October 9, 2023 • 26 mins
Have you ever wondered about the pivotal phase between AI's potential and its real-world impact? Join us for exciting episodes of Partners in Possibility hosted by CEO of Plus Company Brett Marchand as he’s joined by renowned digital economy expert and bestselling author, Avi Goldfarb!
We're at a crucial phase in AI's development, and Avi takes us through it in a special two part series. Avi and Brett discuss what we’ve witnessed in terms of the potential of AI, but we haven’t yet seen its widespread adoption and effects. Learn about Canada's leadership in AI, the privacy vs. innovation trade-offs, and what it means to live in 'The Between Times.
Tune in now to be ahead of the AI curve! Partners In Possibility, where innovation meets marketing excellence. Listen and subscribe wherever you get your podcasts and stay at the forefront of AI's impact!
We're at a crucial phase in AI's development, and Avi takes us through it in a special two part series. Avi and Brett discuss what we’ve witnessed in terms of the potential of AI, but we haven’t yet seen its widespread adoption and effects. Learn about Canada's leadership in AI, the privacy vs. innovation trade-offs, and what it means to live in 'The Between Times.
Tune in now to be ahead of the AI curve! Partners In Possibility, where innovation meets marketing excellence. Listen and subscribe wherever you get your podcasts and stay at the forefront of AI's impact!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:01):
Chat GPT more people are talking about kids are cheating
on their homework. Well, maybe we need to rethink homework.
If we now, for now and forever, we're going to
have machines that can write. We all see all the
exciting things that this technology can do, but it hasn't
happened yet in most industries.
Speaker 2 (00:24):
In an ever changing world.
Speaker 3 (00:25):
That's all about stay connected, building connections and seeing where
the next collaboration takes a marketing campaign from.
Speaker 2 (00:32):
An initial brief to the follow through.
Speaker 3 (00:34):
What paths are going to make a campaign success more
than a possibility? Hi, I'm Brett marchand CEO of Plus
Company is Partners and Possibility. At this moment in time,
we sit in a striking phase in the development of AI.
We have witnessed its potential, but have not yet seen
(00:55):
its widespread impact. Our guest today, ab Be Goldfarb, calls
this between times. Abby is the Rotman Chair and Artificial
Intelligence and Healthcare at the University of Toronto and Chief
Data Scientist at the Creative Destruction Lab. A renowned expert
on the digital economy, he has testified before the US
Senate and co authored the best selling book Prediction Machines.
(01:18):
The simple economics of artificial intelligence and its follow up
power and prediction, the disruptive economics of artificial intelligence today.
In part one of our conversation, Abby and I talk
about how Canada has led the world in AI development,
some of the trade offs between privacy and innovation, and
what it means to be living in the between times.
(01:45):
I thought, maybe one of the places we would start,
just because of your background, is your work and how
you've tracked the evolution of digital marketing, and maybe touch
a little bit on how this intry compares to other industries.
Speaker 1 (02:02):
So I was a graduate student. I was studying economics
as my PhD in the late nineteen nineties, and I
was looking for an industry to study because that's one
of the things we do. And there was this new
industry called the Internet that no one knew anything about.
And I figured that if I studied something that no
one knew anything about, no matter what I discovered, it
(02:24):
would be new. And so as a graduate student, I said, Okay, well,
let's try to get our heads around a little bit
of the economics of what's going on in the Internet.
And I settled in I worked with a startup at
the time that was out of Carrie, North Carolina, and
worked with what we would now think of as a
(02:46):
the early stage is a big data and looked at
the choices that consumers were making as they surf the
web and the key insights there were. At the time
in the late nineties, the advertising supported internet looked a
lot like the magazine industry in terms of how you
(03:07):
know what people did and how they behaved, and also
in terms of how the advertising worked. And then the
focus over time became on search engine competition, way back
at a time when there was genuine search engine competition.
So we had I had eighteen search engines in my
data set. We call them Internet portals. In the late nineties,
Google was in the data but they were the seventeenth
(03:29):
most popular search engine in the data set.
Speaker 2 (03:31):
Wow.
Speaker 1 (03:32):
And so the world has changed a lot. Yes, the
ad industry has changed a lot, and it the advertising
supported Internet has transformed partly through AI, through what we
might think of as prediction technology and targeting, so that
it doesn't look so much like the old magazine industry anymore.
Speaker 3 (03:52):
Yeah, it's interesting because I think a lot of people
think AI has only really been with us since chat
GPT came along, but in fact it's been around or
a long time, particularly in digital media.
Speaker 1 (04:03):
Absolutely key to this transformation of advertising that we've had
over the past twenty five years has been the ability
to understand the user, to use data to predict what
the user might need and when they need it in
order to get the right ad to the right user
at the right time. And underlying that prediction in many
many cases is in machine learning technology, which is the
(04:27):
core technology underlying today's AI.
Speaker 3 (04:29):
And interestingly enough, I mean, you ended up at U
of T and I don't know if this is a
coincidence or because I know you're from Toronto, but next
door to you was the godfather of AI, right, and
Jeffrey Hinton.
Speaker 1 (04:42):
So I'd love to say it was planned, but it
was completely serendipitous and it turned out to be wonderful.
So I arrived Theater Toronto in two thousand and two,
and for the first decade of my career my focus
was understanding the economic impact to the Internet, inequality on
(05:05):
market outcomes, and especially on the advertising and the rise
of the antech industry, the advertising industry, just measuring the
impact of online ads, understanding the implications for privacy, etc.
