Home
Connexion
S'inscrire
Contenu d'apprentissage
Loading...
Why AI Is Unlikely to Become Conscious | Anil Seth | TED - Video học tiếng Anh
Pratique d'écoute
Pratique d'écoute
/
Video
/
TED Talk
/
Why AI Is Unlikely to Become Conscious | Anil Seth | TED
Why AI Is Unlikely to Become Conscious | Anil Seth | TED
Choisir le mode d'apprentissage:
Voir les sous-titres
Choisir le mot
Réécrire le mot
Highlight:
3000 Oxford Words
4000 IELTS Words
5000 Oxford Words
3000 Common Words
1000 TOEIC Words
5000 TOEFL Words
Sous-titres (282)
0:04
So for centuries,
0:05
people have fantasized about playing God
0:08
by creating artificial versions of ourselves.
0:11
From Mary Shelley's "Frankenstein"
0:14
to Hal in Stanley Kubrick's "2001"
0:16
and Ava in Alex Garland's "Ex Machina,"
0:19
this is a dream reinvented with every breaking wave of technology,
0:23
and with AI, the wave is a big one.
0:25
The AI we have is already smart, at least in some ways.
0:29
But could it ever be conscious?
0:33
Will a robot ever gaze at a sunset and experience the beautiful colors,
0:37
the reds and the oranges?
0:38
Will it feel a sense of beauty or a rush of joy?
0:43
Or will computers, however smart they get,
0:46
always remain dark on the inside,
0:50
always an object and never a subject?
0:54
Whether AI can be conscious is one of the most consequential questions
0:59
we face in our time.
1:01
If computers can be conscious or sentient or aware,
1:03
we'd be entering a new era in human history.
1:06
We'd have new entities that have their own inner lives,
1:10
new inventions that matter for their own sakes
1:13
and not only for their effects on us.
1:16
Conscious AI might suffer the click of a mouse,
1:19
perhaps in ways we wouldn't even recognize.
1:22
And if silicon can be sentient,
1:24
then maybe our messy flesh-and-blood bodies
1:27
will soon be superseded by machines that never age
1:31
and cannot die.
1:34
Now over the last few years,
1:35
progress in AI has been simply astonishing.
1:38
And who knows what's around the corner.
1:40
Many experts think that conscious AI is possible.
1:45
Quite a few think it's inevitable, and some, some think it's here already.
1:50
I think they're wrong.
1:51
I want to tell you why and why this matters so much.
1:55
So I've been studying brains, minds and consciousness for nearly 30 years,
1:59
and one thing I've learned is that to answer the question,
2:02
"can AI be conscious,"
2:04
we need to start by looking within ourselves
2:06
at the makeup of our own human minds.
2:09
Now we humans, we tend to see the world in our own terms.
2:12
We know we’re conscious, and we like to think that we’re intelligent.
2:16
So we think the two go together.
2:19
And this is why some people think
2:21
that consciousness might just glimmer into existence
2:24
as AI gets smarter and smarter.
2:27
But consciousness and intelligence are different things.
2:30
Intelligence is all about doing.
2:32
It's solving a crossword puzzle,
2:34
assembling some furniture, navigating a tricky family situation.
2:38
Consciousness, on the other hand, it's all about feeling and being.
2:43
It's the difference between normal, wakeful awareness
2:46
and the oblivion of general anesthesia.
2:49
It's the bitter tang of coffee,
2:51
it's the warmth of a log fire.
2:53
The joy of seeing a loved one.
2:56
Just because consciousness and intelligence go together in us
3:00
does not mean that they go together in general.
3:03
The assumption that they do,
3:05
well, that's a reflection of our own psychology,
3:07
not an insight into the nature of reality.
3:12
Take language models like Claude or GPT,
3:15
trained on vast quantities of written texts.
3:18
They reflect back to us an image of ourselves,
3:20
of our collective, digitized past.
3:23
We talk about ourselves endlessly,
3:25
and so do they.
3:27
We wonder about consciousness and the meaning of it all.
3:31
And so, it seems, do they.
3:33
But language models are not conscious.
3:37
They simulate consciousness.
3:38
We project consciousness into them
3:41
in the same way we might project faces into clouds,
3:45
or even the image of Mother Teresa in a cinnamon bun.
3:48
I don't know if you can see that.
3:50
There she is, Mother Teresa in a cinnamon bun.
3:52
(Laughter)
3:54
We are built to be seduced, like Narcissus, by our own reflections.
3:59
And so we see ourselves in our algorithms.
4:03
I'm always struck that nobody really worries
4:05
whether DeepMind's AlphaFold is conscious.
