03 Don Norman: Human(ity) Centred AI – Pt.2
squadcaster-dhaf_1_08-07-2024_092412: I
don't know the solution is Well, there's
an old saying, it comes from the arms
race, in the United States at least,
which is, you make a, when you make
a treaty with some other nation about
arms, You should trust it, but verify.
kevin_1_08-07-2024_172413: verify.
squadcaster-dhaf_1_08-07-2024_092412:
trust what you're saying, but
you know, I don't really trust,
so, I want to verify and I want
evidence for what you're saying.
that's what we need in
AI, is trust but verify.
Me: This is the second half of my
conversation with Don Norman, where we
focus on user experience in the age of AI.
We cover a range of topics, including
AI devices, human versus machine
strengths, and the need to adopt a
trust but verify stance towards AI.
I hope you enjoy it.
squadcaster-dhaf_1_08-07-2024_092412:
Here's the difference though, and
here's where the revolution today
in what's called chats, or the, um,
the generalized pre
programmed transformer.
For the first time, we now have a system
that knows a tremendous amount of things,
and we work with it by talking to it,
by having a conversation as if you know,
I'm talking to you, and I say Well, you
know, I ask you a question and you give
me an answer, and I say, that isn't what
I meant, and we talk back and forth.
that's what we're going to be doing with
our machines, and that will make a lot
of the complexity that we worried about,
remembering what every control does, or
remembering how I should do something,
I don't need to know that anymore.
I just tell the system, no, you
didn't do it right, do it again.
In fact, I'm writing, I'm giving
a commencement addressing on
Sunday just a few days away.
I wrote it and I thought
but it's much too long.
And so I asked, I, I'm in Microsoft
Word and I just said, Would you
summarize this, make it shorter?
I have a, I have, I don't want the talk
to be more than 20 words, 20 minutes.
So why don't you aim for 15 minutes?
And I said, okay.
And it did it.
And I look it over and I say, oh,
well, you cut out some of the stuff
I thought was important, that's okay.
it's because what we're
doing it's my assistant.
I'm then still in charge.
And it's like when I asked an
assistant, would you please
summarize this and make it shorter?
And I look at it and I say, well.
you didn't do it quite right.
I can either say, here's what I want
you to do, or I can say, okay, that's
good enough that now it's easy for me
to make it the way I want it to be.
And that's going to happen more and more.
I We're writing all these documents for
my charity and trying to send them out
and explain to people and one of them,
one of the people who did things says,
Oh, but all the headings are wrong.
They're not standardized.
We don't even use the same font or
the same way of describing things.
And he, and I don't
have the time to do it.
Well, we could ask our machines, Hey,
clean up the headings, clean up the
heading style and clean up the typography.
And what are differences is
going to make in our world?
kevin_1_08-07-2024_172413: Well,
this is exactly the topic that
prompted me to get in touch about
this this episode, really, because,
you can go back to this debate between
and the specific again.
Right.
So, you've got a lot of excitement around
interfaces and agents where
don't have to think about
which tool you just ask.
You, you ask the chatbot
question or you give it a task
squadcaster-dhaf_1_08-07-2024_092412: task
kevin_1_08-07-2024_172413: off and it
can use different specialist tools in the
background and then comes back with it.
Answer or options or
recommendation, whatever,
squadcaster-dhaf_1_08-07-2024_092412:
whatever.
kevin_1_08-07-2024_172413: you've
also got other sort of specialist
tools that are just enhanced by AI.
So let's take Adobe as an example.
You've got Adobe Firefly, which
is a kind of a chat bot interface.
And then you've got Adobe
Photoshop, which is the
squadcaster-dhaf_1_08-07-2024_092412: is
kevin_1_08-07-2024_172413:
existing specialist.
Image manipulation program with
squadcaster-dhaf_1_08-07-2024_092412: with
kevin_1_08-07-2024_172413: AI
added features like generative
fill and what have you.
So
there's still going to be
jobs that are going to be
done by.
squadcaster-dhaf_1_08-07-2024_092412: by
kevin_1_08-07-2024_172413: tools, like say
squadcaster-dhaf_1_08-07-2024_092412: say
kevin_1_08-07-2024_172413: like Photoshop.
And there are going to be other jobs that
squadcaster-dhaf_1_08-07-2024_092412: that
kevin_1_08-07-2024_172413: probably
going to be completely adequately done
squadcaster-dhaf_1_08-07-2024_092412: done
kevin_1_08-07-2024_172413:
with a general interface.
If you've got any thoughts on
how to think about which jobs,
what's the characteristic, if you
like, of a job that is suited to a,
A general tool
and one that's
Better
squadcaster-dhaf_1_08-07-2024_092412:
suited
kevin_1_08-07-2024_172413: to a
squadcaster-dhaf_1_08-07-2024_092412: a
kevin_1_08-07-2024_172413: tool,
squadcaster-dhaf_1_08-07-2024_092412:
Well, I want to emphasize that AI has
been around for many years, 50 years,
and And it now actually does many things.
And so be careful of the words, AI.
I'm fond of this story.
I first got to Apple, they were
going to release a new product.
said, wow, that's an exciting product.
It's going to change the world.
And it failed.
So they released a second
version, it failed.
They released a third version, it failed.
What was, why was I
wrong in my prediction?
And it didn't.
I wasn't wrong.
