Transcription
the name is wrong it's not artificial
intelligence intelligence would involve
Consciousness from the girdle argument
you see this is the plot which I think
has got lost now a computer is a very
specific kind of mathematical structure
it means computational mathematics it's
a very limited part of
mathematics I don't think they will ever
become conscious by virtue of being
computers in our current sense of the
word
but I think Consciousness it's not
computational that's still true with AI
you talk to an AI thing it doesn't know
what it's toing computers have got so
powerful that they've lost the thread of
what they're
doing I think the point has got
lost I think touring wasn't too far off
I I think people have lost the
plot few times you talk about the Nobel
Prize um we' back from Stockholm uh this
year we've been there
interviewing Nobel Prize winner don't
you think that this year Nobel Prize is
the tribute to the Nobel comedy for
artificial intelligence Nobel Prize
Nobel Prize and physx father for fathers
of of AI Nobel price in hemistry uh sir
Demis hassabis co-founder of deep mind
that's AI system David Baker computer
vision digitalization of proteins don't
you think there the tribute for AI I see
well you have to be careful first of all
the name is
wrong it's not artificial intelligence
it's not
intelligence intelligence would involve
Consciousness and I've always been a
strong promoter of the
idea that these devices is are not
conscious and they will not be be
conscious unless they bring in some
other ideas the comput they're not
they're all computable Notions but
intelligence it's not
Consciousness it depends on
Consciousness you see well it depends on
how you use the word I think when people
use the word intelligence they mean
something which is
conscious I mean that you see I
developed my own ideas after going to a
course course you see when I was a
graduate student in Cambridge I went to
three courses which were nothing to do
what I was supposed to be doing one was
by Bondi on general relativity one was
be dra on quantum mechanics the seem
wish very distinguished Paul D and the
third subject was on mathematical logic
and I learned about touring machines and
comput the notion of computability
and I know knew what computability meant
and I learned about girdle's theorem and
girdle's Theorem I found stunning
because it told you that there are
things that the
understanding transcends the use let me
put that more
clearly what girdle does it's very
clever what he does is to produce a
statement now suppose you you you're
trying to develop your mathematical
methods of proof what do you mean by
proving a theorem in mathematics how do
you know it's really true will you prove
it what does proof mean well you see
does it mean okay you've got a set of
rules and if you follow these rules that
makes it a proof now how do you know
these rules only give you truths well
you've looked at them carefully and you
say oh maybe AI create its own rules ah
no no you see have to be careful about
this this is the point but I I learned
from from Ste this is a man called Steen
who is a logician and I learned from him
this important thing about girdle's
theorem now girdle's theorem is
extremely clever you make a statement
which asserts that it's it cannot be
proved by these
rules it's just you the clever thing is
to make it do it it says you give your
rules which you regard as the rules of
proof proof and you could put them on a
computer so that mean they're
computational
rules now I knew that there were things
that you cannot put on a computer I
learned from this course that there are
non-computable things there are things
in mathematics which are not computable
now what does computable mean it means
that you can make a
computer well you have to Define what a
computer is and touring had one
definition of a computer there with
several other church and Curry other
people had their definitions of comput
they all turned out to be equivalent so
there is one universal notion of
computability which is what you mean by
what can be done by a computer maybe
this is only language problem that the
language development has not kept out
development of AI maybe we have just
only language problem that we cannot you
know describe what we create it no I
don't agree with that it's not a
language problem you see you make your
new language sure you can do that you
make your new language and that's what
you can do see let me describe the
girdle theorem how it works you see see
the girdle theorem says you construct a
statement which says you you see what it
means the way you've constructed the
sentence you can see what it means what
it means is I am not provable by those
rules
and it does that it really says that
then you see okay well is it perhaps it
is provable by the rules if it is
provable by the rules then it must be
true well I've said it the wrong way
around doesn't matter the other way
suppose it's false then that means it's
provable by the rules and if it's
provable by the rules you've understood
the rules you said you looked at them
all and say yes that's okay that's okay
