00:00:00
It’s no secret that tech companies are racing
to build “artificial general intelligence,”
00:00:04
or AGI that can match a human brain without
needing a lifeline. But here’s the kicker:
00:00:09
our brains already have the home
field advantage. They do the same
00:00:13
heavy lifting with just a fraction of
the resources. Whether it’s energy,
00:00:17
water, land, components, or, you know… money…
human brains are just way better and cheaper.
00:00:24
Scientists figured this out long before we
were arguing with ChatGPT about sandwich
00:00:28
recipes. And now, with the AI race heating
up faster than a server farm in August,
00:00:32
biotechnologists are asking: Why build AI like
a brain when you could just use the real thing?
00:00:38
Right now, you can either buy a
human brain cell-based computer...
00:00:42
or rent time on a remote one. Yep, even
brainpower’s got a subscription plan these
00:00:47
days. So what can these computers
actually do? How do they work? And,
00:00:51
most importantly, should we
be freaking out a little bit?
00:00:54
I’m Matt Ferrell … welcome to Undecided.
This video is brought to you by Brilliant.
00:01:01
Got a spare $35,000 and a lab coat lying
around? You could be the proud owner of the CL1,
00:01:07
which is a biocomputer from Australia’s
Cortical Labs. Unlike your regular computer
00:01:13
running a BIOS (all caps), the CL1 runs
on a biOS — “Biological Intelligence
00:01:20
Operating System.” Because instead of just
mimicking a brain... it literally uses one.
00:01:25
It’s packed with living human neurons, which
are cells that react, learn, and adapt inside
00:01:30
a simulated world. It's like a real brain,
minus the existential dread. (Probably.)
00:01:36
And Cortical Labs isn’t the only one
making living computers. FinalSpark
00:01:39
over in Switzerland is also training human
neurons in the form of brain organoids.
00:01:44
These are tiny clusters of living brain cells
you can rent for research. I even talked to a
00:01:48
few researchers using these systems, and trust
me, it’s as weird and fascinating as it sounds.
00:01:55
But before we get too deep into
the how, let’s ask the obvious:
00:02:00
Why? Why go through all the trouble
of growing brain cells when we have
00:02:05
perfectly good silicon computers. According to
Cortical Labs’ CEO Hon Weng Chong, it’s simple:
00:02:10
“Everyone is racing to build AGI,
but the only true AGI we know of
00:02:14
is biological intelligence, human intelligence.”
00:02:18
But there’s a bigger, messier reason too:
sustainability. Generative AI eats everything
00:02:23
— more energy, more water, more land, more chips.
It’s a runaway resource hog, and it’s only getting
00:02:29
worse. It’s something I dug into recently in
another video if you want the full breakdown.
00:02:33
Tech companies aren’t slowing
down, they’re scaling up. However,
00:02:37
you can only expand so far before
you start stepping on toes. Data
00:02:40
centers aren’t just hogging electricity.
They’re draining water, eating up land,
00:02:44
and putting stress on local communities. In
early 2024, OpenAI CEO Sam Altman put it bluntly:
00:02:51
“We do need way more energy in the world than
I think we thought we needed before…And I
00:02:55
think we still don't appreciate the energy
needs of this technology. The good news,
00:02:59
to the degree there's good news, is there's
no way to get there without a breakthrough.”
00:03:04
FinalSpark’s Fred Jordan thinks he’s found it.
Living neurons. If we can train biocomputers
00:03:09
like traditional AI, Jordan says we could slash
AI’s energy use by thousands of times — making
00:03:14
its carbon footprint almost invisible. And to
really see why that matters, let’s talk numbers.
00:03:20
Meet the Frontier: America’s first exascale
supercomputer. It cranks out 1.1 exaFLOPS — that’s
00:03:26
a quintillion calculations per second —
while weighing more than 266 metric tons,
00:03:31
stretching across 74 cabinets, and costing $600
million to build. Now compare that to your brain:
00:03:38
1.3 kilograms (3 pounds).
20 watts of power usage.
00:03:42
1 exaFLOP of raw performance.
And a price tag that just says: "not applicable."
00:03:47
Seriously — David Byrne was
right: we’re makin’ flippy floppy.
00:03:51
And Cortical Labs gets it. Their motto?