Then in twenty twelve at the Rotman School, we started
this program called the Creative Instruction Lab, which was a
(05:25):
program that helped science based startup scale. The program is
still going strong. We're now in thirteen universities around the
world and a hundreds of startups coming through. But in
our very first tiny cohort back in twenty twelve, there
was this company adom Wise, run by a PhD student
out of Hinton's lab in our Compside department, that was
using artificial intelligence to discover new pharmaceuticals. To discover new
(05:48):
drugs and put yourself back eleven years ago, that just
seemed crazy. The idea that you could use an artificial
intelligence technology to discover new molecules and cure disease.
Speaker 3 (06:00):
That was.
Speaker 1 (06:02):
Way beyond the scope of anything I'd heard of at
the time. But it turned out he was onto something
we dug in and we understood when he said it
was artificial intelligence, what he meant was a branch of
machine learning, and machine learning is prediction technology is computational statistics,
so using information you have to generate information you don't have.
He said, Okay, wow, this is really interesting. They were
(06:25):
one of twenty five companies in our cohort, one of
sort of seven to ten that really succeeded that year. Okay,
move on to the next year. The next year we
had two more startups also using AI, and by twenty
fifteen we had this flood of AI startups in our
lab to the point where we added a whole nother
cohort focused on AI that at the time, as far
(06:48):
as we know, we had the most AI startups of
any incubator in the world. And at that point my
co authors and I said, hey, this is something we
need to get our heads around, and so we changed
our career focus from being about understanding the economic impact
on the Internet and how it affected industry and in
(07:08):
my case, the marketing industry in particular, to thinking more
about this new technology AI. So it came out of
our compside department. We sow it through the lab. Honestly,
if I was an ormal professor and not running that lab,
I think what would have happened is I would have
opened the school newspaper and said, hey, isn't it wonderful
if great things are happening our compside department, And that
would have been it. But this combination, because we were
(07:30):
helping these startup scale and we saw the startups in
our lab. That's what led to this change in my
research agenda to really try to get my head around
this technology.
Speaker 3 (07:40):
And I don't think people around the world really understand
how long and how much money had been going against AI,
particularly in Canada, because when it was being abandoned in
most universities in the world, Canadian government leaned into AI
thirty years ago, I guess, and you know obviously what's
(08:03):
happened at UFT with Jeffrey Hinton and all the great
work that's come out of that, but also University or
Montreal Universal Alberta. I mean, most people don't realize that
this was really the center of AI for a long
long time.
Speaker 1 (08:16):
In the seventies and eighties, the core AI technology was
something called expert systems, which you can think of as
a whole bunch of if ND statements to try to
have the computer do what an expert would do. So
you'd have researchers following doctors as they did all sorts
of stuff, and then try to say, okay, well, if
(08:37):
you with a bunch of if N statements, let's try
to imitate the doctor. It turns out the world is
really complex and those systems broadly failed, and so that
led to what we call the AI winter, where around
the world investment in AI dried up.
Speaker 3 (08:55):
That's right, an AI winter. Due to the ineffectiveness of
those early export systems. All around the world people got
off the AI train, except in Canada, and that choice
put Toronto at the forefront of AI for the next
several decades until the rest of the world caught up.
Speaker 1 (09:15):
In Canada. Uh there, the Canadian government decided, you know,
through through luck or strategy is an open question, to
invest in this alternative AI technology, which was instead of
trying to figure out exactly what an expert did, what
you do is you use statistics, use this tool called
(09:39):
machine learning to make a good guess effectively to predict
what the right decision is or what might happen in
a given situation. That technology was co developed well largely
University of Toronto through Professor Hinton's work and later at
(10:00):
you know, also at the University of Montreal. A parallel
technology called reinforcement learning that is a real compliment to
the deep learning technology invented here in Toronto, was invented
Alberta with Rich Sutton and together Canada was you had
this huge lead. Now in the nineteen nineties, the lead
that Canada had was irrelevant to the world because everyone
(10:22):
else had said this technology doesn't work and they'd given up.
But we invested and got better and better, to a
point where it actually, even as of twenty ten, seemed
like a massive waste of effort. Then in twenty twelve,
a team of doctoral students working with Hinton won what's
called the Imaginet competition, so a competition to label pictures.
(10:46):
And they didn't just win, they blew the competition away.
Their technology was just clearly much much better than anything
else out there. And it was at that moment that
the research community and sort of the venture capital community
started paying attention to what was happening here Toronto and
to some extent in Montreal and Alberta. And then by
twenty fifteen the tech world knew we had something and
(11:10):
it was really exciting.
Speaker 3 (11:12):
Somewhere in that journey for you, you also got involved
in privacy, right because I know that you give testimony
expert testimony to the Senate and the US etc.
Speaker 2 (11:23):
How do you end up in front of the Senate.
