📱

Get Our Mobile App

Take your business learning on the go!

Download on the App StoreGet it on Google Play

“Using Living Human Neurons Is the Best Way to Build AI” | FinalSpark (Swiss Startup)

This Is The World16:23

Transcription

We realized that the best way [music] to

do a better AI was to use living

neurons. Biological processor are made

of living cells. This has never been

done in in history. This is basically

what is running CHP [music] for

instance. It was taking several kilowatt

of power. Your brain is 20 W for 100

billion of neurons. We used more

sophisticated [music] versions of these

artificial neurons. What's about the

ethical issue?

The picturesque town of Ve, Switzerland,

blends natural landscapes with

>> [music]

>> urban fabric. It's home to the

headquarters of one of the most

visionary biotech companies [music]

working to create a so-called living

computer. Is it possible to combine the

computational power of computers [music]

with the efficiency of the human brain?

Final Spark is a Swiss bio computing

startup that is the first in the world

to build a living processor, creating a

bioprocessor capable of [music] learning

while consuming over a million times

less energy than conventional silicon

systems.

>> What is it human brain organo?

It's basically a small part of nervous

tissue made of neuron coming from

human beings. [music] What is it your

human brain project?

So [music]

first when you talk about human brain

project this is first a uran project.

[music] Okay. Um and uh that somehow

inspired [music] us. Okay. But it's not

really a project related to human brain

actually. Um there are many people uh

doing human brain organoids in the world

and in Europe also. Uh [music] this is

not as such this is not something um

even I would say very advanced. Okay. uh

what is a bit special in our company

[music] is that what we are trained to

do with this brain organoids is to use

them for computation.

So it's the goal that we have and the

work that we do which is original. You

see what is behind the corner.

Well, I don't know if I see what's

behind the corner, but that's right.

That's a it's a very original thinking

[music] uh to think that human neurons

can be used for computation. [music]

How do you collect human cells?

Okay. uh we buy human cells from

companies that are specialized to

produce them and they [music] are pur

potent stem cells. So they usually they

come from fiberblast so skin cells that

are reprogram it to [music] become purip

potent. So this means that they can

become whatever type of cells. [music]

How is it possible to create the neuron

cells from using you know skin cells?

>> Yeah, exactly. Is because we

there are methodologies that we can use

to uh the differentiate the skin cells.

[music] So bring them back to the pur

potent state and then from this point we

can differentiate them in neurons. to

give them um specific molecules in the

culture medium

from uh puripotent [music] stem cells

and then we

make sure that they are pur potent. So

we check some markers to make sure that

they will be able to differentiate in

neurons and astroite and other type of

cells in the brain. And then we yeah we

culture them with special uh molecules

that will induce the [music] cells will

tell the cells to become a neuron. So we

are in the cell culture room and is the

place where we grow the brain organoids

and they are in the incubator where we

found like um

precise temperature [music] and CO2

concentration. Inside of these plates we

have different type of organoids that we

are growing. First of all, actually we

um grow them in a special plate that

[music] allow them to create a 3D

structure and then uh step by step we

give different type of uh reagents or

molecules as we we can say that will

tell them to [music] to become

specific type of cells and also they

will reorganize in structures [music]

that are similar to part of the brains.

The function will be similar to um fatal

[music] brain.

Some are in the orbital shaker because

the shaking uh helps the nutrients and

the oxygens to enter inside the

organoid.

>> How do you keep your neurons [music]

cells alive?

>> So, we give them a special um solution.

So [music] it's a it's called cell

culture medium that contains nutrients

and uh that help [music] them to to

live.

[music]

>> How do you keep this on organoid alive

in that system? I know as you can see

there is uh maybe not in really in the

picture but uh we have a fluidic system

below. So uh so we have pipes and bring

the the medium actually. So we have pump

[music] and um so the medium is really

flowing below the the organoid. So

[snorts] the organ exactly will be on

this on on the top of uh

>> so all the time they are feding.

>> Yeah. All the time. Yes. So with this

system we can keep them alive usually

one two three it can be months sometimes

it's but mostly it's like two weeks you

can see the the pump the pump speed

actually of all the MA so for example

you can see that we are running at

around I don't know 30 or 50 microL per

minute [music]

>> how long time you can keep these cells

alive 1 year 2 years 3 years

>> we managed to keep them al almost 3

years alive.

>> How we can connect that organoid with

your neuron platform?

>> We [music] put the organoid on top of

the electrons and we record their

activities [music]

what we can see. So this actually

through you when you go to the website

you can see the live actually so these

pictures. So we have four different MEA.

So actually it's a closeup on this part.

>> Okay. MEA is

>> multi-elerode array.

>> So because there is a lot of different

electrode that you can see maybe better

on this picture you see this littleer.

[music]

>> What you can see it's uh we have five

MEA. [music] So the black one is a new

one that you are trying now. [music] But

what you can see on the website is the

that one the clear one. You see this one

on the bottom the two on the bottom. So

difference. So that one it mean you put

one or two organoid here. So you can

record from eight electro they will be

independent. We call this it's a

membrane actually with with a hole in

the middle. Uh we call it confetti. Uh

but um why it's just a membrane actually

it helps to to for [music] me to put in

the middle the um the organoid and also

it helps to uh to reduce the noise.