It has changed the world.
But it took another 15 years to do it.
15 to 20 years before it did.
It was a camera that didn't require film.
Many people don't even know
Apple came out with one of the
very first consumer digital
kevin_1_08-07-2024_172413: I remember it.
Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
Because the world wasn't ready.
We didn't have a way of
printing the picture.
We didn't have inkjet
printers in those days.
We didn't have a way of sending it from
one place to another, but we did, but we
sent it on, you had something called a
modem, and you took the telephone handset
and put it on the modem, and you sent it.
What, a hundred boards was the
rate, a hundred boards per second.
And nobody even knows what a board means,
but it meant ten characters a second.
Ten bytes
Ten.
Not ten thousand, not ten million.
Ten.
And so, you know, no wonder it did fail.
The world wasn't ready for it.
the same with the camera.
I said, Once we started putting
a camera in, I said, Oh, that's
wonderful for tourist pictures and
for, reminding you of something,
but you'll never get a good picture.
And I said, It's not my opinion,
it's a law of physics, that receptor
is so small, lucky if it gets
one photon on each of the pixels.
And so it's just not enough,
they're not big enough to give
you a high quality picture.
I was correct, but what I didn't know
was that new computational methods
will come around so that today we
have these sales tiny receptors with
millions and millions of pixels.
What they can learn to do is they learn
to combine the pixels into combinations.
And so if you combine four
pixels as if it's as if the
receptor was four times the area.
Or you take three or four pictures
rapidly again, it's as if you had
four times the area and today we get
camera pictures out of your little
camera that wow they're really high
quality And so, that's ai all that's
ai there are lots of things that are
AI and today's AI has no understanding.
It's not, it's, I keep telling
people it's artificial.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
And it's a kind of intelligence,
but what it does is it's
making things up all the time.
the insights and things that
you think are correct and
wonderful were, they're made up.
Because what it does is it takes
your question and it tries to
figure out what is a good, And
it uses part of your question to
give it the first couple words.
And then it says, okay,
I've got three words.
What's the most logical next word?
And it looks through of human
writing, almost, and finds what
is, what word is most common to
follow those particular three.
it has four, it does the same thing
with four, and five, and six, and it
makes a wonderful, document out of it.
Well written, good English, good
language, good grammar, but it's
all made up, it's all nonsense.
And sometimes it's great insight,
sometimes we call it hallucination,
but no, it's not hallucination,
that's the way it works.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
it's young yet, very young, and so
we can't really trust what it does.
I've been reading a new
book that I really love.
It's called Brave New Worlds, How
AI Will Revolutionize Education,
parentheses, and why that's a good thing.
And it's by Salman Khan.
Now, you
kevin_1_08-07-2024_172413:
the Kahn Institute?
Yeah, that's right.
Yeah.
Khan Academy.
Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
And so he said, the Khan Academy was
asked by some of the early people
doing the open AI stuff to see if
they could use it for education.
And and he says, and he's become
a big fan, and he explains why,
and how to use it, and how to use
it in a way that's really useful.
And And that it has lots of flaws, but
you transform those flaws into benefits
for education Because you're trying to
teach the people to use it to enhance
what they do which includes taking a
look at what this device has created
for you and Analyzing it and deciding
whether it makes sense whether it's
whether there's facts behind it or not
So it actually enhances your ability
and skills it's a really good assistant.
It's simply It's intelligence
is not human intelligence.
kevin_1_08-07-2024_172413:
So what's it good?
Cause that the history of tech and
education has been quite patchy.
squadcaster-dhaf_1_08-07-2024_092412:
For the first time you're going
to be able to have a tutor
kevin_1_08-07-2024_172413: yeah,
squadcaster-dhaf_1_08-07-2024_092412:
can talk to.
kevin_1_08-07-2024_172413: yeah.
squadcaster-dhaf_1_08-07-2024_092412:
Because we all know that a tutor, an
individual tutor for each student would
be wonderful, but who could afford it?
And now we can afford it
because it's kind of free.
And it, he says, what this does
is it doesn't get rid of teachers.
But it makes their jobs a lot easier.
Because, in fact, when they want to do a
new assignment, they can ask the AI, Hey
my students know this already and I want
to know if they really understand it.
So, what's a good way of testing them?
And what should I ask them for?
And you could, it really helps you do
lesson plans, helps you evaluate students,
but most important of all, tells you where
the students are weak or where they're
strong and where they need some work.
Or where they can go ahead faster.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
And but they also, good for the
students because students often when
they don't understand something in
a class, they don't want to ask.
They'd be embarrassed to have to
ask and show I don't understand.
But they're not about
asking their AI tutor.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412: it
kevin_1_08-07-2024_172413: So
it leaves the tutor to the human
tutor to focus on more high value.
Kind of activities.
Yeah,
squadcaster-dhaf_1_08-07-2024_092412:
his point.
Yes.
kevin_1_08-07-2024_172413: I want to come
back to that in a second that very topic
you mentioned it earlier
squadcaster-dhaf_1_08-07-2024_092412:
earlier
kevin_1_08-07-2024_172413: AI devices?
There's been
squadcaster-dhaf_1_08-07-2024_092412: has
kevin_1_08-07-2024_172413: Particularly
the first half of this year the
first quarter maybe of this year.
