electric sheep

The internet has given us access to more information than any generation in history. It has also created a difficult question: how do we know what to trust?

A claim can travel around the world before most readers have had a chance to ask where it came from. Search engines can summarise it. Social networks can repeat it. Artificial intelligence can turn it into a polished paragraph. Yet presentation is not evidence.

That is the problem behind Public Knowledge Ledger (PKL): an experimental approach to making public claims easier to trace, inspect, review and challenge.

Why Is It So Difficult to Know What Is True Online?

The problem is not simply that false information exists. Information loses context as it moves. A headline may simplify a study. A statistic may be repeated without its original definition. A genuine photograph may be paired with the wrong event. A confident AI answer may contain a claim whose source the reader never sees.

Good verification therefore means more than asking whether a website looks respectable. It means following important claims towards their evidence, checking context, comparing independent sources and making the reasoning visible.

What Is a Public Knowledge Ledger?

PKL starts with a simple idea: treat the claim as a first-class object.

Instead of burying a factual statement inside thousands of words, a knowledge system can record the claim explicitly and connect it to supporting or challenging evidence. That record can also preserve information about where the claim came from, how it was reviewed and how its status changes.

The word ledger matters. A ledger is not merely a page displaying the latest answer. It is a record. In PKL, the ambition is an append-only approach in which knowledge can develop without quietly erasing the path that produced it.

Claims, Evidence and Provenance

Three ideas sit at the heart of the project.

1. Claims

A claim should be clear enough to inspect. “Exercise is good” is broad. A specific statement about an intervention, population and measurable outcome is much easier to evaluate. Clear claims make disagreement clearer too: people can argue about the evidence rather than arguing past one another.

2. Evidence

A claim is not made reliable simply because many webpages repeat it. The important question is what those repetitions ultimately depend upon. Evidence might include original research, official datasets, primary documents, systematic reviews or other relevant sources. Different evidence has different strengths and limitations, and disagreement should remain visible.

3. Provenance

Provenance is the trail behind information: where it came from, what supports it, who reviewed it and what happened to it over time. In an information environment increasingly mediated by AI, that trail becomes particularly valuable.

Why Source Transparency Matters

When readers can see the route from a claim to its evidence, they do not have to accept an authority on trust alone. They can inspect the route themselves.

This does not magically settle every argument. Evidence can conflict. Studies can be flawed. Definitions can change. New discoveries can overturn old conclusions. But a transparent system can expose those uncertainties instead of hiding them behind a single confident verdict.

Fact Checking Is a Process, Not a Badge

It is tempting to imagine fact checking as attaching TRUE or FALSE to a sentence. Real knowledge is usually more complicated. Verification often means tracing a statement back to an original source, examining the context in which the evidence was produced and comparing it with independent material.

That makes history important. If a claim changes, the previous version and the reason for the change can matter. If reviewers disagree, the disagreement itself may be useful information. A knowledge ledger can preserve that history.

What Happens When AI Enters the Information Chain?

Generative AI makes this question urgent. AI systems can summarise huge quantities of material and make difficult subjects accessible, but fluent language is not proof of accuracy. An AI system can misunderstand evidence, lose context or produce an incorrect statement convincingly.

The answer is not to abandon AI. It is to build stronger connections between machine-generated answers and inspectable evidence.

Imagine an AI answer in which an important factual claim can be followed into a structured public record: the precise claim, its evidence, provenance, reviews, challenges and revision history. The AI becomes an interface to knowledge rather than an invisible replacement for its sources.

Independence and Review

A trustworthy ledger also has to think carefully about review. Ten approvals are not necessarily ten independent confirmations if all ten ultimately come from the same source, organisation or evidence family.

That is why provenance and reviewer independence matter together. A useful system should make conflicts of interest visible and avoid creating an illusion of consensus by counting closely related confirmations as though they were wholly independent.

Can a Ledger Eliminate Misinformation?

No. No database, fact checker, AI model or technical standard can eliminate misinformation.

PKL is interesting for a different reason. It changes the question from “Who should I believe?” towards “What is the claim, what evidence supports it, where did that evidence come from, who has reviewed it, and what has changed?”

That is a healthier architecture for uncertainty. A system does not need to pretend to possess final truth in order to make the evidence easier to inspect.

How to Verify Information Online Today

Even without a public knowledge ledger, the same principles are useful. Identify the exact claim. Find the original source rather than relying on a screenshot or repost. Check whether the source actually says what is being claimed. Look for independent confirmation. Check dates and context. Distinguish evidence from interpretation. Be suspicious of certainty when the underlying evidence is uncertain.

Most importantly, preserve the chain. A citation should not merely decorate an article; it should allow a reader to travel backwards towards the evidence.

From Pages of Information to Networks of Evidence

The web was built around documents and links. The next step could be a richer layer in which individual claims have durable relationships to evidence.

That would be useful to researchers and journalists, but it could also matter to search engines, educators, public institutions and AI systems. A machine trying to answer a question could potentially retrieve not merely a popular webpage but a structured history of the claim it is about to repeat.

Public Knowledge Should Be Challengeable

There is another principle behind PKL: public knowledge should not become untouchable.

A claim that cannot be questioned is not strengthened by technology. A useful knowledge system should make it possible to submit new evidence, challenge existing evidence and record uncertainty. Strong claims should survive scrutiny because their evidence is strong, not because a platform has made them difficult to contest.

The Bigger Idea

PKL is still an evolving project, but the underlying problem is already here. We are entering a world in which creating information is becoming extraordinarily cheap. Generating text, images, audio and video can take seconds. The scarce resource may increasingly be something else: traceable trust.

When information is abundant, knowing where a claim came from and why anyone should believe it becomes more valuable, not less.

A public knowledge ledger is one attempt to explore that future: claims connected to evidence, evidence connected to provenance, review made visible, disagreement preserved and knowledge allowed to change without losing its history.

We may never build a machine that tells us what is true once and for all. Perhaps we should not try. A better goal may be to build systems that make it easier for people — and machines — to see why something is believed.