>> Why [music] did you start this project?

Why biomputing? Why so futuristic vision

of the reality?

[music]

The first thing you know I'm an engineer

so I always try to solve problems. Okay.

and take the best solution for a given

problem. Okay. And initially when we

started to work on this, you know, I

have a PhD in signal processing,

basically mathematics, nothing to do

with neurobiology. Okay. So I can't tell

you if I could have avoided [music]

to learn an entire new field, I would

have preferred. But when we started the

company, we worked a lot with artificial

neural networks.

Artificial neural networks are

simulations of neurons. This is

basically what is running CHP for

instance, okay, an old generative AI.

[music]

And we used more sophisticated versions

of these artificial neurons. And they

were consuming so much energy that we

realized it [music] would be impossible

to scale this

like like with 100 simulated [music]

neurons, it would take it was taking

several kilowatts of power. Imagine

[music] your brain is 20 watts for 100

billion of neurons.

86 billion neurons we have in a human

brain. Yeah. Uh you think 20 [music]

watts for calculation. That's the reason

that you uh decided to open that

company.

No, the main reason of opening this

company was first to do fundamental

research in AI and in the course of this

fundamental research, [music] we

realized that the best way to do a

better AI was to use [music] living

neurons.

>> What's the difference between biological

and classical processor? between

bological and classical processor.

Well, biological processor are made of

living cells which is totally

disruptive. Indeed, [music] this has

never been done in the history.

[music]

I know that I I saw your butterfly

experiment. What is it?

>> Yeah, the butterfly experiment. um is a

way

to show in the browser internet browser

of a user uh how he can interact

directly from his internet browser in 3D

with [music] a brain of sitting in

Switzerland in our lab. This is a visual

illustration of this [music] and you see

this butterfly

flying in three dimension and you can

actually control its [music] direction.

The idea is when the

the butterfly see the light. So the

these [music] blue these big blue dots

you send stimulation and if the

butterfly reply to the stimulation you

go straight. So it mean you go to the

light and if not so you are randomly

going everywhere and when he he reply so

[music] you can see actually here that

after each stimulation this big line you

can see that there is some reply

probably a lot of neuron are discussing

u when you saw this big wave actually we

call burst because there are probably

more than one neuron that are sending

information so it's essentially here we

see a lot of communication

>> you have a communication with organoids,

right?

>> Yeah. Yes, exactly. We have but not

between each of them. So like for

example here you see again a burst of a

lot of neurons [music]

>> they are processing the information. So

so how do you enforce such neurons to

process such information?

>> Yeah, dopamine.

>> So in the [music] way you the words that

you use here makes a lot of sense to me.

Okay. Uh if you talk about processing

information and enforcing a behavior

okay uh one so this is called learning

okay enforcing a behavior okay and this

is the main challenge that everybody

every research in the researcher in the

world of bio computing has to challenge

[music] today okay um so your question

one way that we are indeed pursuing here

is to use for [music] instance dopamine

to reward the brain if it does what we

expect

but there are other strategies which are

used by [music] the by others. Do you

have any progress?

>> Very slow progress. Okay. Because we

need to confirm experiment. [music] Even

if something works once, it does not

mean it works at all. Okay? We are

engineers more [music] than scientists

in the sense that we want something that

works.

What's about the ethical issue?

>> So,

you know, first we are scientist. Okay.

Uh any questions about science? Uh I may

be able to answer

probably

actually not most of the important

questions in science we are not able to

answer but some of the technical

question I may be answer. [music] Uh now

if you think about ethical questions

this is really out of the field of

expertise of final spark. Um and for

this uh we recognized this is a concern

and what we did is that we made a number

of contact with universities and eition

because they are experts in this field

of ethics people [music] there this is

their job actually to study and what we

do is that we went also to conference in

November last year for instance [music]

to present in an ethic conference to

additions what we were doing so So that

those people who are experts in this

[music] can start to think what would be

the implication of using living neurons

for computation. Why human cells?

[music] Yeah, that's a very good

question actually. If you think about

computation probably rat cells could

could also work very well. Okay. The

thing is when you create a startup you

never know what are going to be um the

outcome. Sometimes you may discover

something along the way that is very

useful

and one of the things that we could

discover along the way could have

therapeutic applications

and I'd better have therapeutic

applications [music] for human beings

than for rats.

moment that shocked you that you uh that

surprised you positively [music]

here in this frame of sport?

>> Well, when we were able to record some

activity from neurons, it was already a

fantastic uh thing. When we were able to

to cells, you know, from liquid nitrogen

and multiply them ourselves, we were

able to see some reaction

>> of the nervous tissue to the pamine and

the waves.

>> Yeah. Yeah. We see we saw some activity

which is typical induced by the release

of the pine. This was [music]

one

one I would say major achievement for

us. Okay. But just always put in

perspective that we were starting from

ground zero.

>> Do you really believe that biomputing is

the future of data processing is a

future of computing?

>> Yes absolutely [music]

specifically for one type of computing.

Okay, computing of AI. Okay, like

quantum computing may be restricted some

some types of computing, bio computing

will be restricted also to some task of

computing. But AI is absolutely perfect

for an obvious reason. AI is based on

simulations of neurons.

[music]

>> [music]