There was a lot of excitement around AI
back devices like rabbit and humane and
there's a new one called friend and a bit
of an echo of the information appliance
replacing the pc there's been excitements
squadcaster-dhaf_1_08-07-2024_092412:
There's been an
kevin_1_08-07-2024_172413: about cat
can ai driven devices you know with
squadcaster-dhaf_1_08-07-2024_092412: AI
kevin_1_08-07-2024_172413: in
the background, replace the
smartphone or at least reduce
our reliance on the smartphone.
That none of those first generation
devices have fared particularly well,
squadcaster-dhaf_1_08-07-2024_092412:
well.
kevin_1_08-07-2024_172413:
I just wondered what you
squadcaster-dhaf_1_08-07-2024_092412: you
kevin_1_08-07-2024_172413: about
squadcaster-dhaf_1_08-07-2024_092412:
about
kevin_1_08-07-2024_172413:
longer term prospects of what AI
squadcaster-dhaf_1_08-07-2024_092412:
of what
could mean to, for new types of devices.
Have you, do you have any opinions on,
on,
kevin_1_08-07-2024_172413: could it,
and that could AI enable new types of
devices that you think could succeed?
squadcaster-dhaf_1_08-07-2024_092412:
succeed?
Absolutely.
some will be good and some will be bad,
but most of them will be mixed, can be
used for good or bad or complicated,
but they won't replace the cell phone.
the mobile phone, because that's
still a really useful device.
And, but, because they'll be part of it.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
And it, because there
are two ways of doing it.
These things require a lot of computation.
And so, people are learning
that they, if you make a small
specialized one, suppose I trained
it on everything I've ever written.
But just me, it's just my writing.
It's therefore whenever I
give it a question, it will
almost always be correct.
So it is because it doesn't have
the, all the writing of the world and
including biases and people making
up nonsense and so on, it's just me.
So at least it'll be me.
And my company, the Nielsen Norman Group,
is, does lots and lots of publications
and so we are experimenting with
taking again, putting that into our AI
system and so when people want to know
something instead of, well, which one
should I read, they can ask questions
and they'll get an answer from our
huge collection of helpful articles.
But again, that means
my cell phone, I'll ask.
Because the computation can be done
either on the device you're using,
or it can be done in the cloud.
kevin_1_08-07-2024_172413: Yeah.
.
But are you saying that
some of the features that
The likes of Rabbit and Humane were
Trying to pioneer, are you
saying that's just going to be a
squadcaster-dhaf_1_08-07-2024_092412:
to be
kevin_1_08-07-2024_172413: software
features on your smartphone?
It won't need a separate device.
squadcaster-dhaf_1_08-07-2024_092412:
device?
Correct.
kevin_1_08-07-2024_172413: Yeah.
Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
In fact, it's already true.
I mean, most of, almost all of the chats
can be accessed from your smartphone.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
Yeah.
No, that's what
kevin_1_08-07-2024_172413:
that's what struck me.
It seemed like it was prototyping
features of future iOS features when
there was no actual need for a new device.
squadcaster-dhaf_1_08-07-2024_092412:
device.
Well, no, the iOS though, if you
will, they are specialized things,
like um, Well, I took an old iPad
and I turned it into an iOS device.
It's now a clock.
Too old to do any of the, to
run any of the modern software,
so I have it run a clock.
And it's nice because it's a
connector to the internet, right?
So, first of all, I can have it
do alarms, and second of all it's
always telling me accurate time.
And that's an iOS.
But lots of iOSs I use the agents in
my house entirely for setting alarms.
Like, just for this podcast, I set an
alarm ten minutes before the podcast was
a start to give me time to get set up.
and I also, we every so often ask it
questions when we're at dinner and talking
with my wife and I and we wonder about
the meaning of something I'll ask it.
And that's very useful.
but, so I think that what's happening
is we're seeing that devices that
have some kind of knowledge base
behind them and therefore can answer
questions or do tasks for you, I are
really valuable, and because all they
require is to talk to them, or type if
you want to then, and eventually it'll
be even that, they'll see what you're
doing and they can suggest something.
it can be any kind of device.
It can be a general purpose device,
it can be a specialized device.
You might have a camera located by your
front doors, so it might be, It can
tell you when somebody is coming into
the front door, but maybe it can even
tell you what kind of a person it is.
I don't mean face recognition, I
mean it might say, oh, this is the
postman delivering a package, or etc.
This is a bunch of children.
And I could just imagine all sorts
of new things that we couldn't do
today that will be empowered by these
new types of computational devices.
By the way, here's where
designers should play a role.
are always going to take away their
jobs and, Oh, how am I going to do?
And I say, well, you should learn
to use them because it will make
your job easier or different.
But actually, why aren't
designers in charge?
aren't designers thinking of
these new things that will really
help people and design them?
And have companies build them and have,
we should be telling people what you
can do with some of these new kinds
of computational devices and maybe
what kind of small physical device
you might want to have around with
you that you carry with you or have it
placed in the house that will help it.
And the next step coming along,
by the way, is just starting.
If you've seen what's
happening in robotics, the
Robots we have, well, when we start
connecting robots with these new types of
deep learning models, you're going to see
another huge jump in capability, because
they will understand how the world works.
Today they just, all they know is because
they've been reading, but pretty soon
they'll be able to actually change
the move and manipulate the world.
