The original post appeared on LinkedIn. You can view it below on Linkedin or scroll below for the web version.
Back when I was in college, AI was just becoming the next big thing. Anguilla, a small Caribbean island, had just hit an unexpected goldmine that boosted its GDP by over 20%—the ownership of the domain (TLD) .ai. This surge driven, mainly, by startups realizing that shifting from .com to .ai could significantly boost their chances of securing funding
Fast forward to last week, and AI's impact has crossed yet another milestone—claiming a Nobel Prize in Physics. This milestone shows just how much the field has grown, but the story behind it stretches back much further.
Imagine trying to recreate the human brain inside a computer—sounds like science fiction? About 80 years ago, that’s exactly what scientists began to explore. They hoped to create artificial neural networks that could learn just like our brains do.
In 1982, John Hopfield invented a new algo based on something similar. It worked just like associative memory of our brain where incomplete or slightly distorted patterns are filled up with the most similar stored pattern. Hopfield used his foundation in physics and the spin of atoms to make a model network with nodes and connections of different strengths. This network could worked like associative memory to predict what a distorted or blurred image originally was using energy in spin systems. It came to be known as Hopfield Network
Geoffrey Hinton expanded the Hopfield Network using ideas from statistical physics. It means it is easier to define the collective properties than that of seprate molecules. The states in which individual components can jointly exist was analysed using statistical physics and their probabilities were calculated. in 1985, Hinton published the Idea of Boltzmann Machine which used Ludwig Boltzmann’s equation to predict which states are more probable than the others. While the energy of the network of nodes as a whole remained same, the Boltzmann Machine updated individual node values, one at a time, to generate new pattern. This was an early example of generative model which we now witness in ChatGPT, Gemini etc
Hopfield in 1980s used a network of 30 nodes giving way to 435 connections. He tried to create a network with 100 nodes but it was too complicated for the computers in use back then. Today, large language models contain more than one trillion parameters. Though a machine can never think, we have come so close to mimicking the brain that we have experts warning against AI going sentient - capable of thinking about thinking!
Machine Learning and artificial intelligence have invaded all spheres of our lives. This is the 4th Revolution. This Revolution had its own humble beginning in labs. The Nobel Prize in Physics announced last week “for foundational discoveries and inventions that enable machine learning with artificial neural networks” to John Hopfield and Gerffrey Hinton honors these efforts.
We are happy but Anguilla must be happier!
Pic: Working with Excel, not AI :P