Speaker 1 (11:25):
I'll make that a short story. It's a long started,
it'll make it a short story. So I was studying
online advertising, and the key advance in online advertising relative
to how advertising worked in the nineties and earlier was
this rise of targeting technology, so we could identify we
could use data to figure out which users might want
(11:45):
what and when. What does targeting technology require requires? You know,
as they said, data, And so our work on that said, hey,
you know what, this data is really useful and if
there's restricts and who's allowed to use what data, that
might affect the ability of our advertisements to work. And
so Catherine Tucker and I put together a research paper
(12:09):
that said, hey, Europe has strict privacy regulations. In the US,
this is before GPR, this is some regulations that came
into to practice in two thousand and four two thousand
and five. Let's see how that affected European advertising relative
to American advertising. It's just a as a research question, Hey,
if you restrict data flows, if you add privacy regulation,
(12:30):
does that affect advertising outcomes? And we found that it
had a massive effect, a negative effect on how ads
worked in the sense that American ads over the course
of that decade got slightly better in terms of their
effectiveness under a few different measures, and European ads got
(12:52):
much much worse, and largely it seemed to be driven
by this regulatory change. That paper led to some attention
and saying, hey, you know, we can measure the costs
of regulating privacy, not just the benefits. That led to
two different ideas that we explored. The first was, okay,
(13:17):
that's the cost side, what about the benefits, Like, there
must be good things about privacy too, and can we
start trying to understand and measure what consumers get out
of privacy? And we did some work on that in
the sense that when we target ads using data in
ways that make it clear that the advertiser has an
(13:40):
intent to manipulate the user, users don't like that.
Speaker 2 (13:44):
Yea, it makes sense.
Speaker 1 (13:45):
And then yeah, no one likes being manipulated. That's you know,
we all know that. And then the other side of
it was that first paper show that there's this trade
off between privacy and innovation because the innovation and advertising
slowed down the European regulation. We had another paper that
showed a trade off between privacy and competition that these
(14:05):
privacy regulations seem to lead to market power for dominant
tech firms. The communication of those trade offs is say, hey,
you know, yes, privacy is really good, and here's the
reasons why privacy is good. But it's not free, and
here are the costs. That's what led to my testimony
at the us M Committee Judiciary.
Speaker 3 (14:27):
You know, and some people say that some of those
privacy regulations in Europe actually put them back significantly when
it comes to you know, digital advertising, et cetera. And
I mean, let's let's be honest, right where if you
look at where all of the economic rent has come
from in this industry, you know, Google, Meta, Microsoft, and
(14:48):
then some in China, as we know, very little of
it's come out to Europe. Do you think that that
was part of the reason.
Speaker 1 (14:54):
I do think that was part of the reason. There's
some excellent research that pretty carefully documents a few things. One,
the regulation, especially GDPR to some extent the earlier regulation,
but especially GDPR, hurt the advertising supported software industry in Europe,
so less venture capital investment in that industry, fewer new
(15:16):
entrants even outside of venture capital, and challenges in the
existing companies to grow the smaller European ones and at
the same time, there's evidence that shows that GDPR seemed
to lead to an increased role of Google in Europe
even relative to the US.
Speaker 2 (15:33):
You've given some great history on this.
Speaker 3 (15:36):
When we're talking about AI in twenty twenty three, what
do we mean, like, how's it evolved?
Speaker 1 (15:43):
When we read about AI in the media and you
hear these petitions we should stop AI. The essence of
those stories is something we see in science fiction, right
that we are on the cusp of the Terminator, these
machines that can do everything we can do, and they
won't listen to us, and it's going to take over
the world and destroy us. Or maybe more optimistically, we're
in the cusp of the Jetsons, which are machines that
(16:05):
can do everything that we can do, but they do
listen to us. Those visions. I don't think those visions
are crazy. But that's not the technology we have today
or in the very near foreseerle future. What we have
today is very very good computational statistics. It's prediction technology.
It takes data, the information you have, and then it
(16:27):
fills in missing information. And those first applications of machine
prediction about a decade ago, commercially were good old fashioned
prediction problems, like you walk into a bank and you
want a loan, and the loan officer asked to predict
whether you're going to pay them back. But as the
prediction technology has gotten better, we've started to realize there's
new applications of machine prediction. It's still machine prediction using
(16:48):
data you have to generate in data you don't have
that we didn't think of before. So at first you
know things like medical diagnosis. That's what your doctor is doing.
They're taking a data about your symptoms. They're filling in
the missing information to the those symptoms. That's still prediction,
recommending what somebody might want you. Recommendation engines that kind
of thing, And it turns out in a last year
(17:09):
we figured out that not just identifying the label for
an image, but creating an image can be solved with prediction,
and writing a paragraph or an essay short essay and
response to a query is also a prediction. What chat
GPT is doing is it's predicting the set of words
(17:30):
that is the best response to a particular query if
you ask it for something that's similar to what's in
its training data in the documents that it's read before.
It's going to give you a useful response, and it's
going to be grammatically correct, and it's going to be
able to do things that many of us find really hard,
like instantly create rhymes, okay, because it has a database
(17:51):
and that's of rhymes, and it can match what a
rhyme might be. Once you recognize its prediction it's not
a sentient being, you'll understand its strengths and weaknesses.
Speaker 2 (18:03):
Wow, we're going to take a quick break.
Speaker 3 (18:06):
When we come back, we'll talk about the similarity between
what's happening today and what happened in the forty years
following Thomas Edison's patent of the light bulb. Welcome back
to Partners and Possibility. I'm Brett marschand we're living during
(18:30):
a phase that Abvi Goldfarb calls between times. We've discovered
and witnessed the potential of AI, but we haven't yet
seen its widespread adoption and effects. And you use this
great analogy about how electricity, even though I think the
patent was in what eighteen seventy nine for the light
(18:50):
bulb by Thomas Edison, how it didn't really have an
impact on industry in the business until much later.
Speaker 1 (18:56):
The eighteen eighties were an incredibly innovative time in electricity.