00:03:54
“What digital AI models spend tremendous
resources trying to emulate, we begin with.”
00:03:59
Forget AGI. They're chasing SBI, or
Synthetic Biological Intelligence.
00:04:04
But how are they doing this? Well, before we dive
into how these brain-powered computers actually
00:04:08
work, it’s a good idea to know how to program
them. And there’s something that can really help
00:04:13
you do just that with today’s sponsor, Brilliant.
Brilliant helps you get smarter every day,
00:04:17
with thousands of interactive lessons in
math, science, programming, data analysis,
00:04:22
and AI. The best part is how Brilliant teaches
through hands-on problem solving that lets you
00:04:27
play with the concepts. Learning by doing is
proven to be 6x more effective than watching
00:04:32
a lecture. You build up your critical thinking
skills, which helps make you a better thinker.
00:04:36
It’s been crafted by an award-winning team of
teachers, researchers, and professionals from
00:04:41
Stanford, MIT, Caltech, Microsoft, Google,
and more. Build up your problem solving
00:04:47
skills with courses in a wide range of topics,
like programming, which is constantly evolving
00:04:52
(just like we’re talking about today). Learn
to think like a programmer by breaking down
00:04:56
complex problems into manageable chunks of code.
Developer your intuition for computer logic,
00:05:00
so you can design and debug real programs. And get
familiar with Python to start building programs on
00:05:06
your first day. To try everything Brilliant
has to offer for free for a full 30 days,
00:05:10
visit brilliant.org/Undecided or scan the QR
code onscreen — or you can click on the link
00:05:16
in the description. You’ll also get 20% off
an annual premium subscription. I absolutely
00:05:20
love Brilliant and think you will too. Thanks
to Brilliant and to all of you for supporting
00:05:24
the channel. Alright, so how do you turn
skin cells into neurons inside a computer?
00:05:32
Biocomputing sounds promising for efficiency,
but what actually is a biocomputer? In 2024,
00:05:37
Hon Weng Chong called the CL1 a “body in
a box.” And no, it’s not a horror movie
00:05:43
prop floating in a glass jar. The neurons
inside are actually stem cells — or rather,
00:05:48
they were stem cells, before
scientists reprogrammed them.
00:05:52
Here’s the basic rundown. The human
body has a few types of stem cells.
00:05:55
Embryonic stem cells exist during early
development, while adults have stem cells
00:06:00
in places like the skin and bone marrow,
producing new cells on demand. Once these
00:06:05
stem cells specialize into a particular
role (say, skin cells) they normally stay
00:06:09
that way. That ability to specialize
into anything is called pluripotency.
00:06:14
For a long time, researchers thought
mature human cells couldn’t revert back
00:06:17
to their original state. Then in 2012,
Kyoto University’s Shinya Yamanaka won
00:06:22
a Nobel Prize for proving otherwise.
By flipping a specific set of genes,
00:06:26
he turned ordinary mouse skin cells back
into pluripotent stem cells, ready to become
00:06:31
anything again — a complete biological
reset. And yes, it works for humans too.
00:06:37
That’s exactly how Cortical Labs and FinalSpark
00:06:39
grow their neurons. The process
kicks off with dedifferentiation:
00:06:43
Researchers sample blood or skin
cells from adult volunteers.
00:06:46
They reprogram them into induced
pluripotent stem cells — iPSCs.
00:06:51
Then, through weeks of careful
incubation and gene tweaking (sadly,
00:06:55
no epic training montage),
the cells slowly transform.
00:06:59
Finally, researchers differentiate the
iPSCs again, this time steering them
00:07:03
toward becoming neural progenitor cells.
Basically, baby neurons-in-training that,
00:07:08
if they pass the tests, move on to the next stage.
00:07:11
As Cortical Labs explains it, once the cells are
about ready to turn into neurons, the researchers
00:07:16
move them onto a multi-electrode array (MEA) chip.
The cells attach to the chip, allowing electrical
00:07:22
signals to be both sent to them and received from
them. For a few months, they continue to develop,
00:07:27
and then bam, human neurons, merged with silicon.
It doesn’t get much more sci-fi than that.
