Synthetic History and the Silent Memory-Hole

We have spent centuries worrying about people burning books, and for fairly obvious reasons. Book burning is wonderfully uncomplicated as a metaphor for censorship. You can see the flames, you can identify the person holding the match, and, if necessary, you can point dramatically at the smoke while making a speech about civilisation. It is difficult to misunderstand what is happening.

The modern version may be considerably less theatrical.

Nobody necessarily has to burn anything. The old newspaper can remain in the archive. The local historian’s website can remain online. The obscure forum thread can continue sitting on some forgotten server, complete with its arguments, accusations and spelling mistakes. The photographs can remain in a digital folder, and the handwritten letters can remain in a cardboard box in somebody’s loft. Technically, the information is still there.

The difficulty is that being available and being discoverable are no longer quite the same thing.

That distinction matters more than it first appears, particularly as search engines and generative AI increasingly become the intermediaries between ordinary people and the information they are trying to find. We have moved from a world in which people searched through archives, libraries, newspaper collections and increasingly chaotic corners of the internet towards one in which we ask a machine a question and expect it to return the relevant history in a handful of paragraphs. It is convenient, fast and, most of the time, remarkably useful.

It also introduces a curious problem.

What happens to the parts of history that do not survive the journey into a neat answer?

Local history has always been a messy business, which is probably one of the reasons I like it. Ask three people in an old village about an event that happened fifty years ago, and there is a reasonable chance you will receive four versions of what happened. One family remembers the local councillor as a thoroughly decent man who tried to resolve a difficult situation. Another remembers him as a corrupt bastard who spent twenty years feathering his own nest. Somebody’s grandfather swore that the whole affair was deliberately covered up, while somebody else’s grandmother regarded the entire thing as nonsense.

Then, just as everyone is beginning to lose patience, someone produces an old newspaper clipping from a drawer.

It contradicts everybody.

This is not necessarily a failure of historical memory. It is often what historical memory actually looks like. Human beings remember differently. They forget dates, exaggerate incidents, protect relatives, settle old scores, and occasionally turn an entirely ordinary disagreement into the stuff of village legend. People who were present at the same event can come away with completely different understandings of what they witnessed.

The argument between those memories is itself valuable.

An oral history does not have to be objectively correct in every detail to be historically interesting. A disputed recollection can tell us what a community believed. A family story can reveal something about how an event was understood decades later. A badly researched local blog may contain an error, but it might also contain a photograph, a name or a reference that leads another researcher towards something important.

History is not simply a list of facts that somebody has helpfully labelled TRUE. It is the surviving evidence, the interpretation of that evidence and, rather importantly, the disagreement surrounding it.

For a long time, the internet was surprisingly good at preserving that mess.

The early web was full of people who had no particular reason to be there other than an unhealthy obsession with something nobody else cared about. There were websites dedicated to abandoned railway stations, forgotten military units, local photographs, obscure buildings, family histories and village disputes that had apparently been simmering since the invention of the telephone. There were blogs written by people who would never have described themselves as historians but who had spent twenty years collecting newspaper cuttings and interviewing elderly neighbours.

Some of it was rubbish.

Some of it was magnificent.

Most of it was somewhere between the two.

That was part of the charm. The internet became a gigantic attic containing all the things that professional institutions had never considered important enough to preserve. It was chaotic, repetitive, badly designed and full of broken links, but it allowed ordinary people to become custodians of small fragments of collective memory.

The retired railway enthusiast could preserve photographs that might otherwise have disappeared when his house was cleared. The local woman with a passion for genealogy could record the names of people who never appeared in national histories. The eccentric historian could spend twenty years arguing that the official version of a local event was wrong, complete with enough supporting material to make you wonder whether perhaps the eccentric historian had a point.

The important thing was not necessarily that the eccentric historian was right.

The important thing was that he existed.

His version was part of the record.

Then the internet matured, and gradually the question changed from what information existed to which information was worth showing people.

Search engines were an extraordinary improvement on rummaging around blindly through the digital equivalent of several thousand filing cabinets. They allowed us to find useful information quickly, and the algorithms behind them became increasingly sophisticated at determining relevance, authority and popularity.

But every system which decides what to show you also decides, indirectly, what you are less likely to see.

That does not require malicious intent. It is simply a consequence of filtering.

The same problem becomes more interesting with generative AI because the machine is not merely helping us locate information. It is increasingly being asked to interpret and summarise it for us.

