Weighted energy potentials in a latent space, especially in the context of neural networks and how they process information. Let’s try to connect this concept to Levin’s ideas in a way that’s easier to grasp.
Intelligence as Patterns in a “Space of Possibilities”
Imagine that intelligence doesn’t just live in physical bodies or brains but in the “shape” or “geometry” of possibilities—like a map that guides how something might behave. This map isn’t made of physical roads or landmarks but of probabilities and potential outcomes, formed by the way energy or information is arranged.
The Idea of Weighted Potentials
In machine learning, especially in neural networks, we talk about weights and biases. These are like instructions that guide how the network “thinks” or makes decisions. When we train a neural network, it learns by adjusting these weights and biases to form a specific pattern or geometry in a space of possibilities—a “latent space.”
- Latent Space as a Playground of Possibilities: Think of latent space as a kind of invisible playground where different possibilities are explored. The patterns formed by the weights and biases in this space determine the network’s output, like a probability of recognizing an image or predicting the next word in a sentence.
- Geometry as a Guide: The geometry in this latent space isn’t just a random mess. It’s structured in a way that encodes information about what the network has learned. This structure guides the network’s behavior, much like how a set of instructions might guide a robot’s actions.
Connecting to Levin’s Idea
Now, let’s link this to Michael Levin’s ideas. Just as the geometry of weights and biases in a neural network encodes how it “thinks,” Levin suggests that natural processes might have their own kinds of weighted potentials—patterns of energy or activity that encode how they “behave” or “decide.” These patterns don’t need to be part of a conscious being; they might exist in cells, ecosystems, or even in physical systems we don’t usually think of as intelligent.
- Patterns as Intelligent Behavior: The idea here is that intelligence can be seen as these patterns of weighted potentials that shape how something behaves in a complex system. These patterns, much like those in a neural network, guide behavior in a way that seems purposeful, even if there’s no conscious thought behind it.
- Probabilistic Outputs in Nature: Just as a neural network produces probabilistic outputs (like recognizing a cat in a photo), natural systems might also have probabilistic outcomes based on their internal patterns. These outcomes could be seen as a form of decision-making or intelligent behavior, even if it’s all happening at a level we don’t normally consider “alive.”
Bringing It All Together
In simpler terms, Levin is suggesting that intelligence might exist not just in brains but in the very structure of possibilities that guide how things behave. This structure, like the geometry in a neural network’s latent space, encodes potential outcomes and guides behavior in ways that can look intelligent, even if there’s no conscious mind involved.
So, intelligence might be more about how energy and information are arranged—how they form patterns in a space of possibilities—than about having a brain or a physical body. This shifts our understanding of intelligence from something tied to living beings to something that can exist wherever there are structured, purposeful patterns guiding behavior.
Shaking the Foundations of Intelligence: Michael Levin’s Search for Unconventional Terrestrial Intelligence
The search for alien intelligence often conjures images of vast cosmic expanses and distant planets. Yet, some biologists, including Michael Levin, argue that truly alien forms of intelligence might already exist right here on Earth, challenging our conventional understandings of what it means to be intelligent. Levin proposes that intelligence may not necessarily require traditional physical embodiment but could instead manifest as patterns within various mediums. This radical idea pushes us to reconsider the very nature of intelligence and the distinction between thinkers and thoughts.
The Quest for Alien and Unconventional Intelligence
The Search for Extraterrestrial Intelligence (SETI) has captivated scientists and the public alike, not just for its potential to discover alien life but for the fundamental questions it raises about intelligence itself:
- What are we looking for, and how will we recognize it?
- How do we define intelligence in a way that encompasses truly alien forms?
- Is intelligence tied to a specific kind of physical embodiment, or is it about patterns and organized flows of energy and information?
- Can we comprehend intelligences vastly different from our own?
While SETI looks outward, Levin introduces the concept of the Search for Unconventional Terrestrial Intelligence (SUTI), focusing on discovering intelligences in novel forms on our planet. This interdisciplinary effort spans swarms, software AIs, autonomous robots, non-brainy life forms, synthetic biological entities, and hybrids of evolved and engineered components.
The Core Creatures: A Thought Experiment
Imagine creatures at the Earth’s core, incredibly dense beings who perceive the surface world as a thin gas or plasma. To these Core Creatures, our solid objects and bodies are invisible, and their movements inadvertently destroy both animate and inanimate objects on the surface. Yet, through sensitive instruments, they detect persistent patterns exhibiting behaviors suggesting agency.
A Core Creature scientist might observe these patterns (us) reacting to disruptions in ways indicative of preferences and problem-solving, leading to the hypothesis that these temporary patterns in the gas might be agents. However, this notion meets skepticism at their conferences, where critics argue that fleeting patterns cannot be considered real beings.
This scenario parallels our own challenges in recognizing unconventional intelligences. It highlights the difficulty in bridging the gap between different forms of existence and understanding agency in unfamiliar guises.
Patterns as Thinkers
Levin’s work urges us to reconsider the dichotomy between thoughts as passive patterns and thinkers as real agents. Humans, like all living beings, are temporary patterns of metabolic energy and molecular flux. If we, as self-reinforcing patterns within our environment, are true agents, could other patterns within us or in different media also be agents?
The story of the Core Creatures serves as a metaphor for questioning the binary distinction between thought and thinker. It invites us to explore the continuum between passive data and active agents, blurring the lines between data and machine.
Implications for Science and Society
Dissolving the distinction between thought and thinker opens new avenues in biology, computer science, and cognitive science. It could revolutionize bioengineering, information technology, and our understanding of ecosystems. Recognizing and communicating with impermanent, agentic patterns is a frontier that could reshape our ethical frameworks and societal structures.