00:07:33
The whole point of this process? To build brain
organoids, which are miniature 3D tissue cultures
00:07:38
modeled after real organs. Researchers use
organoids to study how organs like kidneys,
00:07:43
lungs, and even brains develop and
function. Think of them like those
00:07:48
plastic models you saw in biology class…
except these ones are alive. They eat,
00:07:53
they grow, they create waste
… and eventually, they die.
00:07:57
Now, to be clear: these aren’t fully formed
human brains. Not even close. There’s a reason
00:08:01
they’re called organoids and not organs.
They’re organ-like. Organoids are tiny,
00:08:06
limited to specific brain regions, and
cap out around 5 million cells (about
00:08:11
the size of half a centimeter).
By comparison, your brain has
00:08:14
around 86 billion neurons, plus another 85
billion non-neuronal cells. So yeah. Tiny.
00:08:21
When it comes to synthetic neurons,
there’s a lot more nuance than I can
00:08:24
fit here. Cortical Labs actually breaks
it down step-by-step on their YouTube
00:08:28
channel if you want to nerd out. The short
version? It takes months of careful work.
00:08:33
FinalSpark’s timeline clocks in at about four
months to create a single brain organoid.
00:08:38
And just as the CL1 hit the market, MIT
researchers announced something wild. They figured
00:08:42
out how to skip the stem cell stage altogether by
generating neurons directly from skin cells. If it
00:08:48
scales, this shortcut could make neuron production
way faster and cheaper for biocomputing.
00:08:53
In the end, the goal is to tap brain organoids
as an alternative to AI, or what some are calling
00:08:59
Organic Intelligence, or OI. Now, you might
be thinking: What about the ethics of growing
00:09:04
mini-brains in the lab? Good question. And
it’s something researchers across biotech,
00:09:08
neuroscience, and philosophy are struggling
to answer, too. But before we wander too far
00:09:13
down that rabbit hole… Let’s first talk about
what you can actually do with these things.
00:09:19
Outside of building SBI, OI, biocomputing, wetware
(whatever you want to call it), brain organoids
00:09:26
are already making waves in medical research.
Scientists are using them to model diseases,
00:09:31
test new drugs, explore gene therapies,
and push the boundaries of personalized
00:09:35
medicine. And someday, advances here could
even help reduce the need for animal testing.
00:09:40
As for biocomputers like the CL1? Cortical Labs
and FinalSpark are betting big on them as a
00:09:45
greener alternative to today’s resource-hungry
AI. (Or at least, that's their pitch.)
00:09:50
So, what can you actually do with a biocomputer
right now? Well, in the world of computing,
00:09:54
milestones usually come dressed up as games.
Alan Turing’s “imitation game” inspired the
00:09:59
Turing Test. IBM’s Deep Blue made headlines
by beating chess champion Garry Kasparov in
00:10:04
1997. Then DeepMind’s AlphaGo took
down Go master Lee Se-dol in 2016.
00:10:10
And now? Meet DishBrain, Cortical Labs' tiny,
living player. It doesn’t stand tall like Deep
00:10:16
Blue — which its creator famously compared
to "an office refrigerator" back in 1995.
00:10:20
It doesn’t need 1,202 CPUs like AlphaGo
did. And it definitely doesn’t hog
00:10:25
7,300 square feet like Frontier, the
world’s first exascale supercomputer.
00:10:30
DishBrain fits inside… a Petri dish.
00:10:34
By 2022, it could play Pong.
00:10:37
Here’s the Breakout breakdown.
During DishBrain’s development,
00:10:40
Cortical Labs studied human and mouse
neurons grown on MEA chips. The idea?
00:10:45
Based on the Free Energy Principle, which says
intelligent systems prefer predictability,
00:10:49
they trained the neurons to learn how
to play Pong using electrical feedback.
00:10:54
The setup actually mirrors how we interpret
the world: We get sensory input → our brains
00:10:59
translate it into electrical signals → we
respond. DishBrain’s neurons did the same
00:11:03
thing inside a simulated game world. When the
neurons missed the ball, they got hit with random,
00:11:08
unpredictable signals: 4 seconds of a 150 mV
voltage at 5 Hz. Not exactly emotional punishment,
00:11:15
but more like getting a foul called and
hearing static instead of a whistle.
00:11:19
When they successfully intercepted the ball?
