Name: Powerful AI tech missing from your content marketing
Uploaded: 2024-06-13T13:22:26.241Z
Duration: 690.24
Description: AI is already used in marketing for things like data collection, automated decision-making and content generation. But can it do more? Christopher Penn, Chief Data Scientist, shares his ideas about how marketers can get the best from AI technology and points out the downsides everyone needs to be aware of.

Chrisp v4 14.12.mp4

Chris Penn v4 14.12.vtt

English

{{00:00:00}}
Marketers, let's face it,
don't have a great track record

{{00:00:03}}
of using things ethically.

{{00:00:05}}
They don't have a great track record
of respecting people's privacy.

{{00:00:09}}
They don't have a great track record
of thinking things through and asking

{{00:00:12}}
the question, what could go wrong?

{{00:00:24}}
Hi, my name is Christopher Penn.

{{00:00:25}}
I'm the co-founder and chief data
scientist at Trustinsights.ai.

{{00:00:35}}
AI has been around for way
longer than people think.

{{00:00:39}}
AI has been around since the 1950s.

{{00:00:41}}
Artificial intelligence as a field is over
70 years old, so people calling it a trend

{{00:00:46}}
today is a little shortsighted, right?

{{00:00:50}}
The idea of having algorithms
that can create new algorithms

{{00:00:54}}
again, dates back to the fifties.

{{00:00:56}}
What has changed in the last 10 to
15 years is computational power.

{{00:01:00}}
We now have the compute scale to
implement this, to actually bring

{{00:01:05}}
these theories into practice.

{{00:01:07}}
Today's computers, like if you have
a new apple mac, they have processors

{{00:01:13}}
now that are literally trillions of
times faster than what was possible

{{00:01:17}}
when these theories were first created.

{{00:01:19}}
What's happening now in terms of
trends, is now we're seeing the

{{00:01:23}}
productisation and the open sourcing
of artificial intelligence in ways

{{00:01:27}}
that have never been done before.

{{00:01:29}}
Just this past week, so we're
recording this in September of 2022,

{{00:01:33}}
just this past week, the stable
diffusion libraries from hugging

{{00:01:37}}
face were open source to the world.

{{00:01:39}}
Anyone can try them.

{{00:01:40}}
And what's amazing is you give it a
text prompt, like an oil painting of

{{00:01:44}}
charmander from Pokemon eating table
grapes at a dining room table, in

{{00:01:48}}
the style of Henri Matisse, and it
will render that painting for you.

{{00:01:53}}
We are seeing image to text where
you give a machine a photo and say,

{{00:01:58}}
okay, write the captions for the
story that are going to be done here.

{{00:02:02}}
We are seeing things like abstractive
text generation, where you give something

{{00:02:07}}
like a transcript from an interview
like this and say, write me a two

{{00:02:09}}
paragraph summary of everything that we
talked about, just the most important

{{00:02:13}}
parts for, maybe our YouTube captions.

{{00:02:15}}
So the technologies themselves have
advanced substantially in the last

{{00:02:19}}
three or four years, but it's an
evolution of 70 years worth of work.

{{00:02:31}}
AI is being used in marketing right
now, it's a bit like taking an F-35 jet

{{00:02:37}}
fighter to the grocery store, right?

{{00:02:39}}
It's using these technologies for mundane
purposes to start because people have

{{00:02:45}}
not fully grasped what is possible.

{{00:02:48}}
So we're seeing a lot of things like text
summarisation, text generation, having

{{00:02:52}}
machines write blog posts, which again is
a bit like sandblasting a soup cracker.

{{00:02:56}}
We're seeing image generation,
but what is coming, and what is

{{00:03:02}}
available now, in academic context
and scientific context, is generation

{{00:03:07}}
of one set of inputs to another with
a technology called transformers.

{{00:03:11}}
This right now is being used
to say, okay, let's make an oil

{{00:03:15}}
painting of Charmander eating table
grapes and turns it to an image.

{{00:03:18}}
Or, let's take this long blog post and
turn it into a short blog post, or maybe

{{00:03:21}}
translate it from one language to another.

{{00:03:24}}
But, transformers as a technology
are agnostic about what you

{{00:03:27}}
put in and what comes out.

{{00:03:29}}
So what we are going to see, probably
in the next three to five years are,

{{00:03:32}}
here's a bunch of inputs, and I type a
phrase, here's the structure of a virus.

{{00:03:37}}
Design me a vaccine for it, right.

{{00:03:40}}
Here is a rare disease.

{{00:03:42}}
Design me a chemical compound
that can address this disease

{{00:03:45}}
with minimal side effects, right?

