[Music]
so back in 2004 I was driving my lovely
spouse to a conference she is an
obstetrician gynecologist and was
working as such at the time and the big
national conference in her field was
being held in Philadelphia now I’m a
academic computer scientist I’m a
computer science professor I’ve been to
lots of conferences I knew what to
expect and so I was sort of surprised
when I got there and discovered that
right there at the conference where the
attendees had to enter there were
protesters carrying signs and chanting
and I thought to myself that doesn’t
really make any sense this is a ob/gyn
conference they’re about trying to
improve women’s health I don’t see much
to protest here but the fact that matter
is that the stuff that they do in
obstetrics and gynecology is it touches
some some hot-button issues so they work
on abortion they do contraception some
of the protesters there were actually
protesting circumcision so there was I
don’t know why that’s an issue but
that’s that’s what they were concerned
about so they were marching around and I
thought to myself wow I’m I’m really
glad I’m not in a field where people are
protesting and getting all upset but
then I thought about it a little bit
more and I realized ways again what I do
is is studying artificial intelligence
we’re trying to make machines that have
the kinds of intelligence that people
have and if we’re successful there’s
very few things more threatening to our
identity as human beings as our ability
to think intelligently and to think at
all so you know if we were doing our job
well people would be upset probably the
fact that we don’t have protesters means
we’re just not doing a very good job
so just imagining a time in the future
when when in fact artificial
intelligence is being protested and I
thought okay well that would be a sign
that we’re doing better
well anyway flash-forward some 13 years
and this summer here in Providence the
annual meeting of of the governors of
the states in the US all got together
and
met and tried to talk about their shared
problems and one of their speakers one
of their invited speakers was a fellow
by the name of Elon Musk
now Elon Musk is a tech entrepreneur
he’s the head of SpaceX and also Tesla
Motors the electric car company and he
got in front of people and said that of
all the things that we should be worried
about his governor’s AI that’s the thing
artificial intelligence you should
really be very concerned about that it
represents a fundamental existential
threat like human extinction is on the
line if we don’t regulate it and and and
deal with these problems appropriately
so that’s that seems really concerning
he doesn’t he hasn’t he didn’t express
at the time what the argument was why it
why it was it was something that we was
afraid of but he said you know I’m
afraid and you should be very afraid now
I’m not afraid and I want to kind of
convey to you what the issues are why it
is that he’s concerned and why it is
that say I’m not concerned and give you
a sort of idea what the future it might
look like with respect to artificial
intelligence so I think I think the most
important thing to say right off the bat
is that there are at least two different
ways that people use the term artificial
intelligence so artificial intelligence
is a research field that we study in
computer science we we build robots we
make programs that make various kinds of
decisions and that’s something that we
study day to day and it’s been it’s been
underway for quite some time but
artificial intelligence is also a term
that’s used in science fiction to
represent this sort of notion of
machines that have become aware of their
environments aware of what their
capabilities are they can plan ahead
they can they can make their own
decisions and their own goals and and
pursue them and to be clear this this
first kind of artificial intelligence is
something that we have with us today and
it’s as Kip was mentioning it’s it’s
it’s hard to get it to get robots to do
interesting things to get software to
not be stupid but there’s also this
notion of artificial intelligence that
we might call artificial general
intelligence systems that are as smart
as people across the boards and this is
something that doesn’t exist you can
think of artificial intelligence is
being represented by the kinds of robots
that are stumbling around today
and you can think of artificial general
intelligence as maybe the kind of robots
that you might see in a Terminator movie
who are you know out to destroy us and
are very good at their jobs and they’ll
be back
so I think to understand this
distinction and understand why why I’m
not concerned part of it is is the
extrapolation from the AI that we have
now to this kind of hypothetical AI that
we that we don’t know how to build that
we don’t know that it exists so that it
could exist is is that is the main issue
that that I want to try to get across so
let me say a few words about what
artificial intelligence is if you don’t
know a little bit about the history so
back in 1956 a group of researchers
early computer scientists who are just
getting to think about what computers
can do we’re very interested in making
computers that that could exhibit the
kinds of computational processes or