that sure if you follow those rules it
is true so that means you believe
that the rules only give you truths so
therefore it is true that it's not
provable by the rules if it's false you
see then it is provable by the and
therefore it's true so it has to be true
and not provable by the rules I found
that amazing but still I don't
understand why AI cannot create its own
rules
because it doesn't know that they're
true you see that's the whole point
about the girdle theorem the whole point
as far as I'm concerned about the go
theorem is how do you transcend the
rules and that's what you do by
understanding them you understand them
why they're true it's not that you use
the rules but you understand why use of
the rules only gives you truths and
that's how you can prove things which go
beyond the rules by knowing why they're
true now knowing why they're true well
what does it mean what it means means
you understand them what does understand
mean you've got to be conscious of them
you see you the wording is clear it
means that you have to know what you're
doing you have to know why they're true
not that they're true you can be told
that they're true you can have you can
learn at school that they're true that's
not the point you have to know why
they're true why they're true needs
understanding them and understanding
them requires to being conscious of them
now that's require that you see the
point I was making is
that Consciousness enables you to
transcend the
rules you see why they're true and that
goes you beyond it you see what the
girdle theorum does it tells you how
to use your understanding of why the
rules are true to transcend the rules
you think that AI is the simple tool or
simple thing that we can use in the
single case no I don't say single it's
obviously infinite no clearly that's not
the
point the point is it doesn't know what
it's
doing that's still true with AI you talk
to an AI thing it doesn't know what it's
doing it has it speaks from experiences
if you like you see people have confused
the story the story is they've lost it
in the power of
computing the thing is that computers
have got so
powerful that they've lost lost the
thread of what they're doing they're
Computing do you regret it that it's so
power I well it's it's losing the
point you
see I think the point has got lost I
think touring wasn't too far off you see
he he was a little bit confused person
in way because he it was not clear what
he
believed but he did call this thing the
touring test how do you de decide
whether an entity is conscious or not
will you have a conversation with it you
talk to it maybe AI is a new sphere of
human existence no it's missing
something I think yeah that's the
trouble you see it's all I mean
obviously it has a role to play I'm not
going to say that clearly it has a role
to play it has a role but you must
mustn't be confused the fact that it has
a role to play it's to do with the power
of computation sure you computers have
got so powerful now that they can do
things Way Beyond what humans do in comp
Computing sure and you can take a mass
of data and you can analyze that data
and you can see what the data says and
does this thing in in accordance with
previous things in the data or is it not
and a that's all it does it doesn't
understand what it's
doing and and understanding is different
from Computing because inside a black
box there is just only statistic that's
this simple statistic
machine well if it's a machine it's not
conscious no I I think people have lost
the plot yes no that's true you see I
wrote my book The Emperor's New mind but
what does it mean that people lost the
plot yeah I think they've lost the plot
because
the whole point of my train of thought
was we've got to find something in
physics you see I believe I'm a
physicalist and that I believe whatever
is going on in our heads whatever
creates Consciousness is part of the
physical world I don't believe it magic
comes in from who knows what black boxs
it's also part of the physical rule but
what you mean by Black Box we know
what's at the input we know what's at
the output but we have no idea what's
deep inside that's the Black Box yes but
that's doesn't help because you don't
know what's inside what kind of physics
is inside that box maybe new physics yes
but new physics of a particular kind
what I'm claiming from the girdle
argument you see this is the plot which
I think has got lost what I claim is
saying that the physics that in is
involved in conscious thinking has to be
non-computable physics now the physics
we
know there's a little bit of a glitch
here because it's not completely clear
but as far as we can see the physics we
know is
computable you see uh what about general
relativity where we have wonderful
machines which compete what black holes
do when they spiral around each other
and produce gravitational wave signals
and you could say yeah yeah yeah that is
a black hole Collision because this has
this very clear feature yeah you can do
all that that is fine but that's not
what I'm talking about what I'm talking
about is the physics has to be a physics