{{00:03:47}}
So that's what it, what that's
being used for today seems kind

{{00:03:52}}
of silly, but it's a good starting
point because it helps people get

{{00:03:54}}
comfortable with the technology.

{{00:03:57}}
Other things we see AI and machine
learning being used for generally,

{{00:04:00}}
there's two broad buckets.

{{00:04:02}}
There's classification
and there's regression.

{{00:04:04}}
So classification is, I'm gonna give
you a bunch of data and you tell me

{{00:04:08}}
how you would organise it, right?

{{00:04:10}}
So maybe I take a hundred thousand
tweets and say, categorise these.

{{00:04:13}}
How many of them are
about the former queen?

{{00:04:16}}
How many of them are about
marketing automation?

{{00:04:21}}
And it can classify a lot of data
very quickly, and help summarise it.

{{00:04:24}}
And then there's regression, which
is when you have a bunch of numbers

{{00:04:27}}
and you say, okay, here's a target.

{{00:04:29}}
Tell me what of these numbers has
a relationship to this number here.

{{00:04:34}}
So we use this thing for things like
attribution analysis, for marketing mix

{{00:04:38}}
modelling, for anomaly detection, web
analytics, even things like predicting

{{00:04:45}}
what time of year should we send emails.

{{00:04:48}}
So a real simple example, people
Google for Outlook out of office on

{{00:04:54}}
certain times of the year, right?

{{00:04:55}}
And what's about to happen?

{{00:04:57}}
They're about to go on holiday,
they're about to leave.

{{00:04:59}}
They need to figure out
how to turn this thing on.

{{00:05:01}}
So using predictive analytics, you
forecast forward and say, well, when is

{{00:05:06}}
the next time that's gonna peak, right?

{{00:05:08}}
It is September, 2022 right now as
we're recording this, it's going

{{00:05:11}}
to peak next in mid-December.

{{00:05:13}}
Why?

{{00:05:14}}
Because of the holidays,
people are going to leave.

{{00:05:16}}
But, when you can do forecasting
down to the week level, you can say,

{{00:05:19}}
okay, well we're not going to send
emails this week, but also not this

{{00:05:22}}
week either, because nobody's going
to be home, no one's reading email.

{{00:05:26}}
So, those are examples of things
that we're using it for today

{{00:05:29}}
that are very, very effective.

{{00:05:37}}
The best thing about AI that could
be used for marketing, are these

{{00:05:43}}
transformers and diffuser technologies.

{{00:05:45}}
The ability to take any one input
and turn it into another input, as

{{00:05:48}}
long as you are able to tokenise
it, and vectorise it, and embed it

{{00:05:51}}
correctly, that's really complicated.

{{00:05:54}}
A way of just saying, give it a known
set of inputs and a known set of

{{00:05:57}}
outputs, make these two things work
together, make them bond with each other.

{{00:06:02}}
We are seeing some things like
with stable diffusion, being able

{{00:06:06}}
to create photorealistic images.

{{00:06:09}}
One of the things that I've been
sort of yelling at people about

{{00:06:12}}
is, now that we have the ability to
generate realistic looking people,

{{00:06:16}}
say I type in two people sitting at
a conference table talking, right?

{{00:06:20}}
It's a great header image for a blog post.

{{00:06:23}}
There is no excuse anymore for
anyone to have a lack of diversity

{{00:06:28}}
in your marketing, right?

{{00:06:29}}
Because I could just as easily to
say, show me two people, one Indian

{{00:06:34}}
woman and one black man talking
at the conference table, right?

{{00:06:37}}
I don't have to hire actors,
and I can create diversity and

{{00:06:41}}
representation with this technology
that previously was not possible.

{{00:06:46}}
So now, we're talking about changing
our content to reflect the world

{{00:06:50}}
around us without having to change, you
know, dramatically increase our costs,

{{00:06:54}}
which is I think a huge opportunity.

{{00:06:56}}
And brands that don't do this, frankly,
are going to be in a whole world of hurt.

{{00:07:08}}
There are tremendous downsides
to using AI in marketing.

{{00:07:12}}
There are some very frightening ones.

{{00:07:15}}
For folks who market in the
EU, in California, Washington,

{{00:07:20}}
Virginia, the People's Republic
of China, and a few other places.

{{00:07:24}}
There is a set of laws, one, the most
well-known is GDPR, the General Data

{{00:07:28}}
Protection Regulation of the EU, that
says that you may only use data in ways

{{00:07:32}}
that the customer has consented to using.

{{00:07:36}}
When you're using machine learning
and artificial intelligence, if

{{00:07:38}}
you are using customer data to
build models, the customer probably

{{00:07:42}}
has not consented to doing that.

{{00:07:44}}
So if you use that data,
you're in violation of GDPR.