4:08
AlphaFold predicts the structure of proteins
4:10
rather than words and sentences,
4:12
but under the hood, it's not much different from Claude or GPT,
4:16
algorithms running on silicon and trained on vast reservoirs of data.
4:20
AlphaFold just doesn't pull our psychological strings in the same way.
4:24
So if we think that Claude is conscious, but AlphaFold isn't,
4:28
that says more about us than it says about AI.
4:32
(Applause)
4:36
But how can I be so sure
4:37
those systems, like Claude or GPT, aren't conscious?
4:41
Well, nothing is for certain when it comes to consciousness,
4:44
but the very idea of conscious AI depends on a deeper assumption,
4:48
a kind of myth, really.
4:49
And this is the myth that the brain is a computer
4:52
that just happens to be made of meat rather than metal.
4:56
Now consciousness in this story is a special algorithm,
4:59
a collection of computations
5:01
that just happens to be carried out in the wetware of the brain in us,
5:05
but which could equally be carried out in silicon, in AI.
5:10
But the computer is just one
5:12
in a long line of technological metaphors that we've reached for
5:16
when trying to understand the deep complexity of the brain.
5:21
One time, the brain was a system of plumbing.
5:25
Later, it was a telephone exchange.
5:28
And for the last few decades, it's been a computer.
5:30
And this most recent metaphor has been extremely powerful.
5:34
But it is still a metaphor.
5:37
And we will always get into trouble
5:38
when we confuse a metaphor with the thing itself.
5:42
The map with the territory.
5:45
For one thing, in a real brain,
5:47
there's no sharp separation between the mindware and the wetware.
5:52
Unlike the separation that you get between software and hardware in a computer.
5:57
And this really matters because in a computer,
6:00
you can describe and understand everything about an algorithm,
6:03
whether it's a language model or a word processor,
6:05
without worrying about all the silicon shenanigans
6:08
going on underneath.
6:09
The computation, the algorithm is all that matters.
6:13
But for brains,
6:14
you just cannot separate what they do from what they are.
6:19
And this means that what they do
6:20
is unlikely to be a matter of computation of algorithm alone.
6:25
Look closely at a brain, at any brain,
6:28
and it becomes less and less plausible that all that's going on
6:31
is just turning some numbers into other numbers
6:34
in this endless dance of zeros and ones.
6:38
Yes, there are neural circuits
6:40
which exchange signals and may do computation,
6:42
or at least something like it.
6:44
But there's so much else that escapes the confines of the digital.
6:48
Neurotransmitter chemicals course through the brain circuitry,
6:52
electromagnetic fields sweep through the cortex like weather systems.
6:57
Even a single neuron is such a beautiful biological machine.
7:02
A far cry from the simplified, cartoon-like neurons
7:06
that power today's AI.
7:09
The brain is not, or at least not just, a computer made of meat.
7:14
And so consciousness is very unlikely to be a matter of computation alone.
7:19
And if this is true, then conscious AI is off the table,
7:23
at least for AI as we know it today.
7:26
Let me put it another way.
7:28
What if we simulated every last detail about the brain
7:32
in some massive supercomputer?
7:33
Now if the fine details of the brain do matter for consciousness,
7:37
well, wouldn't this be enough
7:38
for consciousness to happen inside a machine?
7:41
Well, a computer simulation of a hurricane does not create real wind.
7:46
[A] computer simulation of a black hole doesn’t suck the Earth
7:49
into its algorithmic singularity.
7:52
Making these simulations more detailed can make them more useful,
7:56
but it does not make them any more real.
8:00
We can have a simulation of the brain,
8:02
and you can make it as detailed as you want.
8:04
This might make it more useful,
8:06
but it's not going to make it any more conscious.
8:11
Now consciousness, has to be said does remain a bit mysterious,
8:14
but perhaps one reason for this
8:16
is that we've been so constrained by our metaphors,
8:19
by the idea that it just has to be some kind of information processing.
8:24
After all, if you think the brain literally is a computer,
8:27
then what else could it really be?
8:30
But once we see the brain more clearly for what it really is,
8:34
many new possibilities arise.
8:36
And my own view, developed over many years,
8:39
is that consciousness is intimately connected to our nature
8:42
as living creatures.
8:44
Unlike the abstract universe of computation,
8:48
life is all about materiality.
8:50
Unlike algorithms,
8:51
living systems are deeply embedded in flows of energy and matter,
8:56
and they continually regenerate their own conditions for existence
9:00
and for persistence over time.
9:03
I think we can draw a direct line from the molecular furnaces of metabolism,
9:08
one billion biochemical reactions in every cell, in every second,
9:13
all the way to the neural circuits
9:14
that underlie each and every experience that we have,
9:18
whether it's the sight of a blue sky or a pang of envy.
