we're trying to do is figure out why
195
00:15:03.759 --> 00:15:07.919
do some people have these rapid,
unexpected drops of blood pressure and others don't.
196
00:15:09.399 --> 00:15:11.840
But we don't just want to predict
it, because even if we can
197
00:15:11.879 --> 00:15:16.159
predict it, that leads the anesthesiologists
wondering, Okay, so what do I
198
00:15:16.200 --> 00:15:18.879
do now? Right? They have
a limited set of tools. Which of
199
00:15:18.919 --> 00:15:22.120
their tools should they use? But
it's going to depend on why the blood
200
00:15:22.120 --> 00:15:26.080
pressure drop happens, what the cause
of it is, And so that's what
201
00:15:26.120 --> 00:15:28.720
we're trying to figure out. And
of course we're doing this in a context
202
00:15:28.720 --> 00:15:31.639
where We can't just go in and
do AB testing, right, We can't
203
00:15:31.679 --> 00:15:37.200
just randomly start doing things to people
when they're in surgery, and so these
204
00:15:37.279 --> 00:15:39.840
kinds of methods are invaluable for those. Similarly, you want to be able
205
00:15:39.879 --> 00:15:46.399
to tell founders here are ways to
increase your chances of success that actually are
206
00:15:46.480 --> 00:15:50.159
the causes and drivers of success.
But you don't want to just do a
207
00:15:50.200 --> 00:15:52.440
bunch of AB tests on entrepreneurs.
I think people would be annoyed if you'd
208
00:15:52.440 --> 00:15:58.159
said, here, just try this
and see what happens. Yeah, the
209
00:15:58.159 --> 00:16:03.360
investors wouldn't be happy either. So
let me ask you this too. So
210
00:16:03.480 --> 00:16:07.840
when you look at this, you
know, AI MLO, the whole set,
211
00:16:07.919 --> 00:16:11.799
right, the whole thing that we
categorize as AI, do you feel
212
00:16:11.840 --> 00:16:18.240
like there are some more practical applications
that you could point to that founders that
213
00:16:18.279 --> 00:16:22.519
are listening or investors that are listening
could say, this is a direction we
214
00:16:22.559 --> 00:16:26.440
should look at specifically in AI,
because you know, right now, you
215
00:16:26.440 --> 00:16:30.360
go to webinars, you read papers, all the emails that you get,
216
00:16:30.399 --> 00:16:33.879
everything, and everybody's saying everything is
AI, you know, even though they're
217
00:16:33.879 --> 00:16:37.919
not really AI. And so you
have these founders that are sort of scrambling
218
00:16:37.960 --> 00:16:41.519
saying I need to get into this
AI thing. Where do I point myself?
219
00:16:41.879 --> 00:16:47.240
Do you have any recommendations for what
categories or verticals that they should be
220
00:16:47.279 --> 00:16:52.120
looking at specifically? Well, I
think the most important thing is to think
221
00:16:52.120 --> 00:17:00.600
about what are the tasks that are
often repeated and where you you know what
222
00:17:00.720 --> 00:17:06.680
counts as success, so very know
when you've been successful. It's the kind
223
00:17:06.720 --> 00:17:10.759
of thing that you can build AI
to do. So that's one of the
224
00:17:10.759 --> 00:17:15.839
sort of hallmarks of AI systems is
they really perform best when we know whether
225
00:17:15.880 --> 00:17:22.559
we've succeeded, And so thinking about
the overall goal and workflow and structure of
226
00:17:22.599 --> 00:17:27.839
your organization and asking where are the
pieces that we keep doing this thing over
227
00:17:27.880 --> 00:17:32.680
and over and we know what success
looks like, so we'll know that we've
228
00:17:32.720 --> 00:17:36.319
succeeded when we get there. Those
are the pieces that you want to look
229
00:17:36.359 --> 00:17:40.160
at and say, that's where we
should be using AI. The parts that
230
00:17:40.200 --> 00:17:45.839
require judgment or that are really contact
sensitive or that requires somebody to go you
231
00:17:45.880 --> 00:17:48.400
know, this case is just a
little bit different than that other one.
