Before we build technology that proves a human made something, we need to decide what “human-made” actually means.

A strange little file extension has been following me around lately.

.pol.

Proof of Life.

The idea was publicly floated by musician and technology entrepreneur will.i.am as a possible way of distinguishing human-created work from machine-generated output. An attendee recounting his Cannes discussion described the proposal as a new .pol file type intended to signal the thought, effort and creative process behind human work rather than treating it as equivalent to machine generation. (LinkedIn)

It is exactly the sort of idea that sounds slightly eccentric until you try to engineer it.

Then the questions begin.

What would such a file actually prove?

That a human being was physically present?

That a particular human was present?

That the human touched the instrument or computer?

That they made meaningful creative decisions?

That they originated the idea?

That they authored the final work?

Or simply that they were somewhere nearby while a machine made it?

Those are not the same claim.

And before we begin attaching heartbeats, fingerprints and timestamps to creative work, we should probably decide which one we are trying to make.

A living human is not necessarily the author

Imagine a writer.

She authenticates herself on her laptop. Her fingerprint confirms that she is present. A trusted clock records the time. Her keyboard records genuine human keystrokes for two hours.

Proof of life?

Certainly.

Proof that she wrote the text?

Not necessarily.

She could be manually transcribing something generated elsewhere.

That is not merely a philosophical loophole. Recent research into keystroke-based AI-authorship detection demonstrated exactly this weakness: authentic human typing signals can establish that a person operated a keyboard while failing to establish that the person originated the content being typed. (arXiv)

That distinction gives us our first boundary:

Proof that a human performed an action is not proof that the human authored its meaning.

The same problem appears everywhere.

A drummer physically playing a rhythm has produced strong evidence of performance.

A painter whose stylus records every stroke has produced strong evidence of manual image construction.

A writer who slowly develops a manuscript through thousands of revisions has produced evidence of an evolving writing process.

But even then, none of those signals can directly tell us what happened inside the creator’s mind.

Technology can observe actions.

It cannot inspect intention.

That is where the phrase Proof of Life becomes simultaneously powerful and dangerous.

We are actually asking at least four questions

The AI-authorship debate often collapses everything into one binary:

HUMAN MADE

or

AI MADE

Real creative practice has already become much messier than that.

A useful provenance system would separate at least four questions.

1. Was a human present?

This is the narrowest claim.

A device might establish that a live person successfully authenticated during a creative session.

Modern authentication systems already demonstrate how this can work without transmitting someone’s fingerprint or face to every service they use. WebAuthn, for example, can perform biometric verification locally on an authenticator and return a signed indication that user verification occurred, rather than sending the biometric itself to the relying service. (W3C)

That can prove something useful:

a verified user was present for this authentication event.

It still does not prove authorship.

2. Did the human participate in making the artifact?

Now we need process evidence.

For a writer:

typing, deletion, revision, navigation, restructuring, source consultation, document versions.

For a visual artist:

strokes, pressure, layer changes, erasure, tool selection, transformation history.

For a musician:

performance events, arrangement changes, recorded takes, editing sessions.

This gives us a stronger claim:

human activity was materially connected to the development of this artifact.

Still not necessarily authorship.

3. What did the human contribute?

This is where provenance becomes interesting.

Perhaps the writer conceived the thesis but used AI to generate ten possible outlines.

Perhaps she wrote the entire first draft but used AI for copy-editing.

Perhaps an artist generated reference images, constructed the composition manually and used an AI tool for one background element.

Perhaps a film director never touches the editing software but makes hundreds of consequential decisions about what remains in the final cut.

Now the question is no longer whether AI was used.

It is:

Who did what?

4. Who held authority over the work?

This may be the most neglected question.

Who chose the goal?

Who decided what the work meant?

Who selected between alternatives?

Who rejected machine output?

Who accepted the final version?

Who bears responsibility for what was published?

A system can contain substantial automation while remaining human-led.

Conversely, a human can perform many mechanical actions while exercising very little meaningful authorship.

Human labour and human authority are related.

They are not identical.

Provenance is already moving beyond simple labels

There is useful technical precedent here.

The Coalition for Content Provenance and Authenticity, or C2PA, provides an open standard for attaching cryptographically bound provenance information to digital assets. Its Content Credentials can record assertions about how an asset originated or changed, actions performed on it, ingredients used in its creation, software involved and cryptographic bindings to the resulting content. Its current specification also explicitly supports identifying content created through generative AI and representing ingredients used as inputs to computational processes. (C2PA Specification)

That is already considerably richer than stamping:

AI ✅

or

HUMAN ✅

And the distinction matters because provenance is not fundamentally a purity test.

It is a history.

Something closer to:

Human:
    originated concept
    selected sources
    wrote sections 1–4
    rewrote conclusion
    approved final publication

AI system:
    generated outline alternatives
    suggested sentence edits
    retrieved references

Image model:
    generated two reference studies

External photographer:
    supplied Figure 3

Final authority:
    human author

Now we can actually discuss the work.

Then what would .pol add?

This is where will.i.am’s question becomes useful.

The public .pol idea is currently better understood as a provocation than a published technical specification. That leaves room for a more interesting engineering question:

What evidence of human involvement could accompany provenance without pretending to prove things technology cannot know?

