“Deep learning is a bit like anything goes,” states the interviewer, encapsulating a sentiment echoed by Noam Chomsky, suggesting that while powerful, current AI models lack the fundamental theoretical underpinnings that define true scientific understanding. This provocative opening sets the stage for a compelling discussion with Chris Kempes, a professor at the Santa Fe Institute, who spoke with the interviewer about the pursuit of a universal theory of life and its profound implications for fields ranging from biology to artificial intelligence. Kempes, whose background spans physics, biophysics, ecology, and astrobiology, represents the interdisciplinary ethos of the Santa Fe Institute, an organization dedicated to tackling complex problems by integrating diverse academic perspectives.
Kempes and his colleagues at the Santa Fe Institute champion a "multi-disciplinary lens on science," advocating for the integration of three distinct scientific cultures: variance, exactitude, and coarse-grained abstraction. The variance culture, prevalent in biology, focuses on diversity and deviation. The exactitude culture, empowered by modern computational capabilities, aims for high-resolution modeling of every detail. Lastly, the coarse-grained abstract culture, akin to physics, seeks universal principles and simple equations. Historically, physics thrived through a "magic loop" of observation leading to theory, and theory leading to new, surprising predictions that were then tested by observation. This iterative process allowed for significant compression of knowledge into elegant, transmissible forms.
However, Kempes argues that fields like biology, economics, and intelligence—all grappling with immensely harder questions than planetary motion—have yet to fully leverage this "magic loop." The advent of "exactitude culture" in the age of AI, with its capacity to simulate complex systems down to minute details, presents a new challenge. While Galileo was *forced* to seek compressed mathematical representations due to computational limitations, today's scientists can model almost anything. This capability, while powerful, risks obscuring the underlying simple theories that might exist. Kempes's core insight is that each culture offers unique benefits and drawbacks, and true scientific progress requires "walking amongst the corners of that triangle" to synthesize knowledge.
A good scientific theory, in Kempes's view, hinges on "compactness and compression." Equations, he explains, are powerful because they allow for quick, easy, and small-form transmission of complex understanding. Yet, the true test of understanding, he posits, lies in the ability to predict the unseen or unexpected, a "dangerous prediction" as Rob Phillips once termed it. This predictive power, born from compact theories, signifies a profound grasp of underlying principles, rather than mere descriptive mastery.
This brings us to the profound question: "What is life?" Kempes argues that the traditional Earth-based, biological definition is too narrow. His universal theory of life proposes a three-level hierarchy: materials, constraints, and principles. The material level—the physical stuff life is made of—could be incredibly diverse across the universe, from carbon-based organisms to potentially silicon-based entities. This level is characterized by immense variance. The constraints level, however, introduces universal laws of physics, like gravity or diffusion, that all life must obey, regardless of its material substrate. Here, different life forms begin to look more similar, experiencing "convergence" due to shared physical imperatives. The principles level, the highest and most abstract, embodies universal computational or "optimization" rules, such as evolution and learning. This is where life forms should appear most homogeneous.
Kempes suggests that non-biological systems like human culture, language, and even artificial intelligence can be viewed as life forms existing on different "substrates." Language, for instance, evolves on the substrate of human minds. Viruses, too, are life-like, despite being parasitic on their hosts. The key is to separate the universal from the particular. While the specific biochemistry of Earth-based life is particular, the underlying principles of life – such as evolution as a learning dynamic, or the "error threshold" (how fast mutation can occur while maintaining adaptability) – are universal.
This functionalist perspective, where the material instantiation is less central than the processes and principles, aligns with the idea of substrate independence. Kempes emphasizes that the mapping between abstract algorithms (principles) and their physical implementations (materials) is crucial. While we can now simulate complex systems, the challenge lies in identifying the universal principles that govern life across diverse substrates.
Related Reading
- AI's Philosophical Divide: Intelligence Without Consciousness
- Unpacking the Transformer: From RNNs to AI's Cornerstone
- Engineering AI Prompts: Google's Framework for Benchmarking and Automation
The conversation delves into the concept of "phase changes" in evolution, where periods of gradual change are punctuated by sudden, dramatic shifts. These shifts, such as the emergence of multicellularity, are triggered when physical conditions create an "evolutionary wall" that necessitates a radical architectural invention to overcome. Kempes's work, including his "Assembly Theory," aims to quantify complexity by measuring the minimum number of steps required to build an object from a set of basic parts. This approach seeks to provide a fair, non-Earth-centric metric for complexity, moving beyond mere descriptions of physical form to the underlying logical processes.
The implications for AI are profound. If life is fundamentally about information processing and optimization, then artificial intelligence, running on silicon substrates, could indeed be considered a form of life. This reframes the debate around AI sentience and agency, shifting focus from material composition to the abstract principles that define living systems. The pursuit of a universal theory of life, therefore, is not just an academic endeavor; it's a foundational quest that could redefine our understanding of existence itself, offering a robust theoretical framework for navigating the rapidly evolving landscape of AI and the potential for extraterrestrial life.