232
00:17:48.799 --> 00:17:51.720
Those are the places that you want
to make sure to reserve the judgment for
233
00:17:51.759 --> 00:17:56.079
the human in your organization because those
are the parts that AI is still struggling
234
00:17:56.119 --> 00:18:00.000
with. And so you know,
kind of around way, I can come
235
00:18:00.039 --> 00:18:04.680
back to your question, which is
when people ask, well, what vertical
236
00:18:06.000 --> 00:18:10.720
is you know, most right for
disruption by AI or these kinds of things,
237
00:18:10.960 --> 00:18:14.839
my response is all of them and
none of them, because in every
238
00:18:14.920 --> 00:18:19.039
vertical, in every place that people
have looked, some parts of the jobs,
239
00:18:19.200 --> 00:18:26.000
some parts of that sector are clear
repeated, easy to turn relatively speaking,
240
00:18:26.039 --> 00:18:30.240
easy to turn into AI, and
other parts are not. I mean,
241
00:18:30.279 --> 00:18:33.359
we're already seeing this, for example, with generative AI systems, where
242
00:18:33.920 --> 00:18:37.200
there was sort of this idea that
Dolly and chat gpt we're going to put
243
00:18:37.240 --> 00:18:44.720
all advertising workers out of jobs because
we'll just have advertisements created by the generative
244
00:18:44.759 --> 00:18:49.559
AI systems. And what people have
very quickly discovered is generative AI systems can
245
00:18:49.599 --> 00:18:55.559
do a lot of the tasks that
somebody does in in advertising office, but
246
00:18:55.680 --> 00:18:59.200
not all of them. And there
have been multiple studies that have found exactly
247
00:18:59.240 --> 00:19:03.920
this, that almost every job has
some parts of it that AI should not
248
00:19:03.960 --> 00:19:07.640
be trying to do because AI is
not going to do it well. And
249
00:19:07.720 --> 00:19:11.839
almost every job has some parts that
AI should be doing, and can do
250
00:19:12.000 --> 00:19:15.680
very well. So to an entrepreneur, to a founder, I guess my
251
00:19:15.759 --> 00:19:23.480
advice is, think about the parts
of people's lives that are boring, repetitive,
252
00:19:25.799 --> 00:19:30.000
easy to turn into just a you
know, the system takes care of
253
00:19:30.000 --> 00:19:33.160
that for me, and that's going
to be the sweet spot in terms of
254
00:19:33.279 --> 00:19:37.839
rapid adoption and happy customers. Some
more like a practical versus like a sentiment
255
00:19:38.160 --> 00:19:42.759
right, because it's exactly does a
terrible job with sentiment. It doesn't know
256
00:19:42.799 --> 00:19:45.720
what you're actually trying to do.
You have to be extreme. You know,
257
00:19:45.799 --> 00:19:48.799
it's a machine, right, so
you're saying like they're practical, like
258
00:19:49.039 --> 00:19:55.160
hey, go find this or tell
me what these numbers mean, or tell
259
00:19:55.200 --> 00:19:57.480
me if my numbers are off compared
to these numbers or that sort of stuff.
260
00:19:57.559 --> 00:20:03.400
Is that it exactly? So you
know, plan five itineraries for an
261
00:20:03.440 --> 00:20:07.319
upcoming vacation. Actually, it would
be really good at that picking which of
262
00:20:07.400 --> 00:20:10.200
those itineraries we want to do.
I don't want to delegate that to an
263
00:20:10.200 --> 00:20:15.160
AI system because that's going to depend
on the context and knowing the details of
264
00:20:15.240 --> 00:20:17.640
you know, what it is that
I like to do on vacation, but
265
00:20:17.680 --> 00:20:21.880
also what my wife likes to do
on vacation. That's a pretty important consideration
266
00:20:22.039 --> 00:20:26.319
as well. And so I think
that that's really the key is to understand
267
00:20:26.400 --> 00:20:33.839
that almost nothing can be completely turned
over to AI, but almost everything can
268
00:20:33.880 --> 00:20:40.319
be partially assisted by or augmented by
AI. And finding those those niches,
269
00:20:40.400 --> 00:20:47.160
finding those places for augmentation both within
an organization and in dealing with customers is
270
00:20:47.279 --> 00:20:49.920
really I think, you know,
whoever figures that part out first is going
271
00:20:49.960 --> 00:20:55.039
to be a rapid winner in this
whole AI race. Wow, you know,
272
00:20:55.200 --> 00:20:57.640
I really appreciate your time. I
could talk to you for hours.
273
00:20:59.079 --> 00:21:02.079
You're just fascinating. I really appreciate
your time. I appreciate you joining the
274
00:21:02.119 --> 00:21:04.799
podcast. Thank you so much,
David. Thanks so much for having me
275
00:21:04.839 --> 00:21:08.880
on. Greg And that's it for
another episode of the Startup Science Podcast.
276
00:21:10.480 --> 00:21:12.960
If you can please leave a five
star review on Apple or Spotify, I'd
277
00:21:14.000 --> 00:21:17.640
really appreciate it. And Founders,
as a thank you for leading the changes
278
00:21:17.680 --> 00:21:22.920
in our world, I'm offering free
access to the Startup Science platform Visionaries by
279
00:21:22.960 --> 00:21:26.240
Gregory Shepherd. It's a one stop
shop for founders. It has everything that
280
00:21:26.599 --> 00:21:32.880
startups need, from academy to investors, to grant finders and pitch Finders,
281
00:21:32.920 --> 00:21:37.160
and a whole lot more. Go
to Gregory Shepherd dot com to get visionaries
282
00:21:37.240 --> 00:21:41.240
access, use the code Founders and
all capitals for free access to the platform
283
00:21:41.359 --> 00:21:45.200
and check it out for yourself.
I'll see you next time, and I'm
284
00:21:45.240 --> 00:21:51.359
as stay my friends. This Startup
Science Podcast with Gregory Shepherd is brought to
285
00:21:51.400 --> 00:21:57.200
you by startupscience dot Io. Founders
can log on to startupscience dot io and
286
00:21:57.319 --> 00:22:03.359
use code thorbes to get free access. To connect and find out more about
287
00:22:03.359 --> 00:22:08.839
greg go to Gregory Sheppard dot com. The Startup Science Podcast is a production
288
00:22:10.160 --> 00:22:11.200
of Forbes Books.