They got a reward: a clean, smooth sine wave
00:11:24
at 100 Hz for 100 milliseconds. And
the more sensory input they had,
00:11:28
the better they performed. When there was
no feedback at all? Performance flatlined.
00:11:33
Cortical Labs argues this shows
DishBrain wasn't just reacting,
00:11:37
it was learning. Not well enough to beat
you at Pong yet... but still, pretty wild.
00:11:42
Like a lot of biocomputing, the
theoretical neuroscience behind
00:11:44
Cortical Labs’ work is a rabbit
hole way too deep for one video.
00:11:48
But if you want a full biotech
deep dive someday, let me know!
00:11:52
For now, let’s put down the journal articles
and talk about something more hands-on. Despite
00:11:56
the CL1’s $35,000 price tag (and the fact
you need to be a legit researcher to buy
00:12:01
one) interacting with biocomputers isn’t as
locked down as you might think. For starters,
00:12:05
the CL1’s API documentation is publicly available
on GitHub. Plus, both FinalSpark and Cortical
00:12:11
Labs run public Discord servers where you can ask
questions, swap notes, and geek out with others.
00:12:16
And if you don’t have a lab budget? Both
companies are offering subscription services
00:12:20
to rent remote access to their living neural
networks. There are some caveats, though:
00:12:24
FinalSpark’s platform is
live but subject to approval.
00:12:28
Cortical Labs’ rental service is still gearing up.
00:12:30
Right now, FinalSpark lists 10 universities
as official users of its “neuroplatform.”
00:12:35
I had the chance to talk to two
research teams using FinalSpark’s
00:12:38
system — which lets them work
with brain organoids remotely.
00:12:42
First up: Dr. Kyle Wedgwood and research
intern Wiktor Wiejak at the University of
00:12:46
Exeter in England. Wedgwood’s a mathematician
specializing in neuroscience, and he’s using
00:12:50
FinalSpark’s organoids to explore what he
calls “the fundamentals” of how neurons work:
00:12:56
“Here it's really trying to ask the question
about what can we import from sort of mathematical
00:13:01
descriptions of things like neuronal networks and
use them to understand how neuronal networks work,
00:13:08
how cells communicate with each other,
but also how can we exert some sort of
00:13:12
control or some modulation in a sort of
targeted way on your neuronal networks.”
00:13:16
When I asked what "training" neurons looks like,
00:13:18
Wedgwood pointed right back
to Cortical Labs’ Pong study:
00:13:21
“Over time, if you stimulate these
networks, they respond in a way by
00:13:26
effectively strengthening and weakening different
kind of connections between neurons. So broadly,
00:13:31
this is called synaptic plasticity, and it's
one of the fundamental ways that brains learn,
00:13:37
remember, acquire new skills, all this kind
of stuff. …Obviously the neural network did
00:13:42
not become really, really good at playing Pong
right? It just got better than it was in the
00:13:46
beginning. But it did show that actually you
can study kind of learning in these systems.”
00:13:50
Then, I sat down with Dr. Tjeerd Olde
Scheper from the Artificial Intelligence,
00:13:53
Data Analysis and Systems (AIDAS) Institute at
Oxford Brookes University. As a computational
00:13:59
neuroscientist, Scheper’s digging
into how biological systems store and
00:14:02
represent information — and whether they might
someday outperform our traditional computers:
00:14:07
“Each cell, individual cell, solves a huge number
of computational tasks every moment in time…We
00:14:14
have a lot of knowledge about the biochemistry,
the molecules involved, the structure of a lot
00:14:20
of those things as well. That's getting more
apparent as well, but how they actually interact
00:14:25
with each other, how they come combine with
each other to create this complex, system and
00:14:30
do that to solve complex problems is, is still
quite of, you know, quite fairly much unclear.“
00:14:36
“So from my point of view is, if we have
a better understanding how each of those
00:14:40
components work by letting each part of
this component decide for itself what its
00:14:47
best behavior should be —and, I mean behave in
a dynamic sense, so how it changes over time.”
00:14:53
So, there’s a glimpse of how researchers
are already putting biocomputers to work.
00:14:56
But here’s the fun part: You don’t need
a PhD, or a grant, to get involved. If
00:15:00
FinalSpark likes your project enough, you
could potentially work with them for free.
00:15:04
They also host a 24/7 livestream of
some of their neurons online. Yes...