Imagine a local historical dispute for which ten surviving sources exist. One says the dispute began over land. Another says it was really a political disagreement. A third claims the land dispute was simply the excuse used to conceal something else. A fourth disputes that interpretation entirely. A fifth contains an eyewitness account, while a sixth argues that the eyewitness was unreliable. A seventh reproduces a contemporary newspaper report. An eighth contains a correction to that report. A ninth records an oral history from someone who was a child at the time, and the tenth is an increasingly furious forum post written by a man who has apparently spent thirty years being annoyed about the whole affair.

Ask an AI system what happened, and you may receive a perfectly respectable summary explaining that the dispute arose from competing interests concerning land and local governance and was eventually resolved through negotiation.

That answer may not technically be false.

It may simply be missing the interesting bit.

The machine has compressed disagreement into coherence.

That is where the idea of synthetic history becomes useful. I don’t mean history that has simply been invented by artificial intelligence. That is a different and considerably easier problem to identify. Synthetic history is more subtle. It is history reconstructed from the information that survives the filtering, ranking, indexing, summarising and repetition processes of modern information systems.

The danger is not necessarily that the machine lies.

The danger is that it becomes increasingly good at producing an answer that is broadly correct while quietly removing everything that made the original historical record complicated.

When information is repeatedly summarised, disagreement is particularly vulnerable. A source says one thing. A summary describes the general consensus. Another writer uses that summary. A search engine indexes the resulting article. An AI system encounters the repeated version and gives it greater weight because it appears consistently across multiple sources.

The process does not require anyone to sit down and decide that an inconvenient interpretation must disappear. The information environment simply becomes increasingly confident about the clean version.

And clean versions have a peculiar advantage.

They are easy to understand.

They are easy to repeat.

They fit neatly into a paragraph.

The awkward history, by contrast, requires caveats. It requires context. It requires someone to admit that the evidence is contradictory and that nobody can say with certainty what happened.

That is not particularly satisfying when the user has asked a machine a question and expects an answer before the kettle has boiled.

There is another layer to this, which is worth considering, because the sanitisation of history does not necessarily happen through factual omission alone. It can happen through language.

Institutions naturally develop their own vocabulary. Governments, corporations, universities, charities and public bodies all have preferred ways of describing complicated situations. Again, this does not require some sinister conspiracy. Bureaucracies simply develop language that makes difficult subjects easier to manage.

Over time, however, that language can begin to influence how events themselves are remembered.

A riot becomes civil unrest. A scandal becomes a controversial episode. A violent political dispute becomes a period of heightened tension. A campaign of intimidation becomes political pressure. A forced relocation becomes population movement. A man who spent twenty years behaving like a complete bastard becomes a controversial local figure.

History has been through the public relations department.

Nobody gets off lightly.

The problem is not that these phrases are always inaccurate. Sometimes they are perfectly reasonable descriptions. The problem comes when the language becomes so polished that it removes the human texture from what actually happened.

Real history contains stupidity, cruelty, courage, cowardice, revenge, loyalty, greed, generosity and astonishing acts of human kindness. It contains people making dreadful decisions for reasons that made sense to them at the time. It contains people doing admirable things for reasons that were not entirely admirable. It contains ordinary people caught in circumstances they never expected to experience.

It is rarely tidy.

Yet tidy is exactly what information systems tend to reward.

The eccentric source is particularly vulnerable here. A badly formatted personal website written in 2004 by somebody with no recognised institutional affiliation is unlikely to compete for visibility with a professionally produced article from an established organisation. A forum thread full of contradictory comments is harder to summarise than an official historical account. An oral history containing uncertainty and personal recollection is much less convenient than a paragraph which confidently explains what happened.

But sometimes the inconvenient source is the one that tells us where the interesting questions are.

The local historian may be wrong about the conclusion while being absolutely right that something deserves investigation. The eyewitness may have remembered the year incorrectly but preserved a detail that does not appear elsewhere. The angry forum poster may be hopelessly biased while nevertheless linking to an original document that everyone else has ignored.

Good history does not necessarily discard such material because it is imperfect.

It examines it.

That distinction becomes increasingly important when information is filtered before most people ever see it.

Consider what happens when an old local website remains online but stops appearing in search results. Nobody has deleted it. Nobody has banned it. The information is technically available.

But the practical difference between information that can be found and information that cannot be found is becoming rather thin.

The same thing can happen when a local newspaper archive moves behind a subscription wall, when a domain expires, when a forum shuts down, when photographs lose their captions or when the people who created the material die without anyone knowing what they had collected.

Digital information gives us an illusion of permanence because it is so easy to copy. In reality, much of the internet is remarkably fragile. Websites disappear. Domains lapse. Hosting bills go unpaid. Formats become obsolete. Links rot. Databases are abandoned. Social platforms change their rules and quietly take entire communities of information with them.

