Why physics Nobel was given to AI researcher ?

So we all may know that this year's Physics Nobel was given to John Hoppfield & Geoffrey Hinton
And it has sparks a great deal of controversy. So let's understand what they have done? Do they really fall into physics Nobel?

To understand their work we have to go back to 1980's to understand the paper that inspires this Nobel.

Hopfield Network

A Hopfield network is a type of artificial neural network inspired by the human brain. It's designed to store and recall patterns.

Let me give you some simple analogies and examples for Hopfield Network:

Analogy: The Beach Sand

Imagine a beach with small holes scattered across the sand. Each hole represents a specific pattern or memory stored in the network. When you drop a marble onto the beach, it will roll down the slope of the sand until it settles into the nearest hole.

• Marble: This represents the input data or query to the network.
• Holes: These are the stored patterns or memories.
• Rolling Process: This simulates the network's process of finding the closest match to the input data. As the marble rolls, it's essentially exploring the energy landscape of the network, seeking the lowest energy state, which corresponds to the most similar stored pattern.

Algorithm:

The above image determines how hopfield network or our brain neurons work.

• The first layer is the input: Suppose we saw something(blurry)
• The next layer is inspired by previous layer: Based on what we see our neurons send signals to other neurons. Anything that matched with our memory sends stronger signal
• More layers: Our brain tries to match something to what we saw
• Final layers: We find a match It's a book(suppose we saw a book) Our brain recognized it because it's in our memory

Let's compare this brainy activity with mathematical algorithm

Imagine we have a Hopfield network with three neurons(for simplicity) , each of which can be either +1 (on) or -1 (off). We'll use this network to store a simple pattern and retrieve

Pattern to Store:
Let’s say we want the network to remember the pattern [+1, -1, +1] for brain comparison that's our brain.

Training the Network (Setting Weights): We create connections between each pair of neurons to store this pattern.Weights are determined by which one matched with the memory most

The weight between neuron 1 and neuron 2, W(1,2), and all other weights are calculated based on the product of their values(It a simple one it can vary) in the stored pattern. For simplicity:
𝑊(1,2)=+1×−1=−1
W(1,2)=+1×−1=−1
𝑊(1,3)=+1×+1=+1
W(1,3)=+1×+1=+1
𝑊(2,3)=−1×+1=−1
W(2,3)=−1×+1=−1

We repeat for all pairs. These weights help the network "remember" the pattern [+1, -1, +1].

Retrieving a Pattern :
Now we give the network a noisy input, say [+1, +1, +1], which is slightly different from the original pattern.

Updating Neurons: Each neuron updates its state based on the sum of the weighted inputs from the other neurons(apart from him):

Neuron 1: Receives input from Neuron 2 and Neuron 3, which are both +1.
New state of Neuron 1 = sign(W(1,2) * Neuron 2 + W(1,3) * Neuron 3) = sign(-1 * +1 + 1 * +1) = sign(0) = +1
Neuron 2: Receives input from Neuron 1 and Neuron 3.
New state of Neuron 2 = sign(W(2,1) * Neuron 1 + W(2,3) * Neuron 3) = sign(-1 * +1 + -1 * +1) = -1
Neuron 3: Receives input from Neuron 1 and Neuron 2.
New state of Neuron 3 = +1 (since inputs confirm it should stay the same).

Done we have retrieved [+1,-1,+1]

so this is the contribution for Hopfield to get the Nobel.And this is the main basic for today's AI boom.Let's move to Hinton's Contribution

Boltzman Method

It's actually build on Hopfield method.In Hopfield we try to achieve a state from memory.But in Boltzman Method we don't directly read from memory we just have a training data and try to find out which ever output is making a sensible pattern (creating something new). This is used in modern generative AI

Analogy:

Imagine a large group of people at a party, each representing a neuron in a Boltzmann machine. The people are milling around, occasionally interacting with each other.
Just like in a Boltzmann machine, each person (neuron) has two possible "states" - they can either be "on" (dancing) or "off" (standing still). The interactions between the people represent the connections between the neurons, with the strength of the connections determining how much influence one person has on another.
The goal of the party, like the Boltzmann machine, is to find a stable configuration where the overall "energy" of the system is minimized. In other words, the goal is to find a configuration where people are arranged in a way that creates the most harmony and least tension between them.
Just as the Boltzmann machine uses a stochastic process to sample different configurations and move towards lower energy states, the people at the party will randomly change their dancing/standing states, moving towards a more harmonious arrangement over time.
The "temperature" of the Boltzmann machine is analogous to the "liveliness" of the party. At high temperatures, the people are more chaotic and willing to change states frequently. As the temperature decreases, the people become more hesitant to change and the system settles into a stable, lower energy configuration.

Algorithm :

If you take previous example that Hopfield method solves. Here it does 2 things differently

1.Hidden layer: Between Output and known layers from training data we have hidden layers whom weight is determined by probability of stable state
2.Back-propagation: Once a output is there it can back propagate to train again to get different output. IN this back and forth process it outcomes new things

Here output depend on the training data. IF we only train on [+1,-1,+1] then output will be that. If we train it like [+1,-1,+1],[-1,+1],[+1,-1,+1,-1,+1][-1,+1,-1,+1] ETC then it can generate something new like [+1,-1] etc

Is it a physics Nobel

In favour arguement:

Well as the initial Hopfield(physicists) is inspired by brain not for PC but to memory efficiency in physical world. And it was the main base for Hinton's work.So It's actually physics thing. Also this technique is used in every field from medical to economy so it's impact is unimaginable. Also many physics well known paper also used Hopfield as extension to their work

In against arguement:

First of Hopfield was not original it was also inspired from associate memory(in reference). Next if we are giving Nobel to work related to physical world then everything should be gone to physics.Also there is turing-award dedicated for CS. Now they won't be giving awards to physicists. Also The work was too old it's only appreciated because OpenAI popularized it.

My opinion:
There are great physics work out there. Giving their core Nobel to CS work is not right. The work is truely impactful but it's not physics core topic. In that sense AI has impact on trading so why not give economy Nobel to them

References:
Paper:
Hopfiled Paper
Hinton most popular paper

Videos:

Who are the winners actually
You Must Watch Her
Yet Another Computerphile
Hopfield Network
Generative Model

D Marr. Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci, 262(841):23–81, July 1971.

Kaoru Nakano. Associatron-a model of associative memory. IEEE Transactions on Systems, Man, and Cybernetics, SMC-2(3):380–388, 1972. doi: 10.1109/TSMC.1972.4309133.

S.-I. Amari. Learning patterns and pattern sequences by self-organizing nets of threshold elements. IEEE Transactions on Computers, C-21(11):1197–1206, 1972. doi: 10.1109/T-C.1972.223477.
W.A. Little. The existence of persistent states in the brain. Mathematical Biosciences, 19(1):101–120, 1974. ISSN 0025-5564. doi: https://doi.org/10.1016/0025-5564(74)90031-5.

J. C. Stanley. Simulation studies of a temporal sequence memory model. Biological Cybernetics, 24(3):121–137, Sep 1976. ISSN 1432-0770. doi: 10.1007/BF00364115.