That was the electric light bulb. Tesla's patent for the
alternating current electric motor was an eighteen ninety. Anybody who
was paying attention in the eighteen eighties could tell that
this technology electricity was going to transform the way we lived,
(19:18):
in the way we worked. Sounds very it sounds very familiar,
but it wasn't until the nineteen twenties. It was another
forty years till half of American households and half of
American factories were electrified. So it took forty years until
this technology that was clearly transformative affected most of us
at home and at work. And so that disconnect is
(19:43):
the underlying challenge. What happened was in the eighteen eighties
it was clear that you know, for example, an electric
motor was really powerful. But what did factory owners do.
They took us steam engine, dropped in the electric motor
at the exact same point and left the workfl the same.
And if you leave the workflow the same, your upside
(20:04):
is limited to however much you end up saving on
energy costs, and you don't save that much an energy
costs you might save five ten fifty percent, but not
much more than that. And so for the fast majority
of factory owners, it wasn't worth the effort of taking
out their steam engine, dropping the electric motor, figuring out
how to get electricity distributed into the factory through wires,
(20:25):
laying out the wires throughout the factory in a way
that was fire safe and effective. It wasn't worth all
that investment to just do what you always did a
little bit better, right, And so even by nineteen hundred,
less than five percent of factories were electrified.
Speaker 2 (20:42):
And a long comes Henry Ford, and along.
Speaker 1 (20:44):
Comes Henry Ford and some others, and they realize that
electricity isn't just cheap power, right. Electricity allows you to
do things differently. And so in the eighteen nineties, the
factory would have been organized around the steam engine. The
most power hungry machines would have been as close as
possible to the steam engine. Because I don't know if
you remember your high school physics, but energy dissipates with distance,
(21:05):
and so you want to create your power, make sure
your power your machines are close to your steam engine,
to your power source, and so and that meant you
didn't move things around your factory that much. You had
pretty skilled artisans doing a lot of the work, and
each person did a lot of different steps in the
production process. Electricity came along and Henry Ford and these
(21:25):
others realized, we can now distribute our machines around our factory.
And so what you might think of as the quintessential
twentieth century factory with inputs in one end and outputs
out the other, with modular production, that is a direct
result of the change that electricity enabled. That's when we
see a change in the trajectory of adoption and sort
of wrap an adoption over the next decade or two.
(21:47):
Right with AI, it feels like we're in the eighteen nineties, right.
We all see all the exciting things this technology can do,
but it hasn't happened yet in most industries, and a
lot of the applications we're imagining are let's do what
we always did, but a little bit better. And that
(22:08):
lack of imagination is natural. That's of course where we
would be, but it means that we haven't figured out
what the organization of the future looks like. And that's
what we mean by this concept of the between times.
We can see the future, but it hasn't happened yet
chat GPT more people are talking about, oh, kids are
cheating on their homework, then well maybe we need to
rethink homework. If we now, for now and forever, we're
(22:31):
going to have machines that can write right. And so
those challenges are fundamental and they need to be overcome,
and they are being overcome. But it's not straightforward because
we have to be much more ambitious with what we're
trying to do with technology.
Speaker 3 (22:48):
Right because people are looking at AIS, how do I
save fifteen or twenty percent of my legal cost because
contracts can be done more effectively and many other applications.
But you're right, I'm not sure anyone's really thinking about
it and how it can fundamentally change a business and
create an evolution a revolution, right, because I mean the
industrial revolution came out of the modern factory. And what
(23:12):
kind of opportunities do you see in front of us
as far as industries go, where businesses go in using
AI to be more transformative?
Speaker 1 (23:18):
As you say, okay, I see them, see them everywhere.
So what's already been transformed by AI? I mean we'll
start there there's two industries that have already seen some transformation.
One is the taxi industry. And so if you look
at the first applications of GPS systems to help predict
(23:39):
how to get from point A to point B of
digital maps with prediction navigational predictions, were Hey, let's look
at professional drivers like truck drivers and taxi drivers and
make them drive five percent more efficient. They are the
ones who are going to get the most benefit because
they spend most time driving, right, And that happens so
like two thousand and eight to twenty twelve, a lot
(24:01):
of the application and a lot of the benefit for
these technologies for GPS systems was for professional drivers. Then
you know, Uber and Lyft and others realized that's crazy, right,
This technology combined with digital dispatch means that anybody can
be as good as a professional driver. So what do
we need to limit the number of people who can
(24:22):
professional drive professionally drive to these people who have licenses
and all these other steps, and that, you know, change
that personal transportation industry is a direct result of digital navigation.
So navigational predictions and handful of other tools that led
to a transformation. The other industry of course is the
(24:44):
advertising industry in order to tell you, but advertising in
the nineteen nineties and earlier was not a data intensive industry.
There was a lot of gut feel and a lot
of intuition. Yeah, we had some demographics, but it wasn't that, Okay,
we want to know exactly here is an individual user.
We want to know what they want at that moment,
(25:06):
and we're going to serve an ad for them based
on their needs. And then along came targeting technology, Along
came transformation and data and some machine learning tools, and
the industry went from these rate cards in magazines and
even on Yahoo and elsewhere, to real time auctions based
on these predictions to price ads and lead to an
(25:29):
entirely new ad tech industry that is a direct result
of what's happened with prediction technology.
Speaker 3 (25:39):
Thank you for listening to Partners and Possibility. That was
part one of my conversation with Avid Goldfarb. Next week,
we'll continue our conversation about AI and prediction technology, the
vital role that human judgment plays in maximizing the efficiency
of the technology, and how it will change our industry forever.