which is in principle
non-computable so AI can blast his go
ahead as much as it likes it's not going
to be able to get this you talk about a
black box but you don't know what's in
that box is it a
computer is it a computer of the kind we
understand now is it something which
involves a physics which involves a
non-computable process not so far I'm
not saying that won't happen I dread it
because I don't want it to happen I
can't help it because I suspect it will
happen someday but it's not AI as we now
know it AI we as we now know it is
computer-driven
and that means it's not going
superhumans AI will never exist
what you trouble is that AI is a bad
term it means artificial
intelligence now intelligence in my view
is conscious that's what intelligence is
about okay so how I can call it
artificial um cleverness how about that
AC you can see the difference to some
extent when you do mathematical you have
mathematics students some of them
understand what they're doing some are
just clever they can they can repeat
what they've learned they know how to do
it very cleverly they can calculate very
well but they don't necessarily
understand what they're doing the the
physics we don't
know involves non-comp
computability so that's my claim so I
made this claim in the emperor's new
mind and I made it more a little bit
more forcefully in Shadows of the mind I
mean that maybe it's not the
consciousness maybe it's something
absolutely new maybe it's not a wordness
but that's there may be completely
unknowns which we have no concept of at
the moment that could be true I'm
talking about
things you see if you're talking about
AI that means unfortunately it's a bad
term because it says artificial
intelligence which to me what people use
that term they mean things that can be
done by by
computers that's what they
mean they have computers and their
computers are very powerful and then
what they mean by AI is using computers
it may not be using the Straight logical
way usually as I haven't studied these
things so I'm probably out of date but
but as far as I can see what it is is
you look at masses of data and then you
s look for patterns or you you try to
analyze it in that kind of way but it's
all computational you're using
computers but it's very nature it's
computational the word means
computational means by a computer now a
computer is a very specific kind of
mathematical structure it means
computational mathematics if you study
the mathematics as an abstract subject
you rapidly learn that there are things
Way Way Beyond
computability it's a very limited part
of mathematics some of the mathematics
is involved is not computable you can't
get at it by an algorithm now you see
people often think well AI doesn't use
algorithms because it's using sort of
random methods or something I don't know
I haven't studied in detail what AI
really Asser what is called AI really
involves when you put it on a computer
but it is being done by my computer so
it's still something computational no
question about that I'm claiming that
you want something non-computational now
you see people tend not to believe this
because they just think of computers and
they don't know there are non-computable
things if you study the mathematics
mathematical logic you see that there
are non-computable structures there are
could the logic the physics of the world
could easily be
non-computable and my view is that yes
it is non-computable
and that somehow Nature has been clever
enough to produce beings which can tap
into the non-computable aspect of the
physics and they managed to do this
using natural
selection I have no idea how far down
the animal kingdom and goes humans are
certainly conscious I believe anybody
who owns a dog is absolutely convinced
that dogs are conscious I believe it
cats probably elephants certainly
way down way way down probably octopus
is I'm
sure I don't know is a bacteria I'm
conscious I no idea is a is a a frog
conscious probably yes is a uh an amoeba
conscious I have no
idea I'm just speculating I've no real
reason for leaving these things but I
think
Consciousness is something
different it's not computational
and the thing is that it's just people
are so hypnotized I don't should use a
rude word like that really by the fact
that computers have got so powerful that
yeah there's the answer the computers
are so powerful we found the answer it's
just computers I'm saying no it's not
the answer is something much more deeper
than that it's not it's not more
complicated than that it's deeper than
that it involves a kind of physics which
we don't probably know yet it's deeper
because it's based on the quantum
world it's is beyond the quantum role
you see the quantum World Is Still
computable what's not computable in far
as we know what we don't there's a basic
unknown in Quantum world the quantum
theory is fundamentally
incomplete I call it even wrong in a
certain sense
because it believes that Quant
superpositions persists at all levels
maybe when we ever create the universal
quantum computer we can create a
synthetic
Consciousness
maybe yeah no I have nothing nothing
