What this covers

Stuart Bartlett, an astrobiologist at Caltech, proposes in conversation with host Sean Carroll a rethinking of how we define and recognize life. Rather than anchoring the concept to Earth-specific chemistry or Darwinian evolution alone, Bartlett argues that life is better understood through four functional pillars—dissipation, autocatalysis, homeostasis, and learning—which he and collaborator Michael Wong call lyfe (spelled distinctly from "life"). By defining living systems as processes rather than components, this framework opens the door to recognizing life in non-terrestrial chemistries, in the lab, or anywhere else in the universe, without being blinded by assumptions baked into our single known example.

The conversation ranges across multiple fronts of evidence and argument. Bartlett examines failures of existing approaches: Spiegelman's RNA experiments showed that selection for efficiency drives complexity downward, not upward, and digital artificial-life simulations plateau rather than continuing to innovate—both suggesting that Darwinian evolution alone may be insufficient. He walks through how each pillar functions, from the thermodynamic necessity of dissipation to the subtlety of learning, which requires not just memory but processing and feedback. The discussion also touches contentious terrain: whether life should be understood at the planetary or ecosystem level rather than the organism level, whether group selection and downward causation play roles that gene-level accounts miss (illustrated via social insects and the Bristlecone pine), and whether evolution itself functions as a form of associative learning at the biosphere scale.

Sharpest takeaway

Bartlett argues that 'life' as we know it is just one instance of a broader category 'lyfe' defined by four process-based pillars—dissipation, autocatalysis, homeostasis, and learning—and that defining life by these reusable functions rather than by Earth-specific components or Darwinian evolution alone gives us a better tool for finding and understanding life elsewhere and explaining its origins.

  • Darwinian evolution alone may be insufficient to explain life's open-ended complexity, as artificial-life and RNA experiments saturate rather than keep innovating.
  • The four pillars are processes rather than components, so they can be realized in non-terrestrial chemistries we might otherwise miss.
  • Life is better understood as a system- or planetary-level phenomenon than at the individual-organism level.

The argument · threads8 threads · 30 claims
0.72

Life requires four essential pillars: dissipation, autocatalysis, homeostasis, and learning as processes.

4 pointscentrality 5/5
  • The dissipation pillar requires that a living system be out of thermodynamic equilibrium and not an isolated closed system, because life always uses sources of free energy and dissipates it—at planetary scale taking in concentrated UV light and emitting less concentrated infrared, with a net loss of free energy.

    “dissipation, which just means that the system cannot be in thermodynamic equilibrium... life always uses sources of free energy and it always dissipates free energy”

  • The autocatalysis pillar means living systems exhibit exponential, non-linear growth when resources are abundant—as bacteria do in a Petri dish—because life's activity under abundant conditions always enhances that activity, creating the potential for positive feedback, whether through replication or exploiting new niches.

    “autocatalysis, which is another way of saying exponential growth. So we would always expect lyfe to exhibit exponential growth when it has access to abundant resources”

  • Homeostasis—negative feedback mechanisms that rein in key physical or chemical variables (like blood glucose or temperature)—is necessary to keep the autocatalytic property in check; multicellular organisms must regulate their cells' replication rate, and a cell that breaks this contract by replicating faster than the others produces cancer.

    “you need homeostasis to keep the autocatalytic property in check... if we have a stray cell which breaks that contract and replicates faster than all the others, then you end up with cancer”

  • Learning—the fourth and trickiest pillar—is defined as three abilities: interacting with the environment to record information, processing that information to change its form, and then using it so the organism increases its utility; a footprint on a beach is mere memory, but only if the recorded information is processed and fed back to benefit the system does it count as learning.

    “we define learning as three abilities. The ability to interact with the environment and record information. The ability to process that information... And then after processing, the use of that information such that the organism, it has somehow increased its utility”

0.68

Darwinian evolution alone may be insufficient to explain life's origin and the emergence of its complexity.

3 pointscentrality 5/5
  • Darwinian evolution—heritable variation leading to reproductive success—may be insufficient on its own to explain the diversity and complexity of life on Earth, so we cannot confidently rely on it to bridge from the origin starting point to the complexity we observe.

    “I'm unsure as to whether just Darwinian evolution is sufficient to explain the diversity of life on earth, so I'm not confident that we can rely on Darwinian evolution to explain the complexity that we see after that origins starting point”

  • In Spiegelman's RNA replication experiments, RNA molecules with heredity, reproduction, and selection for efficiency evolve toward simplicity rather than complexity, ending up at the minimal sequence capable of replication—contrary to the hope that selection would generate emergent novelty.

    “those RNA molecules evolve towards simplicity, not towards complexity. And actually, they end up evolving towards the minimal sequence capable of replication. And then they stop there.”

  • Richard Watson has proposed that evolution functions like a large associative (Hebbian) learning system, in which whole ecosystems and the biosphere learn about their environment and each other by observing correlations, becoming progressively more knowledgeable—a more general explanation than gene-level selection and one that could account for the formation of cooperative groups.