{{00:07:47}}
Even if you think the data is somewhat
anonymised, in June, 2022, Austria,

{{00:07:53}}
Spain, France, Italy, and Denmark.

{{00:07:56}}
No, not Spain, Spain wasn't there.

{{00:07:57}}
Italy and Denmark all had court cases
that basically said, Google Analytics

{{00:08:03}}
data is illegal to use in the EU.

{{00:08:05}}
Four different countries have
ruled that to be the case.

{{00:08:08}}
That that was the previous
version, Universal Analytics.

{{00:08:11}}
So, with artificial intelligence
and machine learning in marketing,

{{00:08:16}}
the challenge for marketers is the
old quote from Jurassic Park, right?

{{00:08:21}}
Your scientists were so preoccupied
with whether they could, that they never

{{00:08:25}}
stopped to think whether they should.

{{00:08:26}}
We have the same problem in marketing
where we're racing ahead to see if

{{00:08:30}}
we can do something, but it doesn't
necessarily mean we should do something.

{{00:08:34}}
All these things are tools.

{{00:08:35}}
They're like blenders, right?

{{00:08:36}}
Or knives or shotguns.

{{00:08:39}}
How they're used depends
on who's using them.

{{00:08:43}}
Marketers, let's face it,
don't have a great track record

{{00:08:46}}
of using things ethically.

{{00:08:48}}
They don't have a great track record
of respecting people's privacy.

{{00:08:52}}
They don't have a great track record of
thinking things through and asking the

{{00:08:55}}
question: what could go wrong, right?

{{00:08:58}}
And when you have technologies
that can literally change

{{00:09:01}}
people's lives, that's a big deal.

{{00:09:04}}
People don't think through.

{{00:09:06}}
And, there's a lot of people who
are creating machine learning models

{{00:09:09}}
who are coders, and very talented
coders, but there's no one tapping

{{00:09:13}}
them on their shoulders saying, let's
talk about ethics and bias guys.

{{00:09:17}}
And that's one of the biggest
dangers right now in machine

{{00:09:20}}
learning, is there's nobody asking
difficult, awkward questions.

{{00:09:32}}
One of the big things that will happen in
the next five years probably is, we will

{{00:09:36}}
need some settled law about AI, because
right now there isn't any, and we don't

{{00:09:42}}
know, for example, who owns the copyright
on an AI piece of artwork, right.

{{00:09:47}}
Particularly if the model was
trained on copyrighted pieces

{{00:09:50}}
of art, which many of them are.

{{00:09:52}}
Who owns that?

{{00:09:53}}
Because there's two different precedents,
there is, you have Microsoft Word and

{{00:09:58}}
you do the typing and you own the output.

{{00:10:01}}
But on the other hand, there are case law
precedents where an elephant or chimpanzee

{{00:10:05}}
painted something and the courts ruled
no, non-humans cannot hold copyrights.

{{00:10:10}}
So we don't know how courts
are going to rule on this.

{{00:10:13}}
And there's two very different outcomes.

{{00:10:15}}
But what has to be settled soon, because
we are seeing a dramatic increase in

{{00:10:21}}
the amount of content, that's one of
the other consequences that marketers

{{00:10:25}}
are going to have to deal with in
the next five years, or now actually.

{{00:10:31}}
These technologies allow for the
creation of so much more content

{{00:10:35}}
than has ever been possible.

{{00:10:37}}
For example, I was playing with
stable diffusion the other night.

{{00:10:40}}
And I gave it a list of 200 prompts
and I said, okay, I'm going to bed.

{{00:10:43}}
You go to work.

{{00:10:43}}
When I woke up the next morning, there
were 3,200 new images that it had created

{{00:10:47}}
overnight, and I went through and sorted
and picked the best 20 that I liked, but

{{00:10:50}}
it created 3,200 new pieces of content
overnight with minimal supervision.

{{00:10:58}}
The sheer amount of content
is going to increase rapidly.

{{00:11:02}}
The amount of good content is going
to increase rapidly, and that means

{{00:11:06}}
that marketers who are not keeping
pace are going to be out of jobs,

{{00:11:10}}
and are going to get zero attention.

{{00:11:13}}
Attention is already the most
scarce resource we have, and

{{00:11:16}}
it's going to get more scarce.

{{00:11:17}}
And if you don't have these capabilities
to keep pace with what the machines

{{00:11:21}}
are doing, you're out of a job.

Marketers, let's face it,
don't have a great track record

They don't have a great track record
of respecting people's privacy.

They don't have a great track record
of thinking things through and asking

I'm the co-founder and chief data
scientist at Trustinsights.ai.

AI has been around for way
longer than people think.