00:15:10
you can literally watch a Petri dish.
Could be better than some screensavers.
00:15:15
Looking for something a little flashier?
FinalSpark also offers “the butterfly,”
00:15:19
a 3D simulation showing about 10,000 neurons
reacting to sensory input in real time. Basically,
00:15:25
it's like controlling a virtual
RC car (except your remote is a
00:15:28
clump of neurons). And don’t worry: heavy
emphasis on virtual. It’s just visuals.
00:15:34
Of course, FinalSpark co-founder Fred
Jordan is quick to remind everyone:
00:15:38
This tech is very early days. It’s not going to
replace your phone or laptop anytime soon … and
00:15:44
it might never. As Jordan’s joked, running Windows
on a brain organoid would be ... unrealistic. No
00:15:50
word yet on whether it could run DOOM, though.
Still, Jordan’s point is worth remembering: The
00:15:54
inventors of the semiconductor had no idea what
the world would eventually build on top of it.
00:16:02
That’s exactly the scary part, isn’t it? All
the what ifs. I’m guessing you don’t need me to
00:16:06
invent hypotheticals … you’ve probably already
thought of a few yourself. The biggest one for
00:16:10
me? How would we even know if a brain organoid
achieved consciousness? And if it did… what then?
00:16:16
Well, I’ve got good news and bad news.
Bad news: We have no idea what we’re doing.
00:16:22
Good news: We have no idea what we’re doing.
00:16:25
We don’t even have a solid definition for
human consciousness yet. Cortical Labs
00:16:29
is actually running a public survey to tackle
this exact problem. Without a shared language,
00:16:33
it’s tough to even describe biotech research
properly. Especially when terms like "sentience,"
00:16:38
"consciousness," and "thinking" trigger emotional
reactions. Who gets to decide what counts as
00:16:43
thinking? What’s the threshold for sentience?
Here’s your chance to toss in your two cents.
00:16:48
That said, and I really can’t emphasize
this enough, we're nowhere close to needing
00:16:52
any emergency shutdown buttons. Brain
organoids, even at their best today,
00:16:56
are tiny compared to real brains. Their networks
aren't even on the same playing field as a bird,
00:17:03
a mouse, a snake... or even an insect.
No disrespect to the humble fruit fly.
00:17:08
As Dr. Scheper put it:
“We have made progress in
00:17:11
trying to understand what we can and cannot
do…at the moment, it's not something that
00:17:17
we can make them push to do anything that is
even remotely related to what intact brains or
00:17:23
even slices or parts of brain can do, because
those have been trained and accommodating.”
00:17:30
So, where does biocomputing land on NASA’s
Technology Readiness Scale? Short answer:
00:17:34
very much still in the lab. Biocomputers
like the CL1 and FinalSpark’s neuroplatform
00:17:39
are under active development. Researchers
using them are basically beta testers and
00:17:43
giving feedback, suggesting improvements,
and helping the developers build in real
00:17:48
time. It’s a true collaboration
between scientists and startups.
00:17:51
Zooming out even further, biocomputing
as a whole is still barely scratching
00:17:55
the surface. It’s an emerging science, a little
like quantum computing... or the quietly growing
00:18:00
comeback of analog computing. Right now,
we’re standing at the top of the canyon,
00:18:04
admiring the view. But what’s waiting
down at the bottom? No one knows yet.
00:18:08
And hopefully not anything with
too many mouths that can’t scream.
00:18:11
Biocomputing, and biotech more broadly, is way
too big to cover fully in one sitting. If this
00:18:16
video fired up your neurons, I’ll be publishing
some of the full interviews over on Still TBD,
00:18:21
which is my follow up podcast. We go much
deeper into the strengths, challenges, and wild
00:18:25
possibilities of brain organoids. And honestly?
There’s still so much more out there to explore.
00:18:31
But what do you think? Does the concept of
SBI bring you hope? Or are we better off
00:18:35
sticking to silicon…before organoids totally
FLOP? Jump into the comments and let me know,
00:18:39
and be sure to listen to my follow up podcast
Still TBD where we’ll keep this conversation
00:18:43
going. Thanks as always to my patrons for
your continued support and helping to keep
00:18:46
the channel going. Keep your mind open, stay
curious, and I’ll see you in the next one.