Perhaps .pol should not mean:

This work is definitely human-authored.

Perhaps it means something narrower:

Verified evidence of human participation in the documented creation process accompanies this artifact.

That is a much more defensible claim.

A future system might attest to combinations of:

• user verification;
• trusted timestamps;
• device integrity;
• signed creation sessions;
• artifact revision hashes;
• creative-tool events;
• process continuity;
• declared AI interventions;
• human approvals;
• provenance relationships.

And crucially, those should remain separate claims.

Not one magical Humanity Score.

A creator might have:

Identity verification        VERIFIED
Human presence               VERIFIED
Artifact/session binding     VERIFIED
Revision history             VERIFIED
AI-tool disclosure           VERIFIED
Creative authorship          CLAIMED + EVIDENCED
Final human authority        VERIFIED BY SIGNATURE

That is much more intellectually honest than:

ORGANIC HUMAN CONTENT™ ✅

The surveillance trap

Unfortunately, once we start trying to prove human creative involvement, an obvious temptation appears.

Record more.

Keystrokes.

Mouse movement.

Stylus pressure.

Face.

Voice.

Heartbeat.

Location.

Watch telemetry.

Eye movement.

Pauses.

Browsing history.

Biometrics.

If one signal is uncertain, collect five.

If five are uncertain, collect twenty.

Eventually we have solved the problem of AI provenance by constructing industrial surveillance of the human creator.

That should worry us.

Researchers exploring process-based authorship verification are already confronting this contradiction. One recent proposal specifically identifies behavioural authorship evidence—typing dynamics, editing behaviour and related process data—as potentially sensitive enough to reveal information about identity, health and cognitive patterns, and proposes zero-knowledge techniques that would allow some properties to be verified without revealing the underlying behavioural record. (arXiv)

Whether that particular architecture ultimately proves practical is less important than the principle underneath it:

Verification does not require maximal disclosure.

That should become a design rule.

Do not record a creator until their humanity becomes undeniable.

Prove the minimum claim necessary.

If a secure device can attest that an authorised human completed a liveness check, perhaps the provenance record needs only the signed attestation—not the fingerprint that produced it.

If a writing system can attest that a document evolved incrementally through authenticated creative sessions, perhaps the public record needs the proof—not every pause, deleted sentence and midnight crisis that occurred along the way.

Privacy and provenance should not be enemies.

A good architecture should make privacy part of the proof.

We cannot prove thought

There is one thing .pol should never promise.

Thought.

A creative system can record that a writer stopped typing for eighty seconds.

It cannot tell us why.

She may have been solving a structural problem.

She may have been rereading the paragraph.

She may have been answering the door.

She may have been staring blankly at the ceiling.

She may have gone to make tea.

The computer does not know.

What technology can preserve is the history through which thought became form.

A paragraph existed.

Then it changed.

A section moved.

A sentence disappeared.

An idea appeared earlier in the manuscript.

A private process note existed before a later revision.

A source was consulted.

An AI suggestion was rejected.

Another was accepted and rewritten.

The system cannot say:

This is what the creator thought.

It may be able to say:

This sequence of authenticated actions produced this sequence of artifact states.

That is provenance.

And provenance does not need to become mind reading in order to be valuable.

Perhaps we are asking the wrong binary question

The arrival of generative AI has produced an understandable desire for a clean boundary.

Was this human?

Or machine?

But creative systems increasingly look more like collaborations among people, software, models, instruments, datasets and institutions.

The important question may therefore not be:

Was AI involved?

It may be:

What was the division of agency?

Who originated?

Who executed?

Who transformed?

Who selected?

Who verified?

Who authorised?

Who is accountable?

Those questions work whether the creator used a pencil, Photoshop, a synthesiser, a language model or a machine shop.

They also preserve something more important than nostalgia for a pre-AI definition of creativity.

They preserve the human’s place in the causal story.

Proof of life should remain modest

I like will.i.am’s .pol idea precisely because it forces an uncomfortable engineering question.

If society really does want a way to recognise human involvement in creative work, what exactly should technology be permitted to verify?

The answer cannot simply be:

everything.

And it cannot honestly be:

thought.

A useful proof-of-life architecture would probably sit beside provenance rather than replace it.

Proof of Life asks:

Was a verified human meaningfully present in this process?

Provenance asks:

What happened to the work, and through which actors and tools?

Authorship asks:

Whose creative contribution is being claimed?

Human-led governance asks:

Who retained meaningful authority and responsibility?

Those are different claims.

Keeping them separate may be the only way to make them trustworthy.

Before we build a file that proves a human made something, then, perhaps we should resist the urge to prove the human.

Prove the contribution.

Prove the process.

Prove the provenance.

Prove the authorization.

And leave the person’s body—and whatever was happening inside their head—as private as we possibly can.

---

Research note

The .pol / Proof of Life concept is attributed here to will.i.am. Our discussion of possible implementations—process attestation, privacy-preserving verification, relationship to OriON-style provenance, and the distinction among presence, contribution, authorship and authority—is an independent exploration prompted by that public concept. Contemporary standards such as C2PA already provide mechanisms for cryptographically bound content provenance, while WebAuthn demonstrates an existing privacy-conscious pattern for verifying a user without sending raw biometric data to a relying service. (LinkedIn)