Artificial intelligence as a field is over
70 years old, so people calling it a trend

The idea of having algorithms
that can create new algorithms

What has changed in the last 10 to
15 years is computational power.

We now have the compute scale to
implement this, to actually bring

Today's computers, like if you have
a new apple mac, they have processors

now that are literally trillions of
times faster than what was possible

What's happening now in terms of
trends, is now we're seeing the

productisation and the open sourcing
of artificial intelligence in ways

Just this past week, so we're
recording this in September of 2022,

just this past week, the stable
diffusion libraries from hugging

And what's amazing is you give it a
text prompt, like an oil painting of

charmander from Pokemon eating table
grapes at a dining room table, in

the style of Henri Matisse, and it
will render that painting for you.

We are seeing image to text where
you give a machine a photo and say,

okay, write the captions for the
story that are going to be done here.

We are seeing things like abstractive
text generation, where you give something

like a transcript from an interview
like this and say, write me a two

paragraph summary of everything that we
talked about, just the most important

So the technologies themselves have
advanced substantially in the last

three or four years, but it's an
evolution of 70 years worth of work.

AI is being used in marketing right
now, it's a bit like taking an F-35 jet

It's using these technologies for mundane
purposes to start because people have

So we're seeing a lot of things like text
summarisation, text generation, having

machines write blog posts, which again is
a bit like sandblasting a soup cracker.

We're seeing image generation,
but what is coming, and what is

available now, in academic context
and scientific context, is generation

of one set of inputs to another with
a technology called transformers.

This right now is being used
to say, okay, let's make an oil

painting of Charmander eating table
grapes and turns it to an image.

Or, let's take this long blog post and
turn it into a short blog post, or maybe

translate it from one language to another.

But, transformers as a technology
are agnostic about what you

So what we are going to see, probably
in the next three to five years are,

here's a bunch of inputs, and I type a
phrase, here's the structure of a virus.

Design me a chemical compound
that can address this disease

So that's what it, what that's
being used for today seems kind

of silly, but it's a good starting
point because it helps people get

Other things we see AI and machine
learning being used for generally,

There's classification
and there's regression.

So classification is, I'm gonna give
you a bunch of data and you tell me

So maybe I take a hundred thousand
tweets and say, categorise these.

How many of them are
about the former queen?

How many of them are about
marketing automation?

And it can classify a lot of data
very quickly, and help summarise it.

And then there's regression, which
is when you have a bunch of numbers

Tell me what of these numbers has
a relationship to this number here.

So we use this thing for things like
attribution analysis, for marketing mix

modelling, for anomaly detection, web
analytics, even things like predicting

So a real simple example, people
Google for Outlook out of office on

They're about to go on holiday,
they're about to leave.

They need to figure out
how to turn this thing on.

So using predictive analytics, you
forecast forward and say, well, when is

It is September, 2022 right now as
we're recording this, it's going

Because of the holidays,
people are going to leave.

But, when you can do forecasting
down to the week level, you can say,

okay, well we're not going to send
emails this week, but also not this

week either, because nobody's going
to be home, no one's reading email.

So, those are examples of things
that we're using it for today

The best thing about AI that could
be used for marketing, are these

The ability to take any one input
and turn it into another input, as

long as you are able to tokenise
it, and vectorise it, and embed it

A way of just saying, give it a known
set of inputs and a known set of

outputs, make these two things work
together, make them bond with each other.

We are seeing some things like
with stable diffusion, being able

Expert Insight

AI is already used in marketing for things like data collection, automated decision-making and content generation. But can it do more?
Christopher Penn, Chief Data Scientist, shares his ideas about how marketers can get the best from AI technology and points out the downsides everyone needs to be aware of.

Trends

Tools

AI & ML

thumbnail_Christopher-Penn_CMWorld-2022-copy_491429_1662484410.png

xsmall_Christopher-Penn_CMWorld-2022-copy_491429_1662484410.png

Christopher-Penn_CMWorld-2022-copy_491429_1662484410.png

Christopher Penn

Discover insights from other experts

Chief Data Scientist

Penn explains that AI has actually been around since the 50s. Open source is the main reason it's now such a hot topic.


 Is AI a trend to watch out for in 2023?

Penn discusses how the current uses of AI are quite mundane. But the ability of AI to deal with huge data sets and create new content opens exciting new possibilities. 


How is AI being used in marketing right now?

According to Penn, the ability to create content that more accurately reflects the world around but at very low cost is one huge advantage.  


What's the best thing about using AI in marketing?

Penn shares the challenges of building models with data while protecting people's privacy. 


What are the downsides of using AI in marketing?

Penn stresses the challenge of the issue of ownership when a lot of content is being created rapidly with no copyright. 