We'll also talk about how we at Plus Company are
working to connect various touch points while continuously measuring creative performance.
(26:03):
To grow business with our newest innovation AOS, stay tuned
Chat GPT more people are talking about kids are cheating
on their homework. Well, maybe we need to rethink homework.
If we now, for now and forever, we're going to
have machines that can write. We all see all the
exciting things that this technology can do, but it hasn't
happened yet in most industries.
Speaker 2 (00:24):
In an ever changing world.
Speaker 3 (00:25):
That's all about stay connected, building connections and seeing where
the next collaboration takes a marketing campaign from.
Speaker 2 (00:32):
An initial brief to the follow through.
Speaker 3 (00:34):
What paths are going to make a campaign success more
than a possibility? Hi, I'm Brett marchand CEO of Plus
Company is Partners and Possibility. At this moment in time,
we sit in a striking phase in the development of AI.
We have witnessed its potential, but have not yet seen
(00:55):
its widespread impact. Our guest today, ab Be Goldfarb, calls
this between times. Abby is the Rotman Chair and Artificial
Intelligence and Healthcare at the University of Toronto and Chief
Data Scientist at the Creative Destruction Lab. A renowned expert
on the digital economy, he has testified before the US
Senate and co authored the best selling book Prediction Machines.
(01:18):
The simple economics of artificial intelligence and its follow up
power and prediction, the disruptive economics of artificial intelligence today.
In part one of our conversation, Abby and I talk
about how Canada has led the world in AI development,
some of the trade offs between privacy and innovation, and
what it means to be living in the between times.
(01:45):
I thought, maybe one of the places we would start,
just because of your background, is your work and how
you've tracked the evolution of digital marketing, and maybe touch
a little bit on how this intry compares to other industries.
Speaker 1 (02:02):
So I was a graduate student. I was studying economics
as my PhD in the late nineteen nineties, and I
was looking for an industry to study because that's one
of the things we do. And there was this new
industry called the Internet that no one knew anything about.
And I figured that if I studied something that no
one knew anything about, no matter what I discovered, it
(02:24):
would be new. And so as a graduate student, I said, Okay, well,
let's try to get our heads around a little bit
of the economics of what's going on in the Internet.
And I settled in I worked with a startup at
the time that was out of Carrie, North Carolina, and
worked with what we would now think of as a
(02:46):
the early stage is a big data and looked at
the choices that consumers were making as they surf the
web and the key insights there were. At the time
in the late nineties, the advertising supported internet looked a
lot like the magazine industry in terms of how you
(03:07):
know what people did and how they behaved, and also
in terms of how the advertising worked. And then the
focus over time became on search engine competition, way back
at a time when there was genuine search engine competition.
So we had I had eighteen search engines in my
data set. We call them Internet portals. In the late nineties,
Google was in the data but they were the seventeenth
(03:29):
most popular search engine in the data set.
Speaker 2 (03:31):
Wow.
Speaker 1 (03:32):
And so the world has changed a lot. Yes, the
ad industry has changed a lot, and it the advertising
supported Internet has transformed partly through AI, through what we
might think of as prediction technology and targeting, so that
it doesn't look so much like the old magazine industry anymore.
Speaker 3 (03:52):
Yeah, it's interesting because I think a lot of people
think AI has only really been with us since chat
GPT came along, but in fact it's been around or
a long time, particularly in digital media.
Speaker 1 (04:03):
Absolutely key to this transformation of advertising that we've had
over the past twenty five years has been the ability
to understand the user, to use data to predict what
the user might need and when they need it in
order to get the right ad to the right user
at the right time. And underlying that prediction in many
many cases is in machine learning technology, which is the
(04:27):
core technology underlying today's AI.
Speaker 3 (04:29):
And interestingly enough, I mean, you ended up at U
of T and I don't know if this is a
coincidence or because I know you're from Toronto, but next
door to you was the godfather of AI, right, and
Jeffrey Hinton.
Speaker 1 (04:42):
So I'd love to say it was planned, but it
was completely serendipitous and it turned out to be wonderful.
So I arrived Theater Toronto in two thousand and two,
and for the first decade of my career my focus
was understanding the economic impact to the Internet, inequality on
(05:05):
market outcomes, and especially on the advertising and the rise
of the antech industry, the advertising industry, just measuring the
impact of online ads, understanding the implications for privacy, etc.
Then in twenty twelve at the Rotman School, we started
this program called the Creative Instruction Lab, which was a
(05:25):
program that helped science based startup scale. The program is
still going strong. We're now in thirteen universities around the
world and a hundreds of startups coming through. But in
our very first tiny cohort back in twenty twelve, there
was this company adom Wise, run by a PhD student
out of Hinton's lab in our Compside department, that was
using artificial intelligence to discover new pharmaceuticals. To discover new
(05:48):
drugs and put yourself back eleven years ago, that just
seemed crazy. The idea that you could use an artificial
intelligence technology to discover new molecules and cure disease.
Speaker 3 (06:00):
That was.
Speaker 1 (06:02):
Way beyond the scope of anything I'd heard of at
the time. But it turned out he was onto something
we dug in and we understood when he said it
was artificial intelligence, what he meant was a branch of
machine learning, and machine learning is prediction technology is computational statistics,
so using information you have to generate information you don't have.