thought processes that people have they
decided they’d get together at Dartmouth
College and they’d work it all out and
maybe over the course of a summer they
would solve this and then you know we
can all get on with with other things
and some of the problems that they
identified were things like how do you
make machines that use language how do
you make machines that can improve
themselves it can learn from their
experience these are great goals and in
fact they’ve kept us busy for some 60
years since then but they didn’t solve
it over the summer obviously but in fact
a lot of progress has been made and so
let me say a few words about various
ways that artificial intelligence is
with us today so one thing is you know
maybe you have an artificial
intelligence piece of software in your
pocket these these devices that actually
let you navigate to locations in the
world are just chock-a-block full of the
kinds of AI systems that we’ve been
designing the field has been designing
for decades so just to spell it out
maybe in a little bit of detail you
might want to ask your phone how to
navigate from here to power Street in
Providence and so to do that you know
you take out your phone and you speak at
it and what the phone receives is a
speech signal so this is the speech
signal of me actually saying take me to
power Street and that’s what it looks
like it’s it’s just a bunch of patterns
or numbers the problem of speech
recognition is the problem of turning
that that sig
that speech signal into words into
symbols and that’s a problem that’s been
with us for a long time and people made
a tremendous amount of progress on it in
the last you know five to ten years and
now speech recognition kind of works but
that’s not the end of the AI just in
getting directions you also need to turn
that that list of symbols into meaning
like what is it that you want the phone
to do and so this process of
interpreting the individual symbols the
language symbols and and and figuring
out what the intent is is sometimes
referred to as natural language
processing it’s another problem of
getting computers to use language and
this also is something where we’ve made
a lot of progress in the last few years
and now we can reliably recognize when
you ask the phone for something you know
what it is that you’re asking for maybe
not entirely reliably but but very
reliable enough so that we’ve fielded
this to millions and millions of people
so in addition to using language the
system also learns from its experience
and so the way that machine learning
comes into play in this kind of system
is when you’re when the system is trying
to figure out what path you should take
to get to power Street what it’s
actually doing is evaluating all the
possible different routes and figuring
out which one is going to lead to the
the fastest travel time to get you there
in the shortest amount of time to do
that it actually has to make estimates
it has to say okay here’s a little piece
of the street say South Water Street how
long is that going to take to traverse
now you can say well just use the speed
limit but that’s not exactly what people
typically do there could be traffic
there could be you know delays of
various sorts and so the ability to
predict based on features like well
where is this piece of road how big a
road is it what time of day is it what
day of the week is it all those things
get factored in together to generate
some kind of prediction and these
predictions are improved by feeding the
Machine lots and lots and lots of data
so anyway in the big picture all these
different little AI subproblems come
together to provide something that is
very valuable to people and it doesn’t
seem so scary right that’s not taking
over the world that’s not destroying us
in any way that’s kind of helping us get
from place to place what’s what’s the
missing step here and so the way that I
like to think about this
well if you think about a system that’s
actually say taking you from place to
place within a city maybe even helping
to drive you around that’s really
subsuming the capabilities of a human
job the job of being say a taxi driver
and so we don’t have this quite yet but
it’s really not hard to imagine in the
near future having a system that can do
all the things that a taxi driver can do
and to get you from place to place
and once that happens we have an AI
system that has all the capabilities
that it has plus it can be a taxi driver
okay well if it can do that what else
can it do it’s not too hard to imagine
that the job of say radiologist would
also be something that we could you know
absorb into the into the AI system a
radiologist takes data about you know
slides or x-rays taken in the human body
and it may can they can make various
kinds of diagnosis that seems like a
very a is sort of problem I think we can
do that one too we can again we don’t
have this yet but it’s it’s not hard to
imagine a time when the system has now
absorbed as well the job of radiologists
okay well what’s what would be next now
Here I am sitting in my lab maybe with
my students and I think to myself okay
well if we can do those things what else
can we do maybe we can maybe we can
absorb my job or maybe I wouldn’t think
about it that we might think I could
absorb the job of all my students
