Here's Why the Nobel Prize In Physics Went to AI Researchers

Scientists won the Nobel Prize
12 followers
Okay, this time, let's discuss a bit of a controversy,
or I guess maybe why it's not a controversy, in regards to the 2024 Nobel Prize in Physics.
Something many of you might have already heard about,
and something that was just announced a few days ago from when I'm making this video.
But if you missed it, well, that's basically what this video is about.
Hello and full person, this is Anton.
Let's discuss the Nobel Prize in Physics awarded in 2024,
why some people are not particularly happy about it,
but why a lot of other people think it's actually appropriate.
And so let's start with the award.
You can find all of the links for this in the description,
but in 2024, the award went to John Hopfield of Princeton University,
and the British-born Geoffrey Hinton, who is now a professor at University of Toronto.
But the reason this was controversial, and not a lot of people were happy with it,
is because Hopfield is technically a theoretical biologist with a background in physics,
and Hinton is known as the computer scientist, cognitive scientist,
and actually more officially known as the godfather of the artificial intelligence,
and up until recently was also employed by Google as one of the artificial intelligence experts.
And so technically, I guess, they're not physicists.
And so quite a lot of people were somewhat unhappy,
and were not entirely sure why they actually got the award.
But the committee decided to give it to them for their contribution to computer science,
and for specifically the development of artificial intelligence.
Essentially, a lot of the ideas proposed by both scientists
are one of the reasons we have things like Chad GPT.
And so the claim here is that their studies were the foundation
for modern applications of artificial intelligence.
And so this was actually a bit of a shock to a lot of people, a lot of scientists.
With the first initial question being,
okay, why does everything have to be about AI nowadays?
With the second question being, why physics?
Or I guess, what does it have to do with physics, when technically this is computer science?
And so I guess let's tackle some of these questions,
and discuss why their studies are so crucial,
and why technically it is actually pure physics,
just not really the physics we think of when we imagine people like Einstein,
or people like Richard Feynman.
And here to understand this, we do have to take a look at history just a little bit,
and specifically the foundation for machine learning.
So prior to Chad GPT, and prior to artificial intelligence,
everything started with these early machine learning networks,
and a lot of them were known by a different name.
Many of them were actually going by the name Boltzmann machine.
In essence, a kind of a network, and specifically a neuron-like network,
that was able to receive inputs and outputs, and was able to learn.
And generally, they can be seen as something like this.
An interconnected network, with each of the nodes possessing different strengths.
And both Hopfield and Hinton played a very big role in developing these Boltzmann machines,
which eventually produced machine learning.
And pretty much all of these early ideas came entirely from physics.
Specifically, John Hopfield, who eventually produced his own version of this,
today commonly referred to as the Hopfield network,
prior to the introduction of these ideas,
based all of his studies on statistical physics,
or sometimes referred to as statistical mechanics.
And to be more exact, if you want to read more about this,
the model known as the Ising model,
named after Ernest Ising, and published back in 1920s,
as a model explaining magnetism and thermal equilibrium.
Here's a very basic schematic,
representing this model for a typical magnetic material.
And so essentially, a lot of these early models were entirely based on something different.
Here the motivation was physics,
because they were trying to understand how certain types of matter changed with time.
So by combining thermal equilibrium with a component of time,
this idea of physical models and physical structures,
that was supposed to just describe magnetism,
eventually led to a new discovery.
These networks could also explain something else.
And so in 1982, Hopfield, who by then was a theoretical biologist,
but did have a background in physics,
came up with a network that provided a detailed description
of how virtual neurons could potentially produce the idea of learning.
And specifically how they could do this using physical forces.
So he essentially reworked some of the math behind statistical mechanics
into the neural networks.
And interestingly, even the formula did not really change that much.
For example, in the physical models,
a lot of these magnetic elements try to maintain minimal possible energy
in order to maintain their network.
And something very similar seems to happen in neural networks as well.
Here the network is trying to achieve the minimal energy level.
And this was of course the basis for this Hopfield network.
And so in some sense, at least for Hopfield,
he received the Nobel because of his recognition of physical models
that could then be applied somewhere else.
And technically not just somewhere else,
but into a tool that is now widely used almost everywhere.
Importantly though, this is not just for materials and for neural networks.