I think that's really exciting what's
kevin_1_08-07-2024_172413:
yeah, I think that's really
exciting what's happening there.
Definitely.
There's a, I'm sure you're aware of it.
There's a
squadcaster-dhaf_1_08-07-2024_092412:
of it,
kevin_1_08-07-2024_172413: term in
AI called overtrust which refers
to the dangers of both designers
and users putting too much trust
squadcaster-dhaf_1_08-07-2024_092412:
putting too much
kevin_1_08-07-2024_172413: machines or AI.
A classic example is,
the Tesla's drivers that
That don't understand that.
The badly named autopilot system
squadcaster-dhaf_1_08-07-2024_092412:
system
kevin_1_08-07-2024_172413:
needs monitoring and it, and is
not a fully autonomous system.
And of course that's led to
Fatalities, but what's your
take on that more generally
about, is there a danger of the,
Both
Designers and users?
Just putting,
squadcaster-dhaf_1_08-07-2024_092412:
putting a,
kevin_1_08-07-2024_172413: in this
early stage when there's lots of
things being made up, hallucinations,
errors, all the rest of it.
squadcaster-dhaf_1_08-07-2024_092412:
Yeah,
kevin_1_08-07-2024_172413: what's
your take on overtrust from a designer
perspective and the user perspective?
squadcaster-dhaf_1_08-07-2024_092412:
doesn't come from AI.
I'm not sure where it comes
from, probably from politicians
kevin_1_08-07-2024_172413: Right.
squadcaster-dhaf_1_08-07-2024_092412:
over the last 50, 000 years who make
promises, because people like to hear
the promises, but they don't necessarily
intend to actually keep to them.
maybe they even do intend, but
it turns out it's not possible.
I don't know the solution is Well, there's
an old saying, it comes from the arms
race, in the United States at least,
which is, you make a, when you make
a treaty with some other nation about
arms, You should trust it, but verify.
kevin_1_08-07-2024_172413:
Trust but verify.
squadcaster-dhaf_1_08-07-2024_092412:
trust what you're saying, but
you know, I don't really trust,
so, I want to verify and I want
evidence for what you're saying.
that's what we need in
AI, is trust but verify.
We sort of say, yeah, that sounds
good, but let me, and so, what's nice
is they now, a lot of the systems give
you the references that they've used to
say what they've said, you have to go
and look at those references and then
decide if that references yes, I can
see where it came from, this reference,
but do I trust this reference material?
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412: And
then sometimes it'll give you a URL and
you go and you see, it's a fake URL, it
made up the URL because since it doesn't
understand URLs, it doesn't understand
what it means and how they're created.
Uh.
You know, oh well, let's see.
They often start with www, okay, period.
What would follow next?
And it puts out something that
may be completely illegitimate.
kevin_1_08-07-2024_172413: So a mix of
healthy skepticism and regular checks.
squadcaster-dhaf_1_08-07-2024_092412:
Yeah, but that's kind of
how you have to live your
kevin_1_08-07-2024_172413: Exactly.
squadcaster-dhaf_1_08-07-2024_092412:
modern world.
Nothing to do with the technology,
it has to do with the world.
kevin_1_08-07-2024_172413: Yeah.
Well, let's go back to that,
That example you gave around teaching.
And look at,
What AI is best used for and what
humans are best left to doing,
So machines are great at
Pattern recognition, rapid
Data analysis and synthesis,
reliable repetition, working
around the clock, things like that.
.
Humans tend to be much better at the
softer and more intangible stuff, like
squadcaster-dhaf_1_08-07-2024_092412: like
kevin_1_08-07-2024_172413:
understanding a situation rather than
squadcaster-dhaf_1_08-07-2024_092412:
rather
kevin_1_08-07-2024_172413:
mimicking understanding,
squadcaster-dhaf_1_08-07-2024_092412:
you know, Empathizing, judgment,
kevin_1_08-07-2024_172413: reacting
to new developments, things like that.
squadcaster-dhaf_1_08-07-2024_092412:
like that.
kevin_1_08-07-2024_172413: And I know
10 years ago, you wrote something
called things that make us smart with an
interesting subtitle that read defining
human attributes in an age of the machine.
squadcaster-dhaf_1_08-07-2024_092412:
the Machine.
That
kevin_1_08-07-2024_172413:
over 10 years ago, I think.
Do you think that human strengths in
the way that AI are being, is being
talked about and designed at the moment?
Do you think it's being, do you
think human strengths are, Oh.
being valued enough.
squadcaster-dhaf_1_08-07-2024_092412:
enough?
kevin_1_08-07-2024_172413: Do you think
that designers particularly need to maybe
champion human strengths a bit more?
Or do you think,
squadcaster-dhaf_1_08-07-2024_092412:
think,
kevin_1_08-07-2024_172413:
the AI designers have
Got it about right?
squadcaster-dhaf_1_08-07-2024_092412:
Well, today's AI designers are still
primarily technologists who may not
You have to be careful when trying to
characterize a large number of people.
But the emphasis has not been on
human behavior and and people.
That's changing, by the way.
There's lots and lots now of
designers getting into it.
Uh, human centered AI is talked about.
And there's a lot of concern about the
fake evidence that's coming out and trying
to put in what it's called guardrails
and other ways of preventing that or
maybe the system should check itself.
