What is life?

Foundations

A question with no answer

Ask a biologist to define life and watch what happens. You will get a list, and then you will get a series of small retreats as you push on each item.

This is not a gap that anyone has failed to notice. One survey collected over a hundred published definitions and tried to distil them into one sentence. The reason there are a hundred is that every one of them either lets something in that we do not want, or shuts something out that we do.

The list, and why it breaks

The school textbook usually gives seven properties: movement, respiration, sensitivity, growth, reproduction, excretion and nutrition. Test them.

Fire moves, grows, consumes fuel, produces waste, responds to its surroundings, and spreads by making more fire. Four or five out of seven, and nobody thinks a flame is alive.

A crystal growing in a saturated solution takes in material, gets bigger, and makes copies of its own structure, template and all. Not alive.

A mule cannot reproduce. Neither can a worker bee, or a person past a certain age, or one who chooses not to. Reproduction cannot be a requirement for an individual to count as alive.

A virus carries genes, replicates, mutates and evolves under selection, and does none of it without a host cell to borrow machinery from. It has no metabolism of its own. Depending on which property you decide is essential, it is either alive or an elaborate chemical.

Every item on the list has a counterexample. The definition problem is not that biologists are careless. It is that life is a category we recognised long before we understood it, and the category may not have sharp edges.

The definition that is actually used

Because the search for life elsewhere needs something operational, NASA adopted a working definition: a self-sustaining chemical system capable of Darwinian evolution.

Notice three things about it. It says chemical, so it is about matter and not about spirit or information alone. It says self-sustaining, which rules out the virus and the crystal. And it says capable of Darwinian evolution, which is the load-bearing clause, because replication with heritable variation under selection is the one thing on the whole list that non-living chemistry has never yet been shown to do on its own.

Notice also the word working. It is a tool for deciding where to point an instrument, not a claim to have settled the question.

What all life shares

Whatever life is, everything alive on Earth does it the same way, and the similarities are far too specific to be coincidence.

Every cell is wrapped in a membrane of fatty molecules. Every one stores information in nucleic acids and builds machines out of proteins. Every one uses the same energy currency, a molecule called ATP, spent and recharged billions of times a second. And every one reads its genes with almost exactly the same code: the same three-letter words mean the same amino acids in a bacterium, a mushroom and you.

There is no chemical reason it had to be this code rather than one of the astronomically many alternatives. The best explanation for why everything uses it is that everything inherited it, from a single ancestral population. That population has a name, the last universal common ancestor, and it is the deepest thing biology can currently see.

What life actually looks like, which is not what you think

Ask someone to picture life on Earth and they picture animals. Here is the accounting, by mass of carbon.

Plants are about eighty-two percent of it. Bacteria are about thirteen. Fungi, archaea, protists and viruses take up most of the rest. All animals together are under half a percent, and within that, arthropods outweigh fish, and fish outweigh everything else with a backbone put together.

Humans are about one hundredth of one percent of the biomass of the planet. And yet the livestock we keep now outweigh all wild mammals several times over, and our poultry outweigh all wild birds. A species that is a rounding error by mass has rearranged the rounding errors around it.

When it started

The Earth was assembled 4.567 billion years ago and had liquid water surprisingly soon after. The oldest widely accepted evidence for life is about 3.5 billion years old, which leaves a window of roughly a billion years for the transition.

There are older claims. Layered structures from Greenland have been argued to be 3.7-billion-year-old microbial mats, which would push life’s origin remarkably close to the end of the heaviest bombardment. That interpretation is contested, because layered rock can also form without biology, and telling the difference in metamorphosed rock this old is genuinely hard. The claim is worth knowing about and worth holding loosely.

Why the question matters

This is not word games. Three real problems depend on it.

If we find something on Mars or in the ocean of Enceladus, we will have to decide whether it is alive, using instruments built around assumptions we have not settled. If someone builds a self-replicating chemical system in a laboratory, we will have to decide whether they have made life. And if life ever arose on Earth twice, with different chemistry, every test we have would miss it, because every test assumes our biochemistry.

The honest position is the useful one. We do not have a definition. We have a working one, we know what it is for, and we know where it fails.