replace them with it with the machine so
we could we could all work together and
design a piece of AI software that can
do the things that an AI researcher can
do can we can come up with ideas for
designing programs we can make those
programs we can test them in the field
like yeah so so I can actually replace
all my my grad students with you know AI
researcher 2.0 and at that point then we
have a brilliant idea then we think okay
we can use our now collective
intelligence together to design even
better AI researchers AI researchers 3.0
now at that point it’s a little hard for
me to keep up with these these AI
researchers that I bought from the store
and they have improved themselves one
knotch and they begin to improve
themselves at a very rapid pace
potentially over the course of a weekend
the theory goes and we experience
something that sometimes referred to as
an intelligence explosion so this
intelligence turned back upon itself is
and more intelligence which can generate
even more intelligence we can ultimately
absorb all kinds of roles of people in
fact maybe all possible roles of people
you know including people who come to
TEDx talks you could imagine anything so
terrible at that point we would have
achieved what you can refer to as
general super intelligence so an
artificial intelligence system that is
smart not only is any individual people
but like anything that a person can do
this system can now do better than us
now this isn’t a threat in and of itself
but it does leave us humanity in a very
awkward position because we no longer
have control of our own destiny right
it’s now the machine that or the group
of machines that can do anything that we
can do better than we can do it and if
they decide that you know we need help
then maybe they’ll help us and that
would be awesome or if they could decide
that we’re in the way then they can get
rid of us and there’s really not much
we’d be able to do in response because
that’s how powerful they are in this
story so this is this is kind of the the
crux of the problem this is the big
concern is so just to put a fine point
on it
the idea is that we get computers up to
a certain point where they can exhibit
recursive self-improvement they can make
themselves better at which point they
can create entities of individual
brilliance the likes of which we’ve
never seen at that point the the fear is
that the things that make us special and
powerful the things that that humans can
do are now being taken over so humans
become obsolete and then finally as a
result we’re no longer the dominant
species on the planet we’re at the whim
of of these machines so that’s that’s
the story that that Elon Musk and others
would like you to be afraid of and it is
kind of a scary story right it’s kind of
straight out of Hollywood
because it’s kind of straight out of
Hollywood this is not coming out of the
science of it but it’s coming out of
speculative thinking about the
directions that these things can go and
I want to say a few words about why I
don’t think this is this is you know
it’s a great story but it’s not it’s not
a path that we’re about to follow
so the first thing I want to point out
is that if you ask a biologist about
who’s the dominant species on the planet
they don’t instantly say well it’s a
sitz people people are awesome I mean I
like people right some of my best
friends are people so it’s not it’s not
like I I think that people are doing a
terrible job at everything but it is
also the case that if you have if you’re
gonna say we’re the dominant species you
need to be careful about your
definitions and it’s the case that for
example in terms of which species have
lasted the longest well horseshoe crabs
have done a lot better job than us at
that maybe that’s not fair right they
had a head start
we’ve been maybe doing well since we’ve
gotten here but bite me again measured
by what measured by the individual
lifetimes well you know people live a
good long time but the the ocean Quahog
lives like four times longer than us so
yeah we’re good but there’s better but
there’s a lot of us but not nearly as
much as there are of Antarctic krill
they’re way ahead of us it’s like oh
that’s not fair they’re they’re tiny
right you know one of us is like I don’t
know a billion of those or something but
if you actually think if you factor
weight into this story while cows
actually beat us on that front there’s
more cow mass on the planet than there
is human mass of course that’s kind of
our fault right we actually gave cows
that boost so you could say what about
impact on the planet as a whole and we
have had a tremendous impact on the
planet but there’s other organisms that
have had arguably more impact on the
planet blue-green algae of all things
produces 330 billion tons of oxygen each
year a small fraction of which we
managed to breathe they’re just lots of
things out there that are having a lot
more impact on the world than we are now
again you know people are still very
important and we have a very big role to
play but but again you have to be
careful about defining your terms now
what about what about the impact of
intelligence on our own destiny if you
ask a historian
they don’t necessarily say that it’s
individual geniuses that have moved