He said, Okay, wow, this is really interesting. They were
(06:25):
one of twenty five companies in our cohort, one of
sort of seven to ten that really succeeded that year. Okay,
move on to the next year. The next year we
had two more startups also using AI, and by twenty
fifteen we had this flood of AI startups in our
lab to the point where we added a whole nother
cohort focused on AI that at the time, as far
(06:48):
as we know, we had the most AI startups of
any incubator in the world. And at that point my
co authors and I said, hey, this is something we
need to get our heads around, and so we changed
our career focus from being about understanding the economic impact
on the Internet and how it affected industry and in
(07:08):
my case, the marketing industry in particular, to thinking more
about this new technology AI. So it came out of
our compside department. We sow it through the lab. Honestly,
if I was an ormal professor and not running that lab,
I think what would have happened is I would have
opened the school newspaper and said, hey, isn't it wonderful
if great things are happening our compside department, And that
would have been it. But this combination, because we were
(07:30):
helping these startup scale and we saw the startups in
our lab. That's what led to this change in my
research agenda to really try to get my head around
this technology.
Speaker 3 (07:40):
And I don't think people around the world really understand
how long and how much money had been going against AI,
particularly in Canada, because when it was being abandoned in
most universities in the world, Canadian government leaned into AI
thirty years ago, I guess, and you know obviously what's
(08:03):
happened at UFT with Jeffrey Hinton and all the great
work that's come out of that, but also University or
Montreal Universal Alberta. I mean, most people don't realize that
this was really the center of AI for a long
long time.
Speaker 1 (08:16):
In the seventies and eighties, the core AI technology was
something called expert systems, which you can think of as
a whole bunch of if ND statements to try to
have the computer do what an expert would do. So
you'd have researchers following doctors as they did all sorts
of stuff, and then try to say, okay, well, if
(08:37):
you with a bunch of if N statements, let's try
to imitate the doctor. It turns out the world is
really complex and those systems broadly failed, and so that
led to what we call the AI winter, where around
the world investment in AI dried up.
Speaker 3 (08:55):
That's right, an AI winter. Due to the ineffectiveness of
those early export systems. All around the world people got
off the AI train, except in Canada, and that choice
put Toronto at the forefront of AI for the next
several decades until the rest of the world caught up.
Speaker 1 (09:15):
In Canada. Uh there, the Canadian government decided, you know,
through through luck or strategy is an open question, to
invest in this alternative AI technology, which was instead of
trying to figure out exactly what an expert did, what
you do is you use statistics, use this tool called
(09:39):
machine learning to make a good guess effectively to predict
what the right decision is or what might happen in
a given situation. That technology was co developed well largely
University of Toronto through Professor Hinton's work and later at
(10:00):
you know, also at the University of Montreal. A parallel
technology called reinforcement learning that is a real compliment to
the deep learning technology invented here in Toronto, was invented
Alberta with Rich Sutton and together Canada was you had
this huge lead. Now in the nineteen nineties, the lead
that Canada had was irrelevant to the world because everyone
(10:22):
else had said this technology doesn't work and they'd given up.
But we invested and got better and better, to a
point where it actually, even as of twenty ten, seemed
like a massive waste of effort. Then in twenty twelve,
a team of doctoral students working with Hinton won what's
called the Imaginet competition, so a competition to label pictures.
(10:46):
And they didn't just win, they blew the competition away.
Their technology was just clearly much much better than anything
else out there. And it was at that moment that
the research community and sort of the venture capital community
started paying attention to what was happening here Toronto and
to some extent in Montreal and Alberta. And then by
twenty fifteen the tech world knew we had something and
(11:10):
it was really exciting.
Speaker 3 (11:12):
Somewhere in that journey for you, you also got involved
in privacy, right because I know that you give testimony
expert testimony to the Senate and the US etc.
Speaker 2 (11:23):
How do you end up in front of the Senate.
Speaker 1 (11:25):
I'll make that a short story. It's a long started,
it'll make it a short story. So I was studying
online advertising, and the key advance in online advertising relative
to how advertising worked in the nineties and earlier was
this rise of targeting technology, so we could identify we
could use data to figure out which users might want
(11:45):
what and when. What does targeting technology require requires? You know,
as they said, data, And so our work on that said, hey,
you know what, this data is really useful and if
there's restricts and who's allowed to use what data, that
might affect the ability of our advertisements to work. And
so Catherine Tucker and I put together a research paper
(12:09):
that said, hey, Europe has strict privacy regulations. In the US,
this is before GPR, this is some regulations that came
into to practice in two thousand and four two thousand
and five. Let's see how that affected European advertising relative
to American advertising. It's just a as a research question, Hey,
if you restrict data flows, if you add privacy regulation,
(12:30):
does that affect advertising outcomes? And we found that it
had a massive effect, a negative effect on how ads
worked in the sense that American ads over the course
of that decade got slightly better in terms of their
effectiveness under a few different measures, and European ads got
(12:52):
much much worse, and largely it seemed to be driven
by this regulatory change. That paper led to some attention
and saying, hey, you know, we can measure the costs
of regulating privacy, not just the benefits. That led to
two different ideas that we explored. The first was, okay,
(13:17):
that's the cost side, what about the benefits, Like, there
must be good things about privacy too, and can we
start trying to understand and measure what consumers get out
of privacy? And we did some work on that in
the sense that when we target ads using data in
ways that make it clear that the advertiser has an
(13:40):
intent to manipulate the user, users don't like that.
Speaker 2 (13:44):
Yea, it makes sense.