It seems to be a very general idea,
potentially applicable in a lot of situations,
even when it comes to much larger structures,
or of course, people in a society.
So technically, the idea of these networks was way more groundbreaking
than we can even imagine today.
I'm sure there are going to be so many more discoveries on this topic
in many decades to come.
But what about Geoffrey Hinton?
Well, in his case, he's just a lot more important
when it comes to the development of modern neural networks.
And specifically in 2010, he published at least one paper
that highlighted the idea behind rectified linear units
being extremely important in Boltzmann machines.
Or just to rephrase this in simple words,
he basically connected a super important concept
that many neural networks rely on today,
which he was able to discover from some of these earlier networks
that were mostly based on biological models
and once again, statistical physics.
And so just the fact that he was able to find this connection,
which then led to a dramatic increase in neural networks
and research on the topic,
essentially makes him the godfather of AI.
And actually, if you actually go through some of these papers,
a lot of them are extremely popular,
with many of them cited over and over again.
So he obviously made a lot of contributions to the field.
For example, even his papers from 1986
on error propagation and backpropagation
were also just as crucial for some of these early neural networks.
And so here we have these two scientists.
One of them invented a kind of a neural network
used in psychology and a lot of cognitive sciences
that was entirely based on how atoms spin
and how they interact inside solid matter.
He applied statistical physics to neuroscience.
And then we had the second scientist
who took some of these ideas
and turned them into the modern field of artificial intelligence.
And naturally, there are a lot of other scientists
that are not mentioned here,
with many other scientists potentially deserving this Nobel as well.
In this case, the Swedish committee
decided to go with just these two people.
Now, whether this is controversial or not,
I mean, I guess time will tell.
But here, I do think that it's kind of difficult to disagree with them
because as you probably know,
AI is pretty much everywhere today.
And so in this case, I think the Nobel committee
wanted to find a way to celebrate this in some way.
But since the Nobel does not have a computer science component,
they had to find a way around it.
And so in some sense, this Nobel is a perfect example of science at work.
It's a reminder that science is very collaborative.
It depends on many different fields.
And in many cases, every new discovery is standing on the shoulders of giants
who developed previous ideas.
And so here, the field of artificial intelligence
actually does take place in statistical physics,
with all this being born as a result of earlier work in material sciences.
And so here, by calculating energy of the system,
just like we do in physics,
we can also exploit this idea to make neural networks learn.
So we can then go on ChatGPT and ask a bunch of stupid questions.
But there's obviously still that other side.
The side that still doesn't agree that statistical physics
should have won this award.
Because many people see statistical physics as a mathematical framework
that uses probability theory when it comes to microscopic entities,
with many physicists online expressing this discontent.
But then there's also the other side of the coin.
According to many of my friends in computer sciences,
every time they talk about neural networks,
every physicist loves to remind them
how modern neural networks came from statistical physics.
So I guess you just can't please everyone, right?
But since statistical physics arose entirely out of classical thermodynamics,
I think it's as physical as it gets.
So anyway, congratulations to these two wonderful people.
And honestly, I think they deserved it.
Especially because historically,
Nobel Prize awards had always had a few controversies here and there.
And not all of them in physics were also very physical.
For example, in 1956, John Bardeen, William Shockley,
and Walter Brattain won the award for developing semiconductors.
And though back then this was presented as discovering the transistor effect,
the invention of transistors led to a major development of industries around the planet.
And likewise, the development of neural networks
is most likely going to do something similar in the decades to come.
And so personally, I kind of see it in a very similar way.
It's a great achievement.
And physics did play a role to some extent.
But it was obviously not theoretical physics,
but a much more obscure field known as statistical physics.
Something that most of us have just never heard of before.
But now you have.
And in some sense, that's the beauty of the Nobel Prize.
It also serves as a super cool tool when it comes to science education.
So once again, congrats John Hopfield and Geoffrey Hinton.
And we'll probably discuss the other Nobel awards
in some of the future videos coming out really soon.
On that note, thank you for watching, subscribe,
share this with someone who loves learning about space and sciences,
come back tomorrow to learn something else,
support this channel on Patreon by joining a channel membership,
or by buying the Wonderful Person t-shirt you can find in the description.
Stay wonderful, I'll see you tomorrow, and as always, bye-bye.