everything forward they’ll give the
example of for example independent
discovery the idea that some brilliant
ideas actually are so important to
society that they appear in multiple
people’s heads that sometimes roughly
the same time and so if you actually
look at this
and I’ve been collecting examples of
these things because I think it’s really
neat there’s quite a range of issues
that actually come up the telephone very
impactful in society the invention of
calculus very impactful on society they
were independently invented by multiple
people really important things but also
some kind of trivial things the the
bacon martini was independently invented
by multiple people a mathematical
notation for specifying juggling
patterns for different groups had the
same idea at the same time for this and
so that’s kind of big and little but all
the way between two beer was
independently invented in multiple times
in history and the neural network
structures that are actually driving a
lot of the progress in AI have been
invented by multiple people over the
over the decades so it’s worth pointing
out that none of these things happen in
a vacuum even even vacuums didn’t happen
in a vacuum they were independently
invented by multiple people at around
the same time so these these are kind of
more general forces in society and not
just not just the the brainchild of
individual people this notion of
recursive self-improvement though is
where people say well this is where it’s
really different from anything else
that’s come before but in fact we try to
do recursive self-improvement all the
time and it helps but it’s not a game
changer so an example is we we know
teaching is really important this is how
we convey knowledge to people and you
can actually try to teach teachers to be
better teachers
there’s teaching schools that teach
teachers and that results in better
teachers but again not completely
different kinds of teachers than we have
now they’re just better you have ways of
coaching coaches to be better coaches
and you have of course ways of parenting
parents to be better parents the most
silly one that I’ve seen is you have
crowdfunding startups whose job is to
better fund crowdfunding startup so they
can found crowdfunding startups and
again this is it makes for progress it
just doesn’t make game-changing progress
and finally I want to address the
question of human obsolescence so as
machines get better at us at the things
that we care about does that make it so
that we don’t have a role to play
anymore and I can think of lots of
examples in technology from recent years
and an AI in particular but one example
that came up that I thought was
really telling is is the following thing
so once upon a time people decided it
was really important to draw sunflowers
and so on the caves that they were
living in at the time they came up with
techniques for drawing sunflowers and
they were they were very lovely
sunflowers very meaningful to people and
over the the the intervening centuries
people got better and better at drawing
sunflowers and by the 1800s individual
people could draw experts could draw
sunflowers in a way that was almost
indistinguishable from real life really
amazing artistry unfortunately around
the same time photography was invented
now now machines could make sunflowers
that were exactly as detailed as reality
they were capturing reality in
indirectly and people’s reaction at that
time some people said okay well folk art
is dead but other people said they don’t
know this is okay this is we just kind
of changed the way we were thinking
about what art is and so they came up
with Impressionism they came up with
with ways of incorporating photography
into their art they came up with ways of
using computers to draw sunflowers and
over time a thousand flowers would bloom
and as a result it didn’t really end the
pursuit of sunflowers what it really did
was it gave us new ways of creating we
got to change the rules we got to change
the goals and as a result we became
richer for it
so I want to mention one last little
example of of the impact of a eye on
society the real AI not this sort of
weird futures sci-fi AI but real AI has
the property that it really does impact
people’s lives so since the the early
days when researchers were just trying
to figure out what AI is it’s now
something that’s impacting millions of
lives and so one of the ways this has
happened is again if you take your your
direction giving software out in the
early days it had weird glitches in it
and you could ask the system you know
take me to an abortion clinic and have
it come back and say there are no
abortion clinics or there the only thing
I’d like to take you to is a crisis
pregnancy center where it which staffed
by anti-abortion people these systems
are actually changing the opportunities
and the affordances that people have as
they’re getting around in the world
and from that perspective it’s very
important that we take AI as it really
exists extremely seriously it’s up to
all of you to put pressure on on us the
technology creators to make sure that
the systems we build reflect our shared
values and this is a responsibility that
that we as technologists need to take
extremely seriously thanks a lot for
your attention
[Applause]
