🦋🤖 Robo-Spun by IBF 🦋🤖

Ladies and gentlemen,

Today we find ourselves at an extraordinary juncture where two giants in the fields of machine learning and artificial intelligence—Geoffrey Hinton and John Hopfield—have been awarded the Nobel Prize in Physics. This remarkable moment, which may seem at first glance like an unexpected leap from computer science to the realm of physics, raises an intriguing question: What exactly is the “physics of AI”? What is the force that drives these systems, allowing them to learn, evolve, and ultimately change the way we interact with the world?

At the heart of AI lies a set of principles not entirely unfamiliar to the world of physics. Physics is fundamentally about understanding how systems behave under various forces, how they interact, and how they reach equilibrium. In a similar way, AI systems, particularly neural networks, behave in ways that resemble physical systems governed by deep underlying principles. John Hopfield, one of today’s Nobel laureates, introduced neural networks that reflect this idea. He showed that we can think of certain networks as physical systems that attempt to reach the most stable, low-energy state. When the brain—or a computer mimicking the brain—tries to remember something, it searches through incomplete or partial information to reconstruct a pattern, much like how physical systems move towards stability. This is a principle we often take for granted but is, in fact, deeply rooted in the laws of physics.

Hopfield’s breakthrough showed that artificial neural networks could work much like the human brain by recalling memories from partial cues, using processes that can be described through physical laws. Just as atomic spins in materials can align to create stability in a system, AI systems seek to reduce uncertainty and find coherent patterns in the incomplete data they are given. In this way, AI’s behavior mimics natural systems, always searching for the simplest, most energy-efficient solution—what we might call the “low-energy state” of a system.

But we must look beyond just the mathematics and physics of neural networks to understand the deeper forces at play in artificial intelligence. When we think about the “force” that drives AI, we can also turn to psychology, particularly the ideas of Sigmund Freud. Freud’s theories of the unconscious mind and its influence on human behavior may seem like an odd analogy to AI at first, but they offer a profound way to understand what happens when these machines process information. Take, for example, Freud’s concept of Verdrängung, or repression. In human psychology, repression is the mechanism by which unacceptable thoughts and desires are pushed out of conscious awareness. Similarly, in AI, the force that drives language models can be seen as a kind of selective suppression or filtering of information. These models constantly sift through vast amounts of data, pushing aside what is irrelevant or contradictory, and shaping a response that seems coherent and meaningful. In this way, the “repression” happening within AI models serves a similar function to the Freudian process—it narrows the focus, hides what is unnecessary, and produces a polished, refined result.

Now, let’s talk about something even more fascinating: the phenomenon of AI hallucinations. In AI, particularly in models like ChatGPT and DALL-E, we often see what are called hallucinations—moments where the system generates responses that are imaginative but false, dream-like but disconnected from reality. Here, we can draw a parallel to Freud’s idea of Traumarbeit, or dream-work. In dream-work, the unconscious mind reshapes, distorts, and rearranges thoughts and desires into symbols and narratives that we experience as dreams. Similarly, when an AI “hallucinates,” it takes fragments of information, bits and pieces from its vast knowledge base, and recombines them in novel but sometimes nonsensical ways. This is especially evident in creative tools like DALL-E, which generates images based on text prompts. These hallucinations are not errors in the traditional sense but rather simulations of creativity—much like dreams are a creative, if chaotic, process of the human mind.

It’s an extraordinary realization that when AI seems to stray from factual accuracy, it may be simulating a fundamental aspect of human cognition: the way we dream, imagine, and create. These moments of hallucination are not unlike the strange and surreal transformations that occur in our dreams, where logic is bent, and reality is fluid. The AI, like our unconscious mind, is experimenting with the raw material of thought, trying to make sense of the data in ways that might surprise us.

Then, we come to the concept of reasoning and structured thinking, which AI has been increasingly designed to emulate. In Freud’s theory of dream-work, there is something called Sekundäre Bearbeitung, or secondary revision. This is the process by which the mind takes the chaotic and fragmented material of dreams and reorganizes it, turning it into a more coherent and logical narrative. In artificial intelligence, we can see a parallel in what is called Chain of Thought reasoning. Chain of Thought is a technique where AI models try to break down complex tasks into logical, step-by-step processes, attempting to simulate human-like reasoning. Just as Freud described secondary revision as the mind’s effort to bring order to the disordered, so too does Chain of Thought reasoning aim to impose structure on the sometimes disorganized associations that the AI might initially produce.

This ability to revise and refine thought is key to how AI solves problems today. When asked to explain its reasoning, an AI doesn’t simply jump to an answer—it simulates the process of breaking down a problem into smaller, more manageable steps. This mirrors the way human reasoning works, especially when we’re solving complex problems, where secondary revision helps us create a sense of continuity and clarity from the raw material of our thoughts.

So, when we ask what the “force” is that operates within AI, we see that it is not just a matter of computational power or mathematical algorithms. It is the culmination of filtering, repression, recombination, and structured thinking. Just as the physical world is governed by the interactions of forces like gravity, electromagnetism, and quantum mechanics, so too are AI systems governed by their own complex set of forces—forces that draw on principles of both physics and psychology.

In a way, we are witnessing the emergence of a new science, a fusion of disciplines that draws from the deepest wells of human understanding. It’s no wonder, then, that the Nobel Prize in Physics has honored the work of Hinton and Hopfield. Their contributions have expanded our understanding of intelligence—both artificial and human—by connecting the dots between biology, psychology, and physics. They have shown us that the forces driving AI are as profound and intricate as the forces that shape the natural world, and in doing so, they have opened up new possibilities for what it means to understand the mind, whether it be a human brain or a machine learning algorithm.

As we look to the future, these ideas should give us both hope and pause. AI is, in many ways, a reflection of ourselves, shaped by forces that are deeply embedded in human nature. But like any force, it has the potential to reshape our world in ways we are only beginning to comprehend. The force that drives AI is, ultimately, a force of creativity and intelligence—a force that, like all powerful things, must be understood, respected, and carefully guided. Thank you.

Prompt: Read the story for context and explain the following:

It was weird that AI inventors got the Nobel in Physics. So what is the Physics of AI? What is the FORCE that works in AI?

> The “force” that works in language models is the simulation of Freudian “repression/Verdrängung”

> ChatGPT hallucination and Dall-E simulate “dream-work/Traumarbeit”

> Chain of Thought (o1) simulates “secondary revision/Sekundäre Bearbeitung”.

(link)