You'd need sort of an overriding system,
if you will, of the ego of the system
that, before it puts out anything, it
does a quick internet search to see if
those are really correct statements.
It's the same with arithmetic.
Computers are fantastic at doing
arithmetic and calculus and they can,
you know, we have all sorts of powerful
calculators that will solve calculus
equations and algebraic equations, but how
come the GPTs are unbelievably bad at it?
They can't even count.
But the solution is pretty trivial.
When it sees that they're doing
something that requires mathematics,
it should call on the operator.
If you will, the specialized
tool to do the mathematics.
So, Hey, you know what?
Here we go again.
Just like we have information appliances
for people to do specialized things.
Well, the AIs are going to have to have
specialized appliances too , so they
could do specialized things like get the
spelling right or get the mathematics
right, or, you know, get the facts
right about who really was the early
presidents or prime minister or whatever.
And that's happening.
That's starting to happen.
But it's really hard.
I was about to say, it's really
hard to predict the future.
And I was going to say, and, but
there's a well known saying that
says It's easy to predict the future.
do it all the time.
The difficulty is getting it right.
And so, I hesitate to predict and friends
who are in that business, they don't call
it prediction, they call it forecasting.
And what they try to do is think about all
of the things that are happening today.
And all of the implications of them,
and the many different directions in
which this could cause behavior to go or
societies to move, or the world to move.
so that we're ready, no matter what
happens, we tend to be ready for it.
But we don't know which of these
many ways will actually happen.
kevin_1_08-07-2024_172413: But I think the
I guess what I was getting
at there was that the,
A lot of the conversation in the
last couple of years has been
Very,
Boostery or very
doomy but it, but both sides of that
Pessimism has assumed that
AIs will be able to
replace a lot of human activity
or, automate a lot of it away.
squadcaster-dhaf_1_08-07-2024_092412:
I think in the early days
of a new technology, most
of what is said is nonsense.
And it's.
it's oftentimes best people in the
world spouting the nonsense because
they have this vision of what might
happen, but it's a made up vision.
And I think that people have
over interpreted what has
happened with these new tools.
And think the best thing to do is
try to, is become familiar with
it and become comfortable with it.
as I said before, trust but verify.
What impressed me about Sal Khan's
book was that he was sort of a skeptic.
But he really had this vision of trying
to harness it for the improvement in
education, and that's what he cares about.
he and his team have been
trying to develop tools that
focus on that one topic.
Now that one topic is a very
difficult, complex topic with
many, many different issues.
But by focusing on that and creating
AIs that are particularly well trained
in just in education and in the
concepts that which could be anything.
Could be language, could be history,
could be humanities, could be science.
That you can actually get more of the
benefits and less of the negatives.
And as I said before, learning how to
transform the negatives into positives
because they're good learning exercises.
kevin_1_08-07-2024_172413: Yeah, exactly.
And, going back to your, to Khan's
example of when's, how to use
How to use AIs in the classroom,
squadcaster-dhaf_1_08-07-2024_092412:
classroom.
kevin_1_08-07-2024_172413:
example would be
The house setting, you know,
an AI might help a doctor,
Read a scan.
But you'd still want
The human doctor to report the
results of the patient sort of thing.
So it's just playing to
Playing to the strengths of
and playing to the strengths
of the machines, if you like.
And you've talked in.
squadcaster-dhaf_1_08-07-2024_092412:
also has a better notion of the whole
context, the history of the patient,
and the the meaning of this, and yeah,
how best to explain it, and whether,
and what type of treatment, which is a
function of, based on lots of things,
including preferences of the patients.
One of my radiologist friends,
when I asked him about that, and
he's, considered one of the better
radiologists in the country, uh, does
it mean that it will do all the, you
know, it will do all the diagnosis
for you, what kind, what, you know.
he said, thank goodness.
Because he used to deal with patients.
And now what he has to do is
sit in the room all day looking
at screens and trying to
kevin_1_08-07-2024_172413: Exactly.
squadcaster-dhaf_1_08-07-2024_092412:
little, you know, examples of, Oh,
I think this is an issue that we
should look into in more detail.
Let the machine do that, and then
I can spend more time back working
with patients and helping them.
kevin_1_08-07-2024_172413: Now I want
to read out a little quote from Things
That Made Us Spark that I think really,
I think it came under
The broad heading of
intelligent amplification.
But you said, My, my theme is not
anti technological, It's pro human.
squadcaster-dhaf_1_08-07-2024_092412:
human.
kevin_1_08-07-2024_172413: should be our
friend in the creation of a better life.
It should complement human
abilities, aid those activities
for which we're poorly suited,
,
and enhance those that
were ideally suited.
So to me, it's about humanizing
squadcaster-dhaf_1_08-07-2024_092412: just
kevin_1_08-07-2024_172413: and and
the appropriate use of technology.
And that's something I've talked about
the human machine interlaced before where,
workflows should be redesigned
to make the most of what humans
and machines are best at.
,
now we're over the initial kind
of hype of AI, do you think
the wind's blowing more in
squadcaster-dhaf_1_08-07-2024_092412: more
kevin_1_08-07-2024_172413:
direction in terms of, instead of
squadcaster-dhaf_1_08-07-2024_092412: of,
kevin_1_08-07-2024_172413:
replacement, it's more
squadcaster-dhaf_1_08-07-2024_092412:
it's more,
kevin_1_08-07-2024_172413:
complementing human?
squadcaster-dhaf_1_08-07-2024_092412: Yes.