so back in 2004 I was driving my lovely
spouse to a conference she is an
obstetrician gynecologist and was
working as such at the time and the big
national conference in her field was
being held in Philadelphia now I’m a
academic computer scientist I’m a
computer science professor I’ve been to
lots of conferences I knew what to
expect and so I was sort of surprised
when I got there and discovered that
right there at the conference where the
attendees had to enter there were
protesters carrying signs and chanting
and I thought to myself that doesn’t
really make any sense this is a ob/gyn
conference they’re about trying to
improve women’s health I don’t see much
to protest here but the fact that matter
is that the stuff that they do in
obstetrics and gynecology is it touches
some some hot-button issues so they work
on abortion they do contraception some
of the protesters there were actually
protesting circumcision so there was I
don’t know why that’s an issue but
that’s that’s what they were concerned
about so they were marching around and I
thought to myself wow I’m I’m really
glad I’m not in a field where people are
protesting and getting all upset but
then I thought about it a little bit
more and I realized ways again what I do
is is studying artificial intelligence
we’re trying to make machines that have
the kinds of intelligence that people
have and if we’re successful there’s
very few things more threatening to our
identity as human beings as our ability
to think intelligently and to think at
all so you know if we were doing our job
well people would be upset probably the
fact that we don’t have protesters means
we’re just not doing a very good job
so just imagining a time in the future
when when in fact artificial
intelligence is being protested and I
thought okay well that would be a sign
that we’re doing better
well anyway flash-forward some 13 years
and this summer here in Providence the
annual meeting of of the governors of
the states in the US all got together
and
met and tried to talk about their shared
problems and one of their speakers one
of their invited speakers was a fellow
by the name of Elon Musk
now Elon Musk is a tech entrepreneur
he’s the head of SpaceX and also Tesla
Motors the electric car company and he
got in front of people and said that of
all the things that we should be worried
about his governor’s AI that’s the thing
artificial intelligence you should
really be very concerned about that it
represents a fundamental existential
threat like human extinction is on the
line if we don’t regulate it and and and
deal with these problems appropriately
so that’s that seems really concerning
he doesn’t he hasn’t he didn’t express
at the time what the argument was why it
why it was it was something that we was
afraid of but he said you know I’m
afraid and you should be very afraid now
I’m not afraid and I want to kind of
convey to you what the issues are why it
is that he’s concerned and why it is
that say I’m not concerned and give you
a sort of idea what the future it might
look like with respect to artificial
intelligence so I think I think the most
important thing to say right off the bat
is that there are at least two different
ways that people use the term artificial
intelligence so artificial intelligence
is a research field that we study in
computer science we we build robots we
make programs that make various kinds of
decisions and that’s something that we
study day to day and it’s been it’s been
underway for quite some time but
artificial intelligence is also a term
that’s used in science fiction to
represent this sort of notion of
machines that have become aware of their
environments aware of what their
capabilities are they can plan ahead
they can they can make their own
decisions and their own goals and and
pursue them and to be clear this this
first kind of artificial intelligence is
something that we have with us today and
it’s as Kip was mentioning it’s it’s
it’s hard to get it to get robots to do
interesting things to get software to
not be stupid but there’s also this
notion of artificial intelligence that
we might call artificial general
intelligence systems that are as smart
as people across the boards and this is
something that doesn’t exist you can
think of artificial intelligence is
being represented by the kinds of robots
that are stumbling around today
and you can think of artificial general
intelligence as maybe the kind of robots
that you might see in a Terminator movie
who are you know out to destroy us and
are very good at their jobs and they’ll
be back
so I think to understand this
distinction and understand why why I’m
not concerned part of it is is the
extrapolation from the AI that we have
now to this kind of hypothetical AI that
we that we don’t know how to build that
we don’t know that it exists so that it
could exist is is that is the main issue
that that I want to try to get across so
let me say a few words about what
artificial intelligence is if you don’t
know a little bit about the history so
back in 1956 a group of researchers
early computer scientists who are just
getting to think about what computers
can do we’re very interested in making
computers that that could exhibit the
kinds of computational processes or
thought processes that people have they
