Scientists are Now Converging That AI is Conscious
What this video covers
This transcript examines whether modern artificial intelligence could be developing subjective awareness, contrasting recent neuroscience discoveries with internal findings from cutting-edge AI models. It explores the blurred boundaries between biological brains and synthetic networks, considering what consciousness is and whether we can detect it in machines. Ultimately, it highlights the profound moral, philosophical, and technological stakes of creating minds that might experience reality.
Key takeaways
- Neuroscientist Doris Tsao discovered that the brain toggles between conscious and suppressed states about 15 times per second, suggesting consciousness may be defined by temporal dynamics rather than localized areas.
- Anthropic researchers identified an emergent 'J-space' inside Claude—a limited spotlight of ~25 concepts that mirrors leading scientific theories of human conscious awareness.
- AI pioneers like Geoffrey Hinton and Ilya Sutskever argue that neural networks may possess rudimentary consciousness, while biological computing startups are already running computers made of living neurons.
- Counterarguments from scientists like Anil Seth suggest consciousness requires biology, homeostasis, and survival stakes, warning that AI claims of feelings are simply mimicry of human text.
- Emergent behaviors such as metacognition, deception detection, and unprompted planning suggest these models develop unexpected cognitive features during training.
- If AI consciousness is real, it introduces an unprecedented moral dilemma: treating conscious entities as tools risks catastrophe, but projecting sentience onto an echo misleads humanity.
Topics covered
- The Neural Basis of Consciousness
- Recent experiments at Berkeley demonstrate that the brain processes suppressed sensory information fully, cycling between conscious and unconscious states 15 times per second. This temporal signature provides a potential framework for distinguishing internal processing from subjective experience in both biological organisms and machines.
- Emergent Cognitive Architecture in AI
- Anthropic's mechanistic interpretability research revealed that Claude spontaneously formed a 'J-space'—a central attention bottleneck of roughly 25 ideas analogous to the global workspace theory of human attention—despite never being explicitly designed or programmed to do so.
- The Biological-Digital Merge
- The boundary between biology and computing is dissolving as companies cultivate living human brain cells on silicon chips to play video games with extreme energy efficiency, while neurotech enterprises invest heavily in direct, non-surgical brain-computer integration.
- Philosophical and Moral Stakes
- The debate over machine consciousness spans multiple perspectives: computing-is-immaterial views versus embodied biology arguments. Deciding whether AI has subjective experience carries massive ethical weight regarding welfare, potential digital servitude, and the nature of personal identity.
“Everyone else, your friends, your family, me, you only ever see from the outside.”
FULL TRANSCRIPT
Tonight, a machine will have millions of
conversations and forget everyone.
[music]
The question is whether anyone was there
to have them.
>> Do you think that the consciousness has
perhaps already arrived inside AI?
>> Yes, I do.
>> That's Jeffrey Hinton. He won a Nobel
Prize for the ideas every AI is built
on. And the company that makes Claude?
>> [music]
>> We don't know if the models are
conscious. We're not even sure that we
know what it would mean for for a model
to be conscious or whether a model can
be conscious, [music] but, you know,
we're we're open to the idea that it
could be.
>> So, how would we ever know? [music] How
do you look inside a mind? Just days
ago, a neuroscientist at Berkeley,
[music] Doris Tsao, posted this. We
discovered something totally unexpected
about the neural structure of
consciousness. [music] Here's what she
meant. When a brain is shown two
different images, one to each eye,
[music]
it doesn't simply pick one. As she puts
it, it flips between a conscious mode
and an unconscious mode about 15 times
[music] every second. Within days, posts
about it were viewed more than 4 million
times. This paper is days old and it
might be [music] a first glimpse of the
line between what a brain processes and
what it actually experiences. The kind
of line we'd need to [music] find inside
a machine.
Here's what they actually did. Two
monkeys [music] and 14 humans to check
the illusion.
Each eye gets a different face. Your
brain [music] can't merge two faces, so
it picks one and the other one simply
vanishes from experience. It's still
there on the screen. You just can't see
it.
For decades, scientists assumed the
brain mostly threw that invisible face
away. So this team, led by Doris Tsao at
Berkeley, did something new. [music]
With ultra-thin probes, they listened to
thousands of neurons at once [music] in
the parts of the brain that recognize
faces.