    “he has proposed that actually evolution is more like a large associative learning system... the whole system, whole ecosystems for example, are essentially learning about their environments and learning about one another and they're learning by observing correlations”

0.63

Astrobiology lacks consensus on life's definition, hindering objective-driven research and mission planning.

5 pointscentrality 4/5
  • Astrobiology, though at least 50 years old, has a broad scattering of opinions and a complete lack of consensus on almost anything, including the definition of life, which matters because so much of modern science is objective-driven and missions need a clear common goal.

    “it's a field which has a broad scattering of opinions and a complete lack of consensus on almost anything”

  • The less something is empirically constrained, the more certain people feel about their opinions on it—a pattern Bartlett observes in the strongly held but unconstrained views in astrobiology.

    “it's one of those situations where the less something is constrained, the more sure people feel about their opinions on it”

  • Defining life via Darwinian evolution and chemistry, as NASA does, risks being too 'acentric' (Earth-biased): because we have only one example of life to study it is hard to generalize, and a wrong or off-the-mark definition could cause us to miss extraterrestrial life that doesn't evolve by Darwinian selection or isn't chemically based.

    “what if there's life out there, which does not evolve by Darwinian Evolution, or maybe it's not chemically based, but it's there, what if we miss it because our definition is wrong or our definition is off the mark?”

  • Because we only have a single example of life (Earth life) to study, it is extremely difficult to generalize about what features are essential to life versus incidental to its terrestrial implementation.

    “Because we only have an example of life to study, that makes it very, very difficult to generalize”

  • The NASA definition of life—the most commonly adhered-to standard—is 'a self-sustaining chemical system capable of Darwinian evolution,' which embeds the assumption that once a prebiotic system shows Darwinian evolution, evolution will take over and produce the rest of life's complexity.

    “the NASA definition of life is a self-sustaining chemical system capable of Darwinian evolution”

0.57

Understanding life requires reconstructing it in artificial systems to generate open-ended novelty.

7 pointscentrality 3/5
  • Following Feynman's dictum that when you understand something you should be able to reconstruct it, if we truly understand the evolution of life on Earth we should be able to reiterate it in artificial systems and reproduce its open-ended generation of novelty.

    “As Feynman said, "When you understand something, you should be able to reconstruct it."”

  • In artificial-life digital evolution experiments, populations evolve to a point but then complexity saturates rather than achieving open-ended innovation, largely because parasites emerge that piggyback on stable replicators, draining the resources those replicators would otherwise use to generate new levels of novelty.

    “you have a set of parasites which come along with them, which emerged in the system, and they will essentially piggyback on the more stable replicators and those more stable replicators then find it difficult to generate new levels of novelty because their resources are being expended”

  • In simulations of the Selkov reaction, simple chemical spot patterns grow, divide and replicate like bacteria, and when two populations are coupled—one producing heat and one absorbing it—they can regulate their temperature and protect themselves from thermal damage by adjusting their relative populations, exhibiting self-replication and homeostasis from pure chemistry.

    “if one set of these spot patterns produces heat and the other population absorbs heat, they can actually regulate their temperature. And if you try and sort of kill them by heating up the system or cooling down the system, they will respond by changing their relative populations, and hence they will basically protect themselves from thermal damage”

  • The Bristlecone pine—possibly the oldest known organism at ~5000 years—has cones activated by fire, giving it a selective advantage because fire destroys non-resistant competitors, leaving abundant resources; whether learned via slow natural selection or a faster associative/Lamarckian within-lifetime mechanism, the species has 'learned' to use fire to its advantage despite fire being destructive to most organisms.

    “its cones, containing the seeds, are activated by fire... The selective advantage of that is that a lot of their competitors are not fire-resistant, so once the fire has blown over, it's quite likely that there will be an abundance of certain resources”

  • Because the four pillars are processes rather than objects or components, framing the origin of life around them frees our imagination from the unproductive 'which function came first' debates; it is likely that life began messily, perhaps multiple times with extinctions and exchanges among survivors, so the pillars may have emerged sequentially or all been present from the beginning—an open question.

    “that was part of our motivation for the four pillars because the four pillars are processes, they're not objects or components”

  • Some Hydra do not senesce—they lack programmed cell death—so they may be biologically immortal, with a 1998 study by Daniel Martinez claiming as much, though it remains controversial and unproven.

    “there are Hydra which, the technical term is they don't senesce... it was claimed that Hydra are immortal. But I mean it's a controversial piece of work”

  • Molecular walker molecules in modern life are powered not by ATP but by random thermal fluctuations: they wait for a kick, and when pushed in the desired direction they consume ATP to lock in the forward position, while doing nothing on a bad kick—an information-engine behavior analogous to Maxwell's demon that converts a measurement into directed motion.

    “their motion is not powered by the ATP. Their motion is powered by random fluctuations in the background, but the sequence of steps that they go through is selective in the sense that if the kick is good, they will move forward and lock that position in by burning the ATP”

0.54

Non-living dissipative systems lack one or more pillars, clarifying what distinguishes life from chemistry.