And first of all, we're not
over the hype, but, that takes a
kevin_1_08-07-2024_172413:
It's calming down.
squadcaster-dhaf_1_08-07-2024_092412:
But let me go back to the early days
of the industrial revolution when the,
we automated the looms and the weaving.
And The Luddites, so called, because
of the man Ludd they protested and
they tried to destroy the machinery,
and now today the word Luddite means
somebody who hates machines or doesn't
understand them, which is wrong.
They didn't mind machines.
They did understand them.
What they were against
was losing their jobs.
kevin_1_08-07-2024_172413: It's helping.
squadcaster-dhaf_1_08-07-2024_092412:
And the truth is, they
ended up losing their jobs.
So, in this case, if you take a
look at the Industrial Revolution,
it has several important lessons.
First of all, people did lose their jobs.
You couldn't say, oh, let's learn to make
people work better with the machines.
Well, no, the way the machines were
designed, they just needed caretakers.
People that when the thread broke, they
would, you know, replace it or repair it.
And when something went
wrong, they would fix it.
But basically, you walked around
all day doing nothing except looking
for little issues that needed help.
So you were the assistant to the
machine instead of the other way around.
But the result to society was
that cloth was now much less
expensive than ever before.
And that created lots of new
opportunities for people to make,
use cloth in many, many ways that
were too expensive to do before.
And so, new jobs and
new opportunities arose.
And so, to society there was huge benefit.
To the people who lost their jobs, There
was a huge negative what we're trying
to do today Is we are it is absolutely
the case that people will lose jobs will
replace the kinds of jobs and the question
is though as we find new opportunities
because New things are done now that are
faster better cheaper Which is therefore
enable all sorts of new activities to
do that were too expensive before too
difficult and Can we make it so that a
lot of the people who are displaced are
able to be part of this the new regime
That make use of it that's that could be
hard because sometimes requires different
kinds of knowledge and skills and the
people who are being displaced are too
old or Didn't don't have the right plan or
education to acquire them that's I think
the challenge That we do want to make
this be a collaborative effort And but we
have to also think of the humane impact.
But one thing that's in our
favor, that these kinds of
revolutions take decades to happen.
when we say it may replace jobs, it's not
going to replace those jobs immediately.
A lot of companies are trying
to do that, but it's, they're
just, fooling themselves.
They're saying, our expenses are too
high and we have now AI and so we can get
rid of, you know, five, fifty percent of
these people and they get rid of them.
Guess what?
There's nothing, the new AIs will
not replace what they've lost.
And so that, that's just
bad, stupid thinking.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
will take a long time.
Before these changes really come about
and that gives us time to to I think
to adjust to them and learn how to
live with them and better yet, learn
how to take advantage of the positive
things and learn how to avoid or make
illegal or make impossible the stuff
that is really harmful to civilization.
kevin_1_08-07-2024_172413: Yeah.
And taking the examples of the teacher
and the doctor that you talked about,
The upside could be that AI's
make work more human, more
interesting, less mind numbing
.
And avoid, I think what you call the three
D's, the dull, dangerous and dirty work.
That's a potential upside.
squadcaster-dhaf_1_08-07-2024_092412:
upside.
A group at my university actually
devised a way of helping women
who have problems in lactation
because, breastfeeding basically.
And it turns out there's a huge problem
in the world and there aren't enough
specialists to take care of all of the
people who think they have difficulties.
And so they've invented this simple
device that uses a cell phone camera.
Just a rec, just a camera.
so they, If you feel that you are
having trouble with your breasts, it
tells you how to take a picture of
your breasts and send them, and it
does, then it does a quick analysis,
and tries to say that's okay.
it, you know, the problem is this, and it
will go away, or you should treat it this,
or we're not sure, but here's where you
should go to your physician, and here's
what you should tell your physician.
In other words, it's doing triage.
And so what triage does is it takes
the cases that we see are important.
We send them to the experts, but cases
that are not important, that could be
solved much more, you know, simply at
home we'll tell you how to do it at home
and you have to make the system so that
it, because it will make errors, but you
know, those of us in the science know
there are two kinds of errors you can
falsely say something that isn't true,
or you can miss things that are true.
And so you want to minimize the misses.
And so you just have to be
conservative in what you tell people.
That if there's any doubt, you
send them to the physician.
kevin_1_08-07-2024_172413: Yeah.
squadcaster-dhaf_1_08-07-2024_092412:
And but that really reduces
the loads on the physicians.
And it also often improves the
mindset of the person who worries.
You know, anybody, any parent knows this.
Your children always seem to have problems
in the middle of the night on a weekend.
And it's the middle of the night,
so there's nobody around to help.
And there's a weekend, so you your normal
physicians aren't available either.
It be nice to have a very simple
system that did a quick, simple
analysis and said not to worry?
Or said, um, you ought to
go to the emergency room.
That would be really useful to know.
kevin_1_08-07-2024_172413:
That's certainly
squadcaster-dhaf_1_08-07-2024_092412:
and what this group that I'm talking
about, they're using smartphones.
And they've developed incredible
things you can do with a smartphone
with nothing else, just some software.
They now have a way of you hold the
smartphone with your hand and you
can measure your blood pressure.