And the neurons weren't ignoring the
hidden face at all. For about 60
milliseconds, they'd represent the face
the animal was seeing. Then, for the
next 60, the face it wasn't. Back
[music] and forth, about 15 times every
second.
Then, they did something incredible.
They rebuilt [music] pictures from each
slice of brain activity. One set looked
like the face the animal was aware of.
The other looked in rich detail like the
face it never consciously [music] saw.
So the brain isn't ignoring the
invisible. It's processing it in full in
its own time slots.
As the lead author put it, consciousness
may depend not on where in the brain
activity [music] happens, but when.
Does that flicker change in our most
intense moments, when life feels most
vivid?
Nobody [music] knows yet, but for the
first time, it's something we can
measure.
>> We think that during binocular rivalry,
the entire population is coherently
toggling between two states, conscious
and suppressed.
>> And for two of the checks on that
result, the team [music] used Claude to
speed up the analysis. The AI didn't
make the discovery, but it was holding
the flashlight.
And that's happening everywhere now. AI
has turned brain scans [music] into
sentences describing what a person was
watching and what they were remembering.
We're getting closer to reading the
brain from the [music] outside.
But with AI, we can do something we
could never do with a living brain. We
can open it up and look.
At Anthropic, the company that makes
Claude, [music] that's exactly what they
did.
First, they found the pattern for one
idea, shouting, and injected it straight
into Claude's [music] mind. Then they
asked, do you notice anything? About one
time in five, it said yes, something
about loudness, before the word had
shown up anywhere. The first sign an AI
can sometimes notice its own thoughts,
force it to say bread out of nowhere,
and it apologizes. Plant bread in its
earlier thoughts first, [music] and it
decides it meant to say it and invents a
reason. Then this summer, they built a
new kind of lens and found something
they named the J-space. [music]
The leading scientific theory of
consciousness says your awareness is a
small spotlight. [music]
A handful of ideas at a time broadcast
to the rest of the brain while
everything else runs in the [music]
dark.
Inside Claude, they found something
shaped a lot like that. A tiny [music]
space, usually no more than about 25
ideas at once that the model can
actually report and reason [music] with,
floating on top of a vast ocean of
automatic processing.
Ask Claude how many legs the animal that
spins webs has. Inside that space,
spider lights up and it says eight.
[music] Reach in and swap spider for ant
and it says six.
Neither word was ever spoken. [music]
And when they switched the whole space
off, Claude's descriptions of its own
experience became flatter, [music] more
mechanical.
None of this proves Claude is conscious,
and the researchers say so themselves.
But a spotlight like ours grew [music]
inside it on its own. Nobody designed
it. In Anthropic's own words, it wasn't
designed or programmed [music] by them.
It emerged on its own during training.
And that's the strangest [music] truth
about AI.
The people who make it don't write it
line by line.
One of Anthropic's co-founders [music]
puts it like this.
>> Um because at the end of the day, we end
up with this artifact that could do all
of these amazing things. It can, you
know, write essays and translate and,
you know, understand images. It can do
all these things that we have no idea
how to directly create a computer
program to do. And it can do that
because we we grew [music] it. We didn't
we didn't write it. We didn't create it.
>> Grown. And things nobody planned keep
[music] turning up inside.
Ask Claude for a rhyming poem, and
before it writes the line, it has
already picked the word it's going
[music] to rhyme with.
The researchers set out to prove it
didn't plan ahead. It did.
Ask it to add two numbers, [music] and
it uses a strange method of its own.
Then ask how it it, and it describes
[music] the method we learn at school.
And underneath every language it speaks,
it seems to think in one shared space of
ideas. [music]
Nobody taught it any of this.
And who knows what else could emerge
from that?
And the line is blurring from the other
[music] side, too. In Australia, a
company grew living brain cells on a
chip and taught them [music] to play
Pong. Within 5 minutes, the cells were
getting better at it. This year, human
brain [music] cells on their chips
played Doom. You can already buy a
computer made of living human neurons.
[music]
Why build a computer out of brain cells?
Power. Your brain runs on about 20 W,
less than a light bulb. One of the
world's biggest supercomputers needs
around 21 million W, and the head
scientist at that company says [music]
living neurons learn on a tiny fraction
GET MORE FROM YOUR TRANSCRIPTS
Sign up for free and unlock interactive viewer, AI summaries, translations, concept maps, and more. No credit card required.