3 pointscentrality 4/5
  • Dissipative structures such as convection cells, hurricanes, and pattern-forming chemical systems like the Belousov-Zhabotinsky reaction arise in driven, far-from-equilibrium systems and look organized and process energy, yet are not alive—serving as crucial counterexamples for distinguishing what life has that they lack.

    “A convection cell being an ordered structure that arises in a differentially heated fluid... my other favorite example is pattern forming chemical systems”

  • Fire is often compared to life because it converts free energy, grows, and shows autocatalysis, but it is not homeostatic in general (it burns itself out) and crucially cannot learn—it cannot adaptively adjust its burning rate to avoid extinction—so it is not alive.

    “fire is not capable of learning, and hence, we don't consider it to be alive”

  • Taking the four pillars, there are eight subsets having fewer than all four; classifying non-living systems by which pillars they have clarifies what they lack—convection cells and hurricanes dissipate energy and may transiently replicate (vortices) or persist homeostatically (Jupiter's red spot) but never learn because no identifiable structure encodes, processes, and feeds back information.

    “we can take all of our four pillars and look at the different subsets and of course there are eight different subsets which have less than all four of the pillars”

0.54

Lyfe is defined by four pillars; Earth life is lyfe using nucleic acids, enzymes, and lipid membranes.

3 pointscentrality 4/5
  • Bartlett and Wong introduce 'lyfe' (L-Y-F-E) as a broad category defined by four criteria—dissipation, autocatalysis, homeostasis, and learning—with 'life' (L-I-F-E) being the Earth-specific subset that uses nucleic-acid information transfer, protein-based metabolism, and lipid membranes; this distinguishes general living processes from terrestrial implementation.

    “that larger set of entities satisfies four criteria which we've introduced... the four criteria to be satisfied for a system to be lyfe is dissipation... autocatalysis... homeostasis... learning”

  • Earth life (life with an i) is distinguished by metabolism based on amino-acid enzymes and chemiosmosis—pumping ions across a membrane and recovering that energy by letting them back through a molecular machine that forms ATP—with the rTCA cycle as the most important metabolic cycle; a system using nucleic acids and rTCA but a different membrane would be lyfe, not life.

    “Chemiosmosis refers to the pumping of ions across a membrane and the reverse process of allowing those ions back through a membrane and converting that energy into a different form”

  • A distinction is drawn between replication and reproduction: if offspring are all identical to the parent it is replication, whereas if there is variation among offspring it is reproduction, with variation being a necessary component of Darwinian evolution.

    “if they're identical then that's replication. If there's variation then it's reproduction”

0.50

Information processing and genetic coding emergence remain deeply enigmatic in life's origin.

2 pointscentrality 4/5
  • The deepest enigma in the origin of life is the emergence of information-processing mechanisms and the genetic code: why a physical system, given the second law's tendency for gradients to dissipate, would invent a coding system to translate genetic information into functional protein sequences remains poorly understood, though information engines and molecular walkers show individual molecules can process information and convert energy.

    “to me, the biggest enigma in the origin of life is the emergence of the information processing mechanisms”

  • Because the ribosome evolved by accretion—adding RNA loops onto earlier versions—researchers can identify where each section was tacked on, strip them away in reverse, and reconstruct a very primitive minimal ribosome that still translates genetic information into protein, potentially narrowing the gap in understanding life's origin.

    “you can keep doing that until you've actually stripped away almost all of the additions, and at that point, you get back to a very, very minimal primitive ribosome, which still has the ability to translate genetic information to protein sequence information”

0.40

Life operates as system-level phenomenon integrating multiple entities rather than organism-level property.

3 pointscentrality 3/5
  • Life is better understood as a system- or planetary-level phenomenon rather than something occurring at the individual organism level; individual entities plucked out of the system often fail to satisfy life's criteria (as viruses are integral to the biosphere though they can't survive alone), but the system as a whole does.

    “I would definitely, I would lean towards life being a planetary phenomenon rather than something that happens at the individual level”

  • The emergence of social behavior in insects is a conundrum for the view that gene-level Darwinian evolution explains everything: a worker ant is highly fit and works tirelessly for the colony but will never pass on its genes, so its cooperation yields nothing for its own genes—suggesting some non-traditional, downward-causal group-level effect is at work.

    “that individual worker ant, which is apparently doing really well is not gonna pass its genes on to the next generation, and it's cooperating in this collective, but its genes are not getting anything out of that cooperation”

  • The 'cooperation barrier' explains why higher organization fails to emerge: when selfish individuals begin cooperating into a more efficient collective, a single individual that reverts to taking resources without contributing is favored in the short term and out-competes the cooperators, collapsing the system back to its lower level of organization unless suppression mechanisms exist.

    “those cheaters, in that short term favorable situation where they can basically steal resources from the group, they will tend to out-compete the cooperators and take over the group, and then you go back to the previous level of organization without the cooperation”