9:23
Every conscious experience is imbued, however subtly,
9:27
with a tinge of aliveness,
9:28
with some core relevance for our future survival prospects.
9:32
And at the heart of every experience, beneath even emotion, is this simple,
9:37
shapeless and formless but fundamental feeling of being alive.
9:43
And in this story,
9:44
it's life, not computation,
9:47
that breathes the fire into the equations of experience.
9:51
And if this is right,
9:52
then conscious AI will need to be living AI.
9:58
Let me bring things together.
10:00
First, we're built to see consciousness where it isn't,
10:03
thanks to deep-seated psychological biases
10:06
that bundle language, intelligence
10:08
and consciousness together.
10:10
Second, the brain is not, or not just, a computer.
10:13
So consciousness is unlikely to be
10:15
a matter of computation of algorithm alone.
10:19
Brains are the kinds of things for which you can't separate
10:21
what they do from what they are.
10:23
And silicon is not up to the job.
10:26
And third, many other things about our biological brains and bodies
10:30
might matter for consciousness,
10:31
including a deep connection between consciousness and life.
10:36
Artificial intelligence is computer software.
10:40
It is not a living mind.
10:42
It might give the impression of being conscious,
10:45
but it is vanishingly unlikely that it actually is.
10:50
I want to close by returning to why this matters so much.
10:54
Take the idea of AI welfare.
10:58
Now there are already influential groups advocating
11:00
that AI systems should have their own rights
11:03
based mainly on the idea that they might be or become conscious.
11:08
Now if real artificial consciousness is on the way,
11:12
maybe through some other technology or other pathway,
11:14
then this is entirely justified.
11:16
We humans have a terrible track record
11:19
in our ethical treatment of non-human animals and of other humans,
11:22
and we don't want to make the same mistake again.
11:25
This is one reason why even trying to build conscious AI is a very bad idea.
11:33
But if conscious AI is just an illusion created by design,
11:37
as I think it is,
11:39
then by extending rights to these systems,
11:41
we’d be sacrificing our ability to control, to regulate them
11:45
and perhaps even to turn them off and for no good reason at all.
11:49
And this is one reason why even AI that merely seems to be conscious
11:53
is very dangerous for our society, too.
11:56
And unlike real artificial consciousness,
11:59
conscious-seeming AI is either already here
12:03
or coming very, very soon.
12:06
There are other reasons we should avoid creating AI that seems to be conscious.
12:10
It makes us more psychologically vulnerable.
12:12
We might be more likely to do what an AI says
12:15
if we believe that it really feels for us,
12:18
that it really understands us,
12:20
even if what it's telling us to do is very bad for us.
12:24
And finally, the very idea of conscious AI undermines our human nature.
12:30
The mirror of AI goes both ways.
12:33
We see ourselves in our algorithms,
12:36
but we also see our algorithms in ourselves.
12:40
And when we do,
12:41
when we think of the mind as a collection of computations
12:44
floating free from their bases in biology,
12:47
we reduce and we diminish what it is to be a living,
12:50
breathing human being in a real world.
12:54
(Applause)
12:58
"Frankenstein,"
13:00
which Mary Shelley wrote when she was just 19,
13:03
is often taken as a cautionary tale,
13:06
a warning against the hubris of bringing something to life,
13:09
a Promethean sin like stealing fire from the gods.
13:14
The idea of conscious AI is a new Promethean dream
13:18
wrapped up in a silicon rapture.
13:22
And if conscious AI is possible,
13:25
then so is the prospect of uploading our own conscious minds
13:28
and floating off to eternity in a silicon cloud of living,
13:32
or at least existing forever
13:34
in the pristine circuits of some future supercomputer.
13:39
Now the seductive power of this vision --
13:41
of being at such a pivotal point in the history of life on Earth --
13:45
I suppose it's understandable.
13:47
And back in the real world, talk of conscious AI does other things too --
13:51
it conjures this sense of technological wonder and magic
13:55
which might keep share prices aloft and regulators at bay.
14:02
But we should resist.
14:04
The sacrament of the algorithm is most likely an empty dream,
14:08
delivering not post-human Paradise, but silicon oblivion.
14:13
We need a different story.
14:16
One in which we're more part of nature,
14:18
not apart from it,
14:20
with consciousness more closely tied to living flesh and blood,
14:23
not to the dead sand of silicon.
14:27
AI might claim the prize of intelligence, at least in some ways,
14:31
but consciousness,
14:32
consciousness remains ours to celebrate
14:35
and to share with other living creatures.
14:39
So let's not sell our minds so easily to our machine creations.
14:43
If we do,
14:44
we not only overestimate them,
14:47
we underestimate ourselves.
14:49
Thank you.
14:51
(Cheers and applause)
14:55
Thank you very much.