Deeper

Why definitions of natural kinds fail

Cleland and Chyba’s argument is that “life” is not the sort of thing a definition can settle. Definitions work on concepts, such as “bachelor”, where meaning is fixed by convention. They fail on natural kinds, where the category is discovered rather than stipulated. Nobody defined water into existence; the theory of molecular structure explained what water is, and only then did H₂O become the answer. On this reading, biology lacks a general theory of living systems comparable to molecular theory, and until it has one, a definition of life is premature rather than merely difficult.

The thermodynamic view

Living things maintain low internal entropy while exporting disorder to their surroundings, which is why Schrödinger described organisms as feeding on negative entropy. This framing captures something real: the boundary between a cell and its environment is maintained by continuous energy expenditure, and death is the relaxation to equilibrium. It also fails as a definition, because hurricanes, flames and convection cells are all dissipative structures that maintain organisation far from equilibrium without being alive.

The replicator view

Koonin’s argument is that the confusion around viruses dissolves once one distinguishes replicators from reproducers. A virus is a replicator that lacks a reproducing cell; a cell is both. On this account the question “is a virus alive?” is malformed, and the useful question is what kind of replicator it is and what it depends on. Viruses are also numerically dominant on Earth, outnumbering cells by roughly an order of magnitude, and carry a substantial fraction of the planet’s genetic diversity, which makes their exclusion from “life” a consequential piece of bookkeeping.

The genetic code as evidence

The standard code assigns 64 triplets to 20 amino acids plus stop signals. Its structure is not random: chemically similar amino acids tend to have similar codons, so single-letter mutations often produce conservative substitutions. This has been read as evidence that the code was optimised by selection early on rather than frozen by accident. Variant codes exist, notably in mitochondria and some ciliates, but they are minor deviations from the same scheme, which is exactly what descent with modification predicts and what independent origins would not.

The tree, and its roots

Doolittle’s central point is that rampant horizontal gene transfer among prokaryotes means different genes give different trees, so the base of the tree of life is better drawn as a network. The last universal common ancestor is therefore a population with a shared gene pool rather than an individual, and reconstructing its gene content is an active problem: different methods yield ancestral gene sets differing by hundreds of genes.

Frontier

A theory, not a definition

The productive move, several authors argue, is to stop seeking a definition and start seeking a general theory of the transition from chemistry to biology. Work on autocatalytic sets, on the thermodynamics of self-organisation, and on minimal cells with synthesised genomes is aimed at that. A theory would make the definition fall out, as molecular theory did for water.

The detection problem

Every proposed biosignature has an abiotic explanation. Isotopic fractionation, complex organic molecules, chemical disequilibrium in an atmosphere and layered sedimentary structures are all produced by non-living processes under some conditions. Current thinking favours combinations of evidence in context rather than any single marker, and the Greenland stromatolite debate is a live demonstration of how hard that is even on our own planet.

Transcript and shot list

Target runtime about 7 minutes. Built as an elimination: each candidate definition is tested and fails, until what is left is an admission and a working tool. Shot types: ANIM is rendered from code with Manim, AI is a generated shot from Higgsfield with the shared style reference, REAL is agency imagery with credit, CAM is the narrator.

[00:00] S01 · ANIM · A biology textbook page listing the seven characteristics of living things. Each one will be crossed out in turn over the next two minutes.

There is no agreed scientific definition of life. Not a simplified one, not a provisional one. There is no definition, and it is not for want of trying: one survey collected more than a hundred published attempts. Here is why every one of them fails.

[00:25] S02 · ANIM · The textbook seven: movement, respiration, sensitivity, growth, reproduction, excretion, nutrition.

Start with the list you were taught. Movement, respiration, sensitivity, growth, reproduction, excretion, nutrition. Seven properties. Let’s test them.

[00:40] S03 · AI into ANIM · A fire spreading through dry grass; ticks appear beside movement, growth, nutrition, excretion, sensitivity, reproduction.

Fire moves. It grows. It consumes fuel and produces waste. It responds to wind. And it spreads by making more fire. That is five or six of the seven, and nobody thinks a flame is alive.

[01:00] S04 · ANIM · A crystal growing in solution, layer by layer, templating its own structure.

A crystal in a saturated solution takes in material, gets bigger, and copies its own structure exactly, template and all. Not alive either.

[01:15] S05 · ANIM · A mule; a worker bee; the reproduction criterion crossed out.

Now push the other way. A mule cannot reproduce. Neither can a worker bee. Neither can plenty of people. If reproduction is required, we have just excluded things that are obviously alive. Cross it off.