decided they’d get together at Dartmouth
College and they’d work it all out and
maybe over the course of a summer they
would solve this and then you know we
can all get on with with other things
and some of the problems that they
identified were things like how do you
make machines that use language how do
you make machines that can improve
themselves it can learn from their
experience these are great goals and in
fact they’ve kept us busy for some 60
years since then but they didn’t solve
it over the summer obviously but in fact
a lot of progress has been made and so
let me say a few words about various
ways that artificial intelligence is
with us today so one thing is you know
maybe you have an artificial
intelligence piece of software in your
pocket these these devices that actually
let you navigate to locations in the
world are just chock-a-block full of the
kinds of AI systems that we’ve been
designing the field has been designing
for decades so just to spell it out
maybe in a little bit of detail you
might want to ask your phone how to
navigate from here to power Street in
Providence and so to do that you know
you take out your phone and you speak at
it and what the phone receives is a
speech signal so this is the speech
signal of me actually saying take me to
power Street and that’s what it looks
like it’s it’s just a bunch of patterns
or numbers the problem of speech
recognition is the problem of turning
that that sig
that speech signal into words into
symbols and that’s a problem that’s been
with us for a long time and people made
a tremendous amount of progress on it in
the last you know five to ten years and
now speech recognition kind of works but
that’s not the end of the AI just in
getting directions you also need to turn
that that list of symbols into meaning
like what is it that you want the phone
to do and so this process of
interpreting the individual symbols the
language symbols and and and figuring
out what the intent is is sometimes
referred to as natural language
processing it’s another problem of
getting computers to use language and
this also is something where we’ve made
a lot of progress in the last few years
and now we can reliably recognize when
you ask the phone for something you know
what it is that you’re asking for maybe
not entirely reliably but but very
reliable enough so that we’ve fielded
this to millions and millions of people
so in addition to using language the
system also learns from its experience
and so the way that machine learning
comes into play in this kind of system
is when you’re when the system is trying
to figure out what path you should take
to get to power Street what it’s
actually doing is evaluating all the
possible different routes and figuring
out which one is going to lead to the
the fastest travel time to get you there
in the shortest amount of time to do
that it actually has to make estimates
it has to say okay here’s a little piece
of the street say South Water Street how
long is that going to take to traverse
now you can say well just use the speed
limit but that’s not exactly what people
typically do there could be traffic
there could be you know delays of
various sorts and so the ability to
predict based on features like well
where is this piece of road how big a
road is it what time of day is it what
day of the week is it all those things
get factored in together to generate
some kind of prediction and these
predictions are improved by feeding the
Machine lots and lots and lots of data
so anyway in the big picture all these
different little AI subproblems come
together to provide something that is
very valuable to people and it doesn’t
seem so scary right that’s not taking
over the world that’s not destroying us
in any way that’s kind of helping us get
from place to place what’s what’s the
missing step here and so the way that I
like to think about this
well if you think about a system that’s
actually say taking you from place to
place within a city maybe even helping
to drive you around that’s really
subsuming the capabilities of a human
job the job of being say a taxi driver
and so we don’t have this quite yet but
it’s really not hard to imagine in the
near future having a system that can do
all the things that a taxi driver can do
and to get you from place to place
and once that happens we have an AI
system that has all the capabilities
that it has plus it can be a taxi driver
okay well if it can do that what else
can it do it’s not too hard to imagine
that the job of say radiologist would
also be something that we could you know
absorb into the into the AI system a
radiologist takes data about you know
slides or x-rays taken in the human body
and it may can they can make various
kinds of diagnosis that seems like a
very a is sort of problem I think we can
do that one too we can again we don’t
have this yet but it’s it’s not hard to
imagine a time when the system has now
absorbed as well the job of radiologists
okay well what’s what would be next now
Here I am sitting in my lab maybe with
my students and I think to myself okay
well if we can do those things what else
can we do maybe we can maybe we can
absorb my job or maybe I wouldn’t think
about it that we might think I could
absorb the job of all my students