Speaker 1 (13:45):
And then yeah, no one likes being manipulated. That's you know,
we all know that. And then the other side of
it was that first paper show that there's this trade
off between privacy and innovation because the innovation and advertising
slowed down the European regulation. We had another paper that
showed a trade off between privacy and competition that these
(14:05):
privacy regulations seem to lead to market power for dominant
tech firms. The communication of those trade offs is say, hey,
you know, yes, privacy is really good, and here's the
reasons why privacy is good. But it's not free, and
here are the costs. That's what led to my testimony
at the us M Committee Judiciary.
Speaker 3 (14:27):
You know, and some people say that some of those
privacy regulations in Europe actually put them back significantly when
it comes to you know, digital advertising, et cetera. And
I mean, let's let's be honest, right where if you
look at where all of the economic rent has come
from in this industry, you know, Google, Meta, Microsoft, and
(14:48):
then some in China, as we know, very little of
it's come out to Europe. Do you think that that
was part of the reason.
Speaker 1 (14:54):
I do think that was part of the reason. There's
some excellent research that pretty carefully documents a few things. One,
the regulation, especially GDPR to some extent the earlier regulation,
but especially GDPR, hurt the advertising supported software industry in Europe,
so less venture capital investment in that industry, fewer new
(15:16):
entrants even outside of venture capital, and challenges in the
existing companies to grow the smaller European ones and at
the same time, there's evidence that shows that GDPR seemed
to lead to an increased role of Google in Europe
even relative to the US.
Speaker 2 (15:33):
You've given some great history on this.
Speaker 3 (15:36):
When we're talking about AI in twenty twenty three, what
do we mean, like, how's it evolved?
Speaker 1 (15:43):
When we read about AI in the media and you
hear these petitions we should stop AI. The essence of
those stories is something we see in science fiction, right
that we are on the cusp of the Terminator, these
machines that can do everything we can do, and they
won't listen to us, and it's going to take over
the world and destroy us. Or maybe more optimistically, we're
in the cusp of the Jetsons, which are machines that
(16:05):
can do everything that we can do, but they do
listen to us. Those visions. I don't think those visions
are crazy. But that's not the technology we have today
or in the very near foreseerle future. What we have
today is very very good computational statistics. It's prediction technology.
It takes data, the information you have, and then it
(16:27):
fills in missing information. And those first applications of machine
prediction about a decade ago, commercially were good old fashioned
prediction problems, like you walk into a bank and you
want a loan, and the loan officer asked to predict
whether you're going to pay them back. But as the
prediction technology has gotten better, we've started to realize there's
new applications of machine prediction. It's still machine prediction using
(16:48):
data you have to generate in data you don't have
that we didn't think of before. So at first you
know things like medical diagnosis. That's what your doctor is doing.
They're taking a data about your symptoms. They're filling in
the missing information to the those symptoms. That's still prediction,
recommending what somebody might want you. Recommendation engines that kind
of thing, And it turns out in a last year
(17:09):
we figured out that not just identifying the label for
an image, but creating an image can be solved with prediction,
and writing a paragraph or an essay short essay and
response to a query is also a prediction. What chat
GPT is doing is it's predicting the set of words
(17:30):
that is the best response to a particular query if
you ask it for something that's similar to what's in
its training data in the documents that it's read before.
It's going to give you a useful response, and it's
going to be grammatically correct, and it's going to be
able to do things that many of us find really hard,
like instantly create rhymes, okay, because it has a database
(17:51):
and that's of rhymes, and it can match what a
rhyme might be. Once you recognize its prediction it's not
a sentient being, you'll understand its strengths and weaknesses.
Speaker 2 (18:03):
Wow, we're going to take a quick break.
Speaker 3 (18:06):
When we come back, we'll talk about the similarity between
what's happening today and what happened in the forty years
following Thomas Edison's patent of the light bulb. Welcome back
to Partners and Possibility. I'm Brett marschand we're living during
(18:30):
a phase that Abvi Goldfarb calls between times. We've discovered
and witnessed the potential of AI, but we haven't yet
seen its widespread adoption and effects. And you use this
great analogy about how electricity, even though I think the
patent was in what eighteen seventy nine for the light
(18:50):
bulb by Thomas Edison, how it didn't really have an
impact on industry in the business until much later.
Speaker 1 (18:56):
The eighteen eighties were an incredibly innovative time in electricity.
That was the electric light bulb. Tesla's patent for the
alternating current electric motor was an eighteen ninety. Anybody who
was paying attention in the eighteen eighties could tell that
this technology electricity was going to transform the way we lived,
(19:18):
in the way we worked. Sounds very it sounds very familiar,
but it wasn't until the nineteen twenties. It was another
forty years till half of American households and half of
American factories were electrified. So it took forty years until
this technology that was clearly transformative affected most of us
at home and at work. And so that disconnect is
(19:43):
the underlying challenge. What happened was in the eighteen eighties
it was clear that you know, for example, an electric
motor was really powerful. But what did factory owners do.
They took us steam engine, dropped in the electric motor
at the exact same point and left the workfl the same.
And if you leave the workflow the same, your upside
(20:04):
is limited to however much you end up saving on
energy costs, and you don't save that much an energy
costs you might save five ten fifty percent, but not
much more than that. And so for the fast majority
of factory owners, it wasn't worth the effort of taking
out their steam engine, dropping the electric motor, figuring out
how to get electricity distributed into the factory through wires,
(20:25):
laying out the wires throughout the factory in a way
that was fire safe and effective. It wasn't worth all
that investment to just do what you always did a
little bit better, right, And so even by nineteen hundred,
less than five percent of factories were electrified.