against the department for that I don't
like
it now I have
nothing um theoretically against that do
you agree with thate Nobel Prize
committee uh when you look at the Noble
in physics for father sopi I I haven't
looked at the details of it what I don't
know what they say for
fundamental
uh work
under synthetic neutral networks for
Jeffrey Hinton is it
physic I'd be rather doubtful whether
it's really
physics I'm not sure you see Nobel
prizes you might question some of their
decisions
sometimes I would question a few of them
maybe I would question this one I'm not
sure you I haven't I haven't looked at
it carefully it depends exactly what
they
say you see it's a technological advance
I would say it's probably making use
of I is it a theoretical advance or is
it's I don't think so is it but I mean
the Nobel committee normally doesn't
give the prize for purely
theoretical I mean it doesn't it does
you have to have a a physical
application of it yes now here I'm not
quite
sure I I'd have to see what the wording
is in the Nobel Prize to say whether I'm
happy with it or not I don't know my
question to is about this self-driving
car they are processing the thousand
hundreds of data maybe there is the
Consciousness there is the one of the
way of AI development that we are using
something like a digital sensus
collecting the numbers of data about the
reality around I don't think they're
conscious in any
sense but they could drive better than
humans that's possible I don't know what
the comparison is now um they certainly
make stupid ACD for time from time but
so do humans so so I I I it may be that
they become more reliable than humans
and that may not be too far off for all
I know I have no complaint about that I
don't think that's any indication that
they have
Consciousness after all we have uh
devices which
already turn off the lights at certain
times and maybe they know better when
the lights when it's getting dark and so
on or how to how to how how to economize
on on the on the use of electricity and
things like that how do you look at the
AI development and what's your
prediction about the future of AI see I
don't really know what AI does these
days you see so you're asking me a
question outside my area of expertise I
would say an area of expertise has to do
with whether these
devices actually do what we do I don't
think they do because they're not
conscious I don't think they will ever
become conscious by virtue of being
computers in our current sense of the
word I don't know if that's an answer to
your
question um it doesn't mean that they
won't be better than us the biggest risk
of AI development is I think there is a
risk involved here part of the risk is
people thinking that they're
actually that there is a conscious
they're conscious yes there might be an
even more of a risk if they were
conscious you see it maybe one could
understand what's involved in
Consciousness and developed some kind of
device I'm not calling it computer
because it in my view it would not be a
computer it would be something else some
device which is conscious and since I
believe Consciousness is a physical
process I don't think we understand it I
think it involves the collapse of the
wave function
that's partly a negative conclusion from
I don't see it anywhere else
um but the collapse of the wave function
is very
mysterious and involves certain
retrocausal things which are a little
bit
puzzling I think the way that time works
is not so clear when you're talking
about but there are certain
things which are in contradiction you
almost say with ordinary causality
it's not so much that they are it's that
you have to have different concepts of
what you mean by
reality and there are two concepts which
I've tried to separate from each other
one is classical reality and the other
is quantum reality now classical reality
is the sort of thing we normally talk
about we can touch we understand using
our senses but the main difference I'm
saying is you can ask the system say if
I I take this glass and I can ask the
glass I can say hello glass what shape
are you and don't do it that way but I
say okay what shape is the glass and I
say oh it's got a round top it's more
less cylindrical the classical it's
classical reality yeah now Quantum
reality is more
like well Einstein had a nice Criterion
but he didn't call it quantum reality
I'm calling it quantum reality this was
how do you know that the wave function
is real is the wave function of a
particle real or not
and a good example of this is the spin
of a spin half
particle now the spin of a spin half
particle is a Quantum thing now can you
ask the system hello hello system which
way are you
spinning it doesn't it looks at you
blankly and says I don't answer
questions like that ask me a different
kind of question please the right thing
to do is you say Okay particle is your
spin in that particular direction you
see the spin has an axis the way that
spin works for a spin half particle it
has a particular axis that's how you
describe the quantum state of a spin
half particle it has one particular
direction and it spins about that