replace them with it with the machine so
we could we could all work together and
design a piece of AI software that can
do the things that an AI researcher can
do can we can come up with ideas for
designing programs we can make those
programs we can test them in the field
like yeah so so I can actually replace
all my my grad students with you know AI
researcher 2.0 and at that point then we
have a brilliant idea then we think okay
we can use our now collective
intelligence together to design even
better AI researchers AI researchers 3.0
now at that point it’s a little hard for
me to keep up with these these AI
researchers that I bought from the store
and they have improved themselves one
knotch and they begin to improve
themselves at a very rapid pace
potentially over the course of a weekend
the theory goes and we experience
something that sometimes referred to as
an intelligence explosion so this
intelligence turned back upon itself is
and more intelligence which can generate
even more intelligence we can ultimately
absorb all kinds of roles of people in
fact maybe all possible roles of people
you know including people who come to
TEDx talks you could imagine anything so
terrible at that point we would have
achieved what you can refer to as
general super intelligence so an
artificial intelligence system that is
smart not only is any individual people
but like anything that a person can do
this system can now do better than us
now this isn’t a threat in and of itself
but it does leave us humanity in a very
awkward position because we no longer
have control of our own destiny right
it’s now the machine that or the group
of machines that can do anything that we
can do better than we can do it and if
they decide that you know we need help
then maybe they’ll help us and that
would be awesome or if they could decide
that we’re in the way then they can get
rid of us and there’s really not much
we’d be able to do in response because
that’s how powerful they are in this
story so this is this is kind of the the
crux of the problem this is the big
concern is so just to put a fine point
on it
the idea is that we get computers up to
a certain point where they can exhibit
recursive self-improvement they can make
themselves better at which point they
can create entities of individual
brilliance the likes of which we’ve
never seen at that point the the fear is
that the things that make us special and
powerful the things that that humans can
do are now being taken over so humans
become obsolete and then finally as a
result we’re no longer the dominant
species on the planet we’re at the whim
of of these machines so that’s that’s
the story that that Elon Musk and others
would like you to be afraid of and it is
kind of a scary story right it’s kind of
straight out of Hollywood
because it’s kind of straight out of
Hollywood this is not coming out of the
science of it but it’s coming out of
speculative thinking about the
directions that these things can go and
I want to say a few words about why I
don’t think this is this is you know
it’s a great story but it’s not it’s not
a path that we’re about to follow
so the first thing I want to point out
is that if you ask a biologist about
who’s the dominant species on the planet
they don’t instantly say well it’s a
sitz people people are awesome I mean I
like people right some of my best
friends are people so it’s not it’s not
like I I think that people are doing a
terrible job at everything but it is
also the case that if you have if you’re
gonna say we’re the dominant species you
need to be careful about your
definitions and it’s the case that for
example in terms of which species have
lasted the longest well horseshoe crabs
have done a lot better job than us at
that maybe that’s not fair right they
had a head start
we’ve been maybe doing well since we’ve
gotten here but bite me again measured
by what measured by the individual
lifetimes well you know people live a
good long time but the the ocean Quahog
lives like four times longer than us so
yeah we’re good but there’s better but
there’s a lot of us but not nearly as
much as there are of Antarctic krill
they’re way ahead of us it’s like oh
that’s not fair they’re they’re tiny
right you know one of us is like I don’t
know a billion of those or something but
if you actually think if you factor
weight into this story while cows
actually beat us on that front there’s
more cow mass on the planet than there
is human mass of course that’s kind of
our fault right we actually gave cows
that boost so you could say what about
impact on the planet as a whole and we
have had a tremendous impact on the
planet but there’s other organisms that
have had arguably more impact on the
planet blue-green algae of all things
produces 330 billion tons of oxygen each
year a small fraction of which we
managed to breathe they’re just lots of
things out there that are having a lot
more impact on the world than we are now
again you know people are still very
important and we have a very big role to
play but but again you have to be
careful about defining your terms now
what about what about the impact of
intelligence on our own destiny if you
ask a historian
they don’t necessarily say that it’s
individual geniuses that have moved
everything forward they’ll give the