Speaker 2 (20:42):
And a long comes Henry Ford, and along.
Speaker 1 (20:44):
Comes Henry Ford and some others, and they realize that
electricity isn't just cheap power, right. Electricity allows you to
do things differently. And so in the eighteen nineties, the
factory would have been organized around the steam engine. The
most power hungry machines would have been as close as
possible to the steam engine. Because I don't know if
you remember your high school physics, but energy dissipates with distance,
(21:05):
and so you want to create your power, make sure
your power your machines are close to your steam engine,
to your power source, and so and that meant you
didn't move things around your factory that much. You had
pretty skilled artisans doing a lot of the work, and
each person did a lot of different steps in the
production process. Electricity came along and Henry Ford and these
(21:25):
others realized, we can now distribute our machines around our factory.
And so what you might think of as the quintessential
twentieth century factory with inputs in one end and outputs
out the other, with modular production, that is a direct
result of the change that electricity enabled. That's when we
see a change in the trajectory of adoption and sort
of wrap an adoption over the next decade or two.
(21:47):
Right with AI, it feels like we're in the eighteen nineties, right.
We all see all the exciting things this technology can do,
but it hasn't happened yet in most industries, and a
lot of the applications we're imagining are let's do what
we always did, but a little bit better. And that
(22:08):
lack of imagination is natural. That's of course where we
would be, but it means that we haven't figured out
what the organization of the future looks like. And that's
what we mean by this concept of the between times.
We can see the future, but it hasn't happened yet
chat GPT more people are talking about, oh, kids are
cheating on their homework, then well maybe we need to
rethink homework. If we now, for now and forever, we're
(22:31):
going to have machines that can write right. And so
those challenges are fundamental and they need to be overcome,
and they are being overcome. But it's not straightforward because
we have to be much more ambitious with what we're
trying to do with technology.
Speaker 3 (22:48):
Right because people are looking at AIS, how do I
save fifteen or twenty percent of my legal cost because
contracts can be done more effectively and many other applications.
But you're right, I'm not sure anyone's really thinking about
it and how it can fundamentally change a business and
create an evolution a revolution, right, because I mean the
industrial revolution came out of the modern factory. And what
(23:12):
kind of opportunities do you see in front of us
as far as industries go, where businesses go in using
AI to be more transformative?
Speaker 1 (23:18):
As you say, okay, I see them, see them everywhere.
So what's already been transformed by AI? I mean we'll
start there there's two industries that have already seen some transformation.
One is the taxi industry. And so if you look
at the first applications of GPS systems to help predict
(23:39):
how to get from point A to point B of
digital maps with prediction navigational predictions, were Hey, let's look
at professional drivers like truck drivers and taxi drivers and
make them drive five percent more efficient. They are the
ones who are going to get the most benefit because
they spend most time driving, right, And that happens so
like two thousand and eight to twenty twelve, a lot
(24:01):
of the application and a lot of the benefit for
these technologies for GPS systems was for professional drivers. Then
you know, Uber and Lyft and others realized that's crazy, right,
This technology combined with digital dispatch means that anybody can
be as good as a professional driver. So what do
we need to limit the number of people who can
(24:22):
professional drive professionally drive to these people who have licenses
and all these other steps, and that, you know, change
that personal transportation industry is a direct result of digital navigation.
So navigational predictions and handful of other tools that led
to a transformation. The other industry of course is the
(24:44):
advertising industry in order to tell you, but advertising in
the nineteen nineties and earlier was not a data intensive industry.
There was a lot of gut feel and a lot
of intuition. Yeah, we had some demographics, but it wasn't that, Okay,
we want to know exactly here is an individual user.
We want to know what they want at that moment,
(25:06):
and we're going to serve an ad for them based
on their needs. And then along came targeting technology, Along
came transformation and data and some machine learning tools, and
the industry went from these rate cards in magazines and
even on Yahoo and elsewhere, to real time auctions based
on these predictions to price ads and lead to an
(25:29):
entirely new ad tech industry that is a direct result
of what's happened with prediction technology.
Speaker 3 (25:39):
Thank you for listening to Partners and Possibility. That was
part one of my conversation with Avid Goldfarb. Next week,
we'll continue our conversation about AI and prediction technology, the
vital role that human judgment plays in maximizing the efficiency
of the technology, and how it will change our industry forever.
We'll also talk about how we at Plus Company are
working to connect various touch points while continuously measuring creative performance.
(26:03):
To grow business with our newest innovation AOS, stay tuned
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
My Favorite Murder with Karen Kilgariff and Georgia Hardstark
My Favorite Murder is a true crime comedy podcast hosted by Karen Kilgariff and Georgia Hardstark. Each week, Karen and Georgia share compelling true crimes and hometown stories from friends and listeners. Since MFM launched in January of 2016, Karen and Georgia have shared their lifelong interest in true crime and have covered stories of infamous serial killers like the Night Stalker, mysterious cold cases, captivating cults, incredible survivor stories and important events from history like the Tulsa race massacre of 1921. My Favorite Murder is part of the Exactly Right podcast network that provides a platform for bold, creative voices to bring to life provocative, entertaining and relatable stories for audiences everywhere. The Exactly Right roster of podcasts covers a variety of topics including historic true crime, comedic interviews and news, science, pop culture and more. Podcasts on the network include Buried Bones with Kate Winkler Dawson and Paul Holes, That's Messed Up: An SVU Podcast, This Podcast Will Kill You, Bananas and more.