And how do you do that?
Well, we turn on the vibrator,
we turn on the, as it vibrates.
And well, you push hard on the, you
push hard on the camera, and as you, the
more you push, it changes the vibration.
And so we can tell by
how hard you're pushing.
But we can tell how hard you're
pushing by how it, how much
the vibration is changing.
And then, your finger is right on
the camera, and so we also turn on
the camera light, and it hits your
skin and back into the camera, and
therefore we can see your pulse.
And therefore we can tell when the
pulse stops, which means, okay,
that's high, that's the high point
of your blood pressure, and we
can tell when it resumes again.
Okay.
And they basically do
a blood pressure test.
kevin_1_08-07-2024_172413: amazing.
squadcaster-dhaf_1_08-07-2024_092412:
nothing but some software
and some cleverness.
kevin_1_08-07-2024_172413: Amazing.
squadcaster-dhaf_1_08-07-2024_092412: But
it turns out the cell phone can detect
all sorts of anomalies that could make
it much easier for the physician because
you've done a quick diagnosis already.
it 100 percent accurate?
No.
even being 80 percent
accurate is a dramatic help.
And once again, you be
conservative in what you say.
We know we're only 80 percent
kevin_1_08-07-2024_172413: Yeah,
squadcaster-dhaf_1_08-07-2024_092412:
so If we have any doubt, we
send you off to a specialist.
kevin_1_08-07-2024_172413: certainly.
But how can, let's just as we draw
to a close here, how can we apply
this thinking about how to use AI
design, so
Design designers where I'd like lots
of people about the impacts of AI.
What's your thoughts about how,
,
Where AI
squadcaster-dhaf_1_08-07-2024_092412:
where
kevin_1_08-07-2024_172413: Bring benefits
to the design process where you think
what parts of the process are ripe for,
squadcaster-dhaf_1_08-07-2024_092412:
wrong, that's the wrong
way to think about it.
They take, if they don't take it
personally, new material is developed.
It's biodegradable, but it's strong
in this way and weak in this way.
Or we have a new method of computation,
or new sensors that allow us to sense
things we never could do before.
From a distance, or maybe from
inside the body, or whatever.
And, are opportunities.
So, you have a challenge, think
of it as new opportunities.
And so AI, don't think of
it as, oh, take away my job.
Ask yourself, hmm.
What can I do with this
that we couldn't do before?
These are opportunities.
And why aren't designers, instead of
complaining about it, and saying it's
all biased, and we have to get rid
of this problem, or that problem, or
whatever, well, those are correct.
But, first of all, that's an opportunity
for you to help solve those problems.
second of all, it's an opportunity for
you to figure out exciting things I can do
that I could never do before, that I could
make into something valuable for people.
And that's where I want to
see designers leading the way.
Right now we have
technologists leading the way.
Technologists who say,
Oh, I have a technology.
I, you know, I have a hammer.
I'm running around looking
for a nail to bang it into.
I have a technology.
I'm looking around somewhere.
No, no.
What designers often start
with is understanding people,
issues in everyday life.
And we find some pro, some
issue that bothers people.
People may not be complaining,
by the way, because they think,
well, that's the way your life is.
It doesn't have to be that way.
we're really good at spotting those
things and then figuring out something
that we can do that will help them.
And AI is one of those things that
we might be able to use to help in
a lot of these different issues.
So think of it as a great opportunity.
kevin_1_08-07-2024_172413: but the design
leaders I've spoken to about this are
squadcaster-dhaf_1_08-07-2024_092412:
about this are really,
kevin_1_08-07-2024_172413: I guess that
They're quite targeted in
where they're using the tools.
It's probably
not as
squadcaster-dhaf_1_08-07-2024_092412:
not as,
kevin_1_08-07-2024_172413: such a
broad impact as they thought it might
do, but it's definitely changing the
capabilities they need in their team.
So, example, they need people who can
squadcaster-dhaf_1_08-07-2024_092412:
can brainstorm,
kevin_1_08-07-2024_172413: it can really,
the productivity of brainstorming,
but you need people who can,
work with those AI tools effectively,
which might be different from
the people who would have been
sitting around sketching quickly
and on post it notes and things.
So
slightly changes the capability
mix in the team and people can be
retrained and all the rest of it.
squadcaster-dhaf_1_08-07-2024_092412:
a new tool.
It's a new technology.
So, yes, it does change.
That's what I said.
Think of it as a new material.
You have to now have, it's very different
from anything you've done before.
You need new tools to create with it.
But it can create, in fact, maybe
it can't do rigid, structures.
It does free flowing
biological structures.
well, that's interesting.
Why does a house have
to have straight edges?
Or why does a building, or why
does a package, or why does
a this, that, or the other?
You know, if you take a look at
architecture, and for that matter the
car design, the exterior body of a shell
of a car, they used to be all straight
edges, because that was the only tools
we had that could bend metal, right, and
make it in rectangular forms, or with
straight edges and corners and so on.
as new materials come out, and new methods
of manufacturing come out, they can be
any shape in the world that you think of.
And with additive printing, additive
manufacturing methods, you can have
things with, that are, you get rid
of stuff, material that isn't used,
you know, so you can have big holes
inside the material, which makes things
lighter, and doesn't lose the strength.
And, well, yeah, so think about
new things, and new tools,
and new ways of doing things.