example of for example independent
discovery the idea that some brilliant
ideas actually are so important to
society that they appear in multiple
people’s heads that sometimes roughly
the same time and so if you actually
look at this
and I’ve been collecting examples of
these things because I think it’s really
neat there’s quite a range of issues
that actually come up the telephone very
impactful in society the invention of
calculus very impactful on society they
were independently invented by multiple
people really important things but also
some kind of trivial things the the
bacon martini was independently invented
by multiple people a mathematical
notation for specifying juggling
patterns for different groups had the
same idea at the same time for this and
so that’s kind of big and little but all
the way between two beer was
independently invented in multiple times
in history and the neural network
structures that are actually driving a
lot of the progress in AI have been
invented by multiple people over the
over the decades so it’s worth pointing
out that none of these things happen in
a vacuum even even vacuums didn’t happen
in a vacuum they were independently
invented by multiple people at around
the same time so these these are kind of
more general forces in society and not
just not just the the brainchild of
individual people this notion of
recursive self-improvement though is
where people say well this is where it’s
really different from anything else
that’s come before but in fact we try to
do recursive self-improvement all the
time and it helps but it’s not a game
changer so an example is we we know
teaching is really important this is how
we convey knowledge to people and you
can actually try to teach teachers to be
better teachers
there’s teaching schools that teach
teachers and that results in better
teachers but again not completely
different kinds of teachers than we have
now they’re just better you have ways of
coaching coaches to be better coaches
and you have of course ways of parenting
parents to be better parents the most
silly one that I’ve seen is you have
crowdfunding startups whose job is to
better fund crowdfunding startup so they
can found crowdfunding startups and
again this is it makes for progress it
just doesn’t make game-changing progress
and finally I want to address the
question of human obsolescence so as
machines get better at us at the things
that we care about does that make it so
that we don’t have a role to play
anymore and I can think of lots of
examples in technology from recent years
and an AI in particular but one example
that came up that I thought was
really telling is is the following thing
so once upon a time people decided it
was really important to draw sunflowers
and so on the caves that they were
living in at the time they came up with
techniques for drawing sunflowers and
they were they were very lovely
sunflowers very meaningful to people and
over the the the intervening centuries
people got better and better at drawing
sunflowers and by the 1800s individual
people could draw experts could draw
sunflowers in a way that was almost
indistinguishable from real life really
amazing artistry unfortunately around
the same time photography was invented
now now machines could make sunflowers
that were exactly as detailed as reality
they were capturing reality in
indirectly and people’s reaction at that
time some people said okay well folk art
is dead but other people said they don’t
know this is okay this is we just kind
of changed the way we were thinking
about what art is and so they came up
with Impressionism they came up with
with ways of incorporating photography
into their art they came up with ways of
using computers to draw sunflowers and
over time a thousand flowers would bloom
and as a result it didn’t really end the
pursuit of sunflowers what it really did
was it gave us new ways of creating we
got to change the rules we got to change
the goals and as a result we became
richer for it
so I want to mention one last little
example of of the impact of a eye on
society the real AI not this sort of
weird futures sci-fi AI but real AI has
the property that it really does impact
people’s lives so since the the early
days when researchers were just trying
to figure out what AI is it’s now
something that’s impacting millions of
lives and so one of the ways this has
happened is again if you take your your
direction giving software out in the
early days it had weird glitches in it
and you could ask the system you know
take me to an abortion clinic and have
it come back and say there are no
abortion clinics or there the only thing
I’d like to take you to is a crisis
pregnancy center where it which staffed
by anti-abortion people these systems
are actually changing the opportunities
and the affordances that people have as
they’re getting around in the world
and from that perspective it’s very
important that we take AI as it really
exists extremely seriously it’s up to
all of you to put pressure on on us the
technology creators to make sure that
the systems we build reflect our shared
values and this is a responsibility that
that we as technologists need to take
extremely seriously thanks a lot for
your attention
[Applause]
Please follow and like us: