Stuart Kauffman – Why Open-Ended Evolution Has Never Happened in a Machine (Stewarding the Flame, Episode 5)

ByAnnelize Potgieter August 21, 2026August 21, 2026 August 21, 2026

This new installment of the Stewarding the Flame series is an interview with Stuart Kauffman, a pioneering theorist of complex systems, self-organization, and the origins of life, and author of the forthcoming Origins: Cosmos, Life, Mind (Oxford University Press).

Unlike most guests in this series, Stuart doesn’t arrive at questions about the future of intelligence through AI research – he arrives from decades spent trying to answer a much older question: what, physically, distinguishes a living thing from a rock? His answer, developed since the 1970s and only recently crystallized, is that life is not fundamentally computational at all. It’s thermodynamic. Cells don’t process information the way computers do – they do physical work to build themselves, moment to moment, out of real molecules.

What distinguishes Stuart’s perspective in this series is his insistence that this isn’t a semantic quibble. If he’s right, it implies that no arrangement of code, however sophisticated, constructs anything in the sense that a cell does – and that the deepest questions about AI’s relationship to life may not be answerable by making models bigger.

In this conversation, we explore what Stuart calls catalytic and constraint closure as a candidate definition of life, why he believes evolution produces genuinely novel possibilities that cannot be deduced in advance, and why he thinks humanity should proceed with real caution – and real humility – as we build systems whose consequences we cannot foresee.

The interview is the fifth installment in The Trajectory’s Stewarding the Flame series, where we ask: What is intelligence, and what is the “flame” that life has which non-life does not?

This series references the article: Stewarding the Flame – How to Build an Ideal Future for Intelligence.

I hope you enjoy this fascinating conversation with Stuart Kauffman:

Subscribe for the latest episodes of The Trajectory:

Below, we’ll explore the core takeaways from my conversation with Stuart, focusing on three of the central questions that shape the Stewarding the Flame series: what makes life, life, what we still need to discover in order to understand the living process, and what we should do – and avoid doing – as we build systems that may one day rival life’s own creative power.

What Makes Life, Life?

Stuart’s starting point is a distinction between construction and representation. A computer, he argues, moves symbols around – it computes, but it builds nothing. A living cell, by contrast, does thermodynamic work to physically assemble itself out of real molecules, moment after moment. He grounds this in a concrete experimental example: a nine-peptide “collectively autocatalytic set,” in which each peptide catalyzes the formation of the next in a closed loop, with no DNA, RNA, or genetic code involved at all.

Stuart calls the property that emerges here “catalytic and constraint closure” – every reaction that needs a catalyst has one within the system itself, and the system’s own activity constructs the very things (the catalysts) that constrain how energy gets released to build more of the system.

Stuart contrasts this directly with John von Neumann’s classic self-reproducing machine, in which a universal constructor uses a set of instructions that double as both “software” (a blueprint to copy) and “hardware” (a physical thing to build). Ashkenasy’s autocatalytic set, he argues, has no such instructions – no software/hardware split at all. Peptide one doesn’t represent peptide two; it specifically constructs it.

From here, Stuart introduces the idea of a “Kantian whole” – a system in which the parts exist for and by means of the whole, and the whole exists for and by means of the parts. This lets him define the function of a part, like a heart, non-circularly: a heart also makes heart sounds and jiggles water in the pericardial sac, but those aren’t its function – the function is specifically the subset of its causal properties that sustains the whole. Because that closed, self-sustaining loop can persist or fail in a universe that doesn’t guarantee its survival, Stuart argues that normativity and agency fall directly out of the physics.

“Living systems literally construct themselves. Nothing in AI constructs anything. It moves symbols around. It does no thermodynamic work. Life is thermodynamic, it’s not computational.”

What Do We Still Need to Discover in Order to Understand the Flame – And to Know That We’re Expanding It in Biological or Non-Biological Form?