[01:35] S06 · ANIM · A virus docking with a cell, injecting genetic material, the cell’s machinery building copies; a tick for evolution, a cross for metabolism.

And then there is this. A virus carries genes, replicates, mutates and evolves under selection. It also has no metabolism whatsoever: outside a cell it is an inert particle. It borrows every machine it needs. Alive or not? It depends entirely on which property you decide is the essential one. That is a choice, not a discovery.

[02:10] S07 · ANIM · The phrase “a self-sustaining chemical system capable of Darwinian evolution” assembling; the word WORKING highlighted.

So what does anyone actually use? Because the search for life on other worlds needs something operational, NASA adopted a working definition. A self-sustaining chemical system capable of Darwinian evolution. Three things worth noticing. It says chemical. It says self-sustaining, which drops the virus and the crystal. And the load-bearing part is Darwinian evolution, because replication with variation under selection is the one thing on this entire list that non-living chemistry has never been shown to do by itself. And notice the word working. It is a tool for deciding where to point an instrument. It is not a claim to have answered the question.

[02:55] S08 · ANIM · Four universals appearing in turn across a bacterium, a mushroom and a human: lipid membrane, nucleic acids, proteins, ATP; then the genetic code table with identical assignments.

Here is what we do know. Whatever life is, everything on Earth does it the same way. Every cell is wrapped in a membrane of fatty molecules. Every one stores information in nucleic acids and builds machines from proteins. Every one runs on the same energy currency, spent and recharged billions of times a second. And every one reads its genes with almost exactly the same code. The same three-letter words mean the same things in a bacterium, a mushroom and you. There is no chemical reason it had to be this code out of the astronomical number of alternatives. The best explanation is that everything inherited it, from one ancestral population.

[03:45] S09 · ANIM · A treemap of Earth’s biomass: plants 82%, bacteria 13%, everything else small; animals a sliver; humans a sliver of the sliver; then livestock outweighing wild mammals.

Now, a correction to the picture in your head. Ask anyone to imagine life on Earth and they imagine animals. By mass of carbon, plants are about eighty-two percent. Bacteria about thirteen. All animals together are under half a percent. Humans are about one hundredth of one percent. And yet the livestock we keep now outweigh all wild mammals several times over, and our poultry outweigh all wild birds. A species that is a rounding error by mass has rearranged the rounding errors around it.

[04:35] S10 · REAL · A Pilbara stromatolite in outcrop, layered and domed (research or museum image; confirm credit line). S11 · ANIM · A timeline: Earth forms 4.567 Ga, liquid water by 4.4, accepted life by 3.5, contested claim at 3.7.

When did it start? Earth was assembled four and a half billion years ago and had liquid water surprisingly soon. The oldest widely accepted evidence for life is about three and a half billion years old, which leaves a window of roughly a billion years for chemistry to become biology. There are older claims: layered structures from Greenland argued to be microbial mats three point seven billion years old. That would put life remarkably close to the end of the heaviest bombardment. It is also contested, because layered rock can form without biology and telling the difference in rock this old and this cooked is genuinely hard. Worth knowing. Worth holding loosely.

[05:25] S12 · ANIM · Three consequences: a lander on an icy moon; a flask in a laboratory; a hidden second biochemistry invisible to a test that assumes the first.

None of this is word games, because three real problems depend on it. If we find something on Mars or in the ocean under Enceladus, we will have to decide whether it is alive, using instruments built around assumptions we have not settled. If someone builds a self-replicating chemical system in a lab, we will have to decide whether they made life. And if life arose twice here, with different chemistry, every test we have would miss it, because every test assumes our biochemistry.

[06:10] S13 · ANIM · The word “water” beside H₂O; the word “life” beside a blank; a caption: a theory first, then a definition.

There may be a better way to think about this. Nobody defined water into existence. The theory of molecules explained what water is, and only then did H two O become the answer. Biology has no comparable theory of living systems, which is why a definition of life is not so much difficult as premature. Get the theory, and the definition falls out.

[06:40] S14 · AI into ANIM · A microscope view of a single dividing cell, held; then the calendar.

So the honest answer to the simplest question in biology is: we know it when we see it, we can list what all of it has in common, we can date roughly when it started, and we cannot say what it is. Next lesson, the harder question. How did it begin?

Shot list

Two AI shots, eleven code-rendered shots, one research image. Estimated Higgsfield use: 3 to 6 generations including retries.