The digital age has not necessarily solved the problem of preserving history.

It has simply changed the ways in which history can disappear.

That brings us back to the memory hole.

The traditional memory hole is an obvious thing. A document is removed. A book is banned. A record is destroyed. Somebody makes a decision that a particular piece of information should no longer be available.

The modern version can be much quieter.

Nobody prevents you from reading the obscure source. It simply does not appear when you search.

Nobody deletes the alternative account. The summary simply does not mention it.

Nobody destroys the argument. The machine produces an answer that makes the argument seem irrelevant.

Nobody tells you that another version exists.

And if you do not know another version exists, you are unlikely to go looking for it.

That is what I mean by a silent memory hole.

It does not destroy information.

It separates information from attention.

That may ultimately be more powerful.

There is an important warning here, because it would be very easy to turn this into another comfortable story about artificial intelligence destroying truth, while humans once lived in some golden age of perfectly preserved historical accuracy.

They didn’t.

Humans have been editing history for as long as they have been recording it. Kings commissioned flattering accounts. Governments controlled archives. Victors wrote histories of wars. Families concealed embarrassing relatives. Institutions decided what deserved preservation. Newspapers chose which stories mattered. Communities forgot things they preferred not to discuss.

The difference now is scale.

One historian can influence a book. One newspaper can influence a generation. One government can control an archive.

An information system used by hundreds of millions of people can influence what an enormous number of people encounter when they ask apparently simple questions about the past.

That changes the stakes.

The real danger may therefore not be synthetic history, which is completely false. False information can be challenged if someone notices it.

The more difficult problem is synthetic consensus.

A version of history that contains the major facts, gets the names and dates broadly right and does not appear to contain anything obviously fabricated, but which has gradually lost the arguments, contradictions, uncomfortable memories and disputed interpretations contained in the original evidence.

It may even be more accurate in a narrow factual sense.

It can nevertheless be less truthful in the sense that matters to human memory.

Because truth is not always a list of agreed facts.

Sometimes the truth is that people remember the same event differently.

Sometimes the truth is that the evidence is contradictory.

Sometimes, the most honest historical sentence is simply that nobody really knows what happened.

That is an uncomfortable sentence for both humans and machines.

We prefer answers.

Perhaps that is why the obscure and the eccentric matter so much. They make certainty difficult. They introduce friction into the historical record. They force us to stop and ask whether the clean version is actually the complete one.

A healthy historical culture should probably contain more arguments than conclusions.

Instead, we are building systems designed to give us conclusions.

That is enormously useful when I want to know how to change the battery in something, why my washing machine is making a noise or whether a particular ingredient can be substituted in a recipe. I am perfectly happy for a machine to save me half an hour of searching when the subject is plumbing.

History is different.

When we ask what happened, we are not merely asking for information. We are asking someone, or increasingly something, to decide which pieces of information matter.

That decision deserves more scrutiny than we tend to give it.

Perhaps the great memory-hole of the twenty-first century will not be a locked archive, a banned book or a government censor sitting in an office with a red pen. Perhaps it will be something far more mundane… a perfectly functioning search engine, an efficient archive, a helpful summary and an AI assistant that confidently explains the past without ever quite telling us what has been left out.

The archive will still exist. The photographs will still exist. The letters will still exist. Somewhere, somebody will still have the original newspaper clipping tucked inside a cardboard folder.

The information will technically remain available.

But fewer people will know that it exists.

Eventually, the generation that remembers why it mattered will disappear, and the machine will be asked what happened.

It will provide an answer.

The answer will probably be plausible. It may even be broadly correct.

And perhaps nobody will notice that the history has become strangely tidy.

That is the part that concerns me.

Not that machines will necessarily invent our history, but that they may become extraordinarily good at compressing it. Not that inconvenient information will always be deleted, but that it may gradually become invisible beneath layers of ranking, summarisation, repetition and institutional language.

The old information will still be somewhere in the attic.

We simply won’t know which cupboard to open.

And perhaps that is the real lesson here. We should not only preserve archives. We should preserve disagreement. We should preserve the eccentric sources, the oral histories, the conflicting accounts, the badly written blogs and the people who insist that everyone else has got the story wrong.

Not because they are necessarily right.

Because they might be.

And because once the argument itself disappears, we lose something that no subsequent summary can reconstruct.

A memory does not necessarily have to be destroyed to disappear.

Sometimes it only needs to become inconvenient to find.

Until Next Time

Dominus Owen Markham