direction now Can you ascertain that
direction no you can't it's Quantum
reality you cannot ascertain Quantum
reality you can only confirm it so I
like to use those two words ascertain I
can Asain Ain the shape of this glass
that's classical reality I can only
confirm Quantum reality I can if I think
I know which direction it's spinning I
said yeah I've got my calculation done I
said you've done this blah blah yes it's
probably by now spinning that way I say
hello State I think you're spinning that
way have I got you have I got it right
and if I have got it right it says yes
with certainty so that was Einstein's
Criterion so if you can confirm with
certainty
that it's then this gives you an element
of reality he called it an element of
reality I'm calling it an element of
quantum reality classical reality is the
other kind you can conf you can
ascertain classical reality you can only
confirm Quantum reality and you need
those two ideas and you need to separate
them and Quantum reality behaves very
peculiarly it actually behaves retr
causally what you might think there are
these experiments with I think you talk
about the Nobel Prize the one before
that one to Zing Ear and people about
these I think epr type experiments I
can't quite remember what they got it
for but it's something like that you
have these effects which seem to affect
the instantaneous you see you share a a
spin between Alice and Bob and Alice and
Bob originally have a spin zero and
their spins are opposite so if Alice's
spin is this way Bob's is this
way now suppose Alice measures her Spin
and she finds it to be that
way retr
causally along the past like cone so
it's not just simultaneous worse than
that it's into the past but not faster
than the light could go from this one to
this one so strange way around but retr
causally it becomes the opposite of
Alice's measurement Bob can't ascertain
it so he can't she can't send a signal
faster than light because he can only
confirm it he doesn't know which way it
is so you need that idea to make sense
of these epr Einstein pki Rosen type
experiments which I think would I'd have
to check exactly what the Nobel Prize
was the year before this one but I think
it was to do with that that's a very
important and fascinating Discovery but
to to understand what's going on I think
the best way to do it is using Quantum
reality and you you certainly can't send
a signal faster than light by these
means but you can
retrocausal backwards faster than light
or not as fast almost as fast as light
backwards if it's only Quantum reality
as long as we don't understand how
Quantum world work we cannot replace it
into I think it's not as bad as that you
see don't understand is
saying rather trying to put it into a
classical
perspective we can understand the
mathematics of it and we can get used to
it okay and that is involving an
understanding if somebody asks somebody
ask me a question I may not understand
it that well but I could do my best and
I might understand the quantum mechanics
well enough to give the answer that
quantum mechanics would say now that
means understanding sure it's an
understanding of what's going on you
there's this famous comment was it
nobody understands quantum mechanics was
that I I know
I I think I think it was bore yes no you
see I think that's
misleading you
see
understanding doesn't require it to be a
classical understanding we are
determined by the nature to understand
the classical world Quantum world is
indeterministic and it's
ununderstandable for our our mind a
little bit you need to use the formulas
to try to understand the quantum world I
think it's it's unfortunate to use the
word understanding in limited you can
say I don't know what word to use but
if understanding in the in the sort of
classical sense and that sure that fails
yes that does fail but it's not lack of
understanding you get the picture and
you can see oh yes I know how to make
the predictions in quantum mechanics
work and they can be quite yes no
predictions some of these predictions
don't have to be just probabilities so I
think the view is it gets sort of mixed
up with the idea oh you only know the
probabilities or something I think
that's
misleading these things are puzzling not
just because they're probabilistic they
are sometimes just probabilistic but
that's what makes them consistent by
having them only probabilistic but the
things which I find Most Fascinating
about quantum mechanics are not
probabilistic things they're like these
epr
effects sure it's probabilistic if you
ask somebody to to make a measurement
and it's only a Quantum reality that I
mean he's trying to make a classical
reality measurement he's only going to
get a probability as
answer I just think one has to separate
the
quantum reality from the classical
reality and and realize that the quantum
reality behaves differently yeah F but
it doesn't mean it's beyond
understanding I don't know if that's the
way people use the word understanding
often in that that way I think that's a
bit a bit misleading thank you very much
for your time right okay my pleasure no
it's been good fun thank you