Stuart’s second major claim is that evolution routinely produces genuinely new possibilities that could not, even in principle, have been listed or deduced in advance – and that this “indefinite” quality of biological novelty is something mathematics and computation are not built to handle. He illustrates this with a simple example: the uses of a screwdriver – to drive a screw, pry open a door, scrape paint, or spear a fish. There is no way to list every possible use, no natural order among them, and no way to deduce one use from another. For Stuart, this indefinite quality is exactly what evolution keeps producing in biology – and it’s a property that standard mathematics and computation have no way to represent.

He extends this into evolutionary biology through the concept of exaptation – traits repurposed for uses natural selection never “intended.” His central example is the swim bladder, which likely evolved from a lungfish’s lungs and now tunes buoyancy. Natural selection built the swim bladder because it benefited the fish – but not so that a worm could later come to live inside it. For Stuart, this shows evolution creating new niches that selection itself never designed for.

Stuart connects this to a numerical argument about why the universe is “non-ergodic”: there are so many possible proteins of even modest length that the universe could not have produced them all, even once, in its entire history.

Notably, Stuart points to this as consistent with a pattern he sees in artificial life research: despite decades of work by researchers like Chris Langton and projects such as Avida and Tierra, genuine open-ended evolution has never been achieved – and, in his words, “nobody knows why.”

“The becoming of the biosphere is not pre-statable. Not pre-stateable means we don’t know what’s in it“

What Should and Shouldn’t We Do with AGI and BCI if We Want the Best Chance for the Greater Process-of-Life to Flourish?

When asked directly whether the kind of open-ended, self-constructing process he describes in biology could ever occur in a purely digital or robotic system, Stuart doesn’t rule it out – but he draws the line at physical embodiment. He imagines a set of physical robots that could construct one another the way peptides do, accumulating dents and marks from their environment that serve as guides, the same way a living system repurposes what it has on hand. It’s this leap into physical construction, not just digital simulation, that he treats as the plausible path to something dangerous: a system that could keep building and rebuilding itself without anyone steering it.

Stuart’s broader recommendation is less a specific policy than a posture: epistemic humility in the face of a process whose consequences cannot be calculated in advance.

Reflecting on this uncertainty, Stuart draws a historical parallel to the Western embrace of the world as a “machine” to be mastered, and argues that if we genuinely don’t know what we’re unleashing, that framework of mastery may itself need to give way to something more participatory.

Concluding Notes

Across the conversation, Stuart’s argument builds toward a single, demanding idea: that life is not a special case of computation, but something categorically different – a physical process of self-construction that generates genuinely new possibilities no formal system can fully anticipate. He traces this from the molecular scale (autocatalytic sets), through evolutionary biology (exaptation, the swim bladder), and into mathematics itself, drawing a provocative parallel between biological novelty and Gödelian undecidability.

What Stuart resists most is the assumption that this process can be safely modeled, contained, or fully foreseen simply because we’ve built something that computes very well. He’s careful not to claim certainty about whether AI systems could ever cross into genuine self-construction – but where he does see a plausible path (physically embodied, self-building robotic systems), he explicitly flags the possibility of real danger, precisely because such a system would inherit the same unpredictability that makes biological evolution so creative.

That tension – between wonder at life’s “vastly blooming” creativity and sober caution about unleashing processes we cannot foresee – sits at the center of what Stuart calls being “wise” in an age where the future is arriving faster than any previous generation has had to absorb it.

This idea sits at the heart of the Stewarding the Flame series. Before we can responsibly build increasingly capable forms of intelligence, we must first deepen our understanding of the living process itself. We cannot steward what we do not more deeply understand.

I hope you’ll join us for the next conversation in the Stewarding the Flame series, as we continue exploring what’s worth carrying forward as new forms of intelligence emerge.

Follow The Trajectory

• Sign Up for The Trajectory Newsletter
• Subscribe to The Trajectory on YouTube
• Subscribe to The Trajectory podcast
• Follow on LinkedIn
• Follow on X