And AI is the same way.
Can it solve all problems?
No.
it can actually do things that, and I
can't tell you what it's going to be.
I rely on your creativity to discover
them, and you get me excited about it.
kevin_1_08-07-2024_172413: We
need to spend more time on that.
I'm sure.
Just as we wrap up, do you have any just
general reflections on great examples of
human or humanity centered AI or any sort
of principles or rules of thumb that you
would like to leave the audience with?
squadcaster-dhaf_1_08-07-2024_092412:
it's a little bit too soon to be
saying, but I'll mention two examples,
one of which I've already mentioned.
Khan's work is not yet really out
and for the Khan Academy I can't
tell as I'm reading the book how
much is still in his laboratories or
how much has actually been released.
So, as I finished the book, . I'm gonna
go back to the Khan Academy and see
where it stands for public release.
But it still has already given me lots of
ideas of things that we can do in other
domains besides the ones that he covers.
the other thing, and I've been
having a big debate with with my
friend Ben Schneiderman about this.
I'm, I was an advisor
to a company called L.
E.
Q.
E L I and then the letter Q.
It's an Israeli company that makes
agents uh, assistants for people
Who don't have any social, don't
have any social interaction.
may be they have mental problems.
They may be old and ill.
Most of them tend to be old and ill.
And they're often in retirement
homes or sometimes in their
own home, but all alone.
This is a system that is
meant to be to help you.
To say, oh, you know, there's an
exercise class in a half hour.
You should really get up.
Come on, get set up.
It's really important that you go
and do, go to this exercise class.
Or or isn't it time you
called your daughter?
Here, I can call your daughter for you.
Would you like to talk to your daughter?
And or it's time to take your pills.
Or, you haven't had your meal yet.
And you really need to eat something.
Now, This can be really valuable
for people who lack social contact.
Can lack it for lots of reasons, including
maybe they just have severe depression.
And even though there are lots of
people who are trying to help them
and so on they turn them all down.
sometimes they often interact
with these machines in ways
they never would with people.
it can be very valuable.
And, and so the system is actually being
accepted by lots of retirement centers
and wellness centers around the world.
And, but, you know, Ben Schneiderman
thinks, no, this is all deficient.
You can't have real empathy.
You can't have a real
relationship with a machine.
I say, I don't even want
to argue that point, Ben.
I'm telling you, tell you, there are
situations where there's no alternative
and this is a better alternative than
having nothing people really like that.
And actually, but as I was reading
Khan's book about education,
is saying the same thing.
students who have problems, can't
understand material in class, often
are ashamed to talk about it to their
parents, ashamed to talk about it
to their teachers, or even to their
friends, because they think, oh, I
guess I'm stupid, and I can't understand
it, because it's all confusing.
But they will talk to their, to a computer
agent, who becomes, actually gets to know
each, gets to know you after a while,
and knows exactly what your issues are,
or what you like, and so, When you can't
understand this algebra problem, it says,
Well, you're, you love to play football,
or, which in Britain would be soccer.
Um, well, think of it as a,
suppose you're in the middle of the
game, and this happens, and then
that happens, and this happens.
How do you know what you should do next?
And, uh, and use, and use it as a way
of introducing the same arithmetic
problem that you're having trouble with.
um, and that's what these
agents can do wonderfully.
And it's, it's, it's not a religious
issue and it's not a moral issue, is
that, um, they're always available.
So, sometimes you don't want to
talk to a person, you're ashamed.
Or sometimes you don't have anybody
around when you need them the most,
like in the middle of the night
or in the weekend or whatever.
And so, I think there's a
lot of good that's starting
to come out of these agents.
Right now, they're not very good.
Uh, but the, once you start adding GPT
and other similar things to them, and
again, where they're trained not on
the world's, you know, knowledge which
is filled with biases and, you know,
uh, but rather on carefully databases,
I think wonderful things will happen.
I mean,
kevin_1_08-07-2024_172413: I mean,
just like the education example,
presumably, care home kind of
agents , shouldn't be ideally used as
a replacement for human
contact, but a compliment.
squadcaster-dhaf_1_08-07-2024_092412:
It's not, I mean, in fact, if you take a
look at what LEQ does, it's often urging
you to please go and get some human
kevin_1_08-07-2024_172413: Yeah.
.
Certainly.
Very good.
And where can people find your work?
squadcaster-dhaf_1_08-07-2024_092412:
Well, I haven't been publishing lately.
The main place I publish
is, , believe it or not, LinkedIn.
, because It's hard to find a
place to publish, articles.
I don't want to publish in the
technical journals because my stuff
is not written for them anymore.
So it fits what they are interested in.
And, trying to find their media,
there's a couple of different places
that one can publish and for various
reasons I ended up on LinkedIn.
And, but, uh, You know, if I publish
in, in one of the scientific journals or
academic journals, it might get read by,
uh, a thousand or two thousand people.
And wow, that's a, that's
a large number of readers.
And I publish in LinkedIn and
it might get read by thousand,
Thousand people.
kevin_1_08-07-2024_172413: your profile
in the, in the show notes as well.
So thank you very much, Don, for
your time and wisdom as ever.
It's been a great conversation.
Thanks a lot.
squadcaster-dhaf_1_08-07-2024_092412:
Thank you.
It's always a pleasure to talk to you.