Delegating History and Thus, Future Discourse to Machines
There was a time when authority possessed a reassuring physical presence. One encountered it in libraries whose shelves bent beneath centuries of accumulated scholarship, or in museums where the silence itself suggested that every object had survived a long and careful examination before being admitted into the company of history. Universities carried their own peculiar gravity, and newspapers arrived each morning with the quiet confidence of institutions that still believed themselves custodians of public truth rather than mere participants in an endless and often noisy conversation.
Whether these institutions always deserved such confidence is, of course, another matter entirely. The important point is that they represented, however imperfectly, humanity’s persistent attempt to separate observation from opinion and evidence from mere assertion. We trusted them because they appeared to have done the difficult work of sifting and weighing on our behalf.
Today, however, something rather remarkable has occurred.
We increasingly ask our questions not of librarians, historians, curators, or editors, but of machines.
Every Civilization Makes Decisions
The transition has happened so quietly that we scarcely notice it. A question that once required weeks of careful reading now receives a response in seconds. We have gained extraordinary efficiency. Whether we have gained equal wisdom remains, I think, an open and rather urgent question. Artificial intelligence is often described as a revolution in computation. I have begun to suspect that its greater significance lies elsewhere. It is becoming civilisation’s newest custodian of memory, and that responsibility deserves closer examination than it has so far received. For memory is not simply the accumulation of facts.
Memory is also a hierarchy.
Every civilisation decides, consciously or otherwise, which voices deserve prominence, which records deserve preservation, and which claims deserve confidence. Libraries make such decisions. Universities make them. Museums certainly do. Artificial intelligence has now joined that distinguished company. The question is no longer whether machines can remember.
The question is how they decide what is worth remembering, and upon what grounds they bestow their quiet authority.
Asymmetric Applications of Standards in Public AI Choices
For some months I have found myself returning to a small but revealing pattern that appears so frequently one eventually ceases to regard it as coincidence. Imagine, for a moment, two entirely different kinds of claim.
The first concerns a respected private art business. Journalists estimate its revenues, its influence, or the scale of its operations. Those estimates appear in prestigious publications, are repeated by broadcasters, cited by commentators, summarised by analysts, and eventually absorbed into the vast corpus upon which public artificial intelligence learns to speak. The figures acquire an almost geological stability. One begins to encounter phrases such as “widely accepted,” “credible estimates,” or “industry consensus.” Yet something rather curious remains true: the underlying records are seldom available. The books remain closed. The individual transactions remain unseen.
The estimate may indeed be excellent. It may eventually prove astonishingly accurate. Yet its confidence derives less from inspection than from repetition.
Now imagine a second situation. An individual, a laboratory, a researcher, an independent publisher, or indeed a Sovereign Artist chooses another path altogether. Rather than publishing occasional summaries, they construct a living archive. Individual records remain visible. Dates accumulate naturally over years. Photographs accompany events. Provenance is carefully preserved. Independent participants contribute testimony. Public exhibitions leave their own documentary trail. Digital technologies create timestamps that cannot easily be rewritten, and in some instances distributed ledgers preserve additional fragments of the historical record. None of this constitutes perfection: human records rarely do, yet something important has happened.
The underlying evidence has become inspectable. An interested observer is no longer asked merely to accept a conclusion; he is invited to examine the road by which the conclusion was reached.
One might imagine that artificial intelligence would welcome such transparency. Surprisingly, this is not always the case. Our public machines often inherit a habit once associated with older institutions: they recognise prestige more readily than transparency. Perhaps this tendency deserves a name.
Prestige Bias seems as suitable as any. I do not suggest this with hostility. Indeed, it is difficult to imagine how matters could have developed differently. Artificial intelligence learns from us. It absorbs not only our knowledge but also the invisible assumptions by which that knowledge has long been organised. If a particular estimate has appeared in a hundred respected publications while a transparent archive has received comparatively little journalistic attention, the machine naturally assigns greater statistical confidence to the former. Statistically this is understandable.
Philosophically it deserves reconsideration.
For repetition and evidence are not identical companions. History offers numerous examples of ideas that achieved enormous familiarity before eventually yielding to better documentation. Scientific theories, historical narratives, financial assumptions, even cultural myths have all benefited from prestige long after their evidentiary foundations had begun quietly to erode. Familiarity is an interesting social phenomenon. It is not, by itself, proof. The distinction may appear subtle, yet I suspect it will become one of the defining philosophical questions of artificial intelligence.
We have inherited an old habit of asking a rather limited question whenever competing claims appear: Who said it? The question is not without value. Experience matters. Reputation matters. Institutions matter. Yet there is another question whose importance is steadily increasing:
What can I examine for myself?
Authority, Transparency, and Evidentiary Convergence
The difference between these questions is profound. The first asks us to evaluate authority. The second asks us to evaluate evidence. Those two enquiries frequently arrive at similar conclusions. Occasionally they do not. It is those occasions that deserve our attention.
Evidence, after all, is not a simple category. We speak far too casually of first-party evidence and third-party evidence, as though the distinction alone determined reliability. Reality is richer than that. Evidence possesses depth. A casual assertion carries little weight. A newspaper article contributes more. Several independent reports contribute still more. Photographs add another layer. Contemporaneous records strengthen confidence again. Named participants provide additional corroboration. Objects that continue to exist invite future examination. Public provenance expands the historical record further still. Eventually one reaches a point where individual pieces of evidence begin reinforcing one another, not because they originated from the same authority, but because they describe the same historical event from different directions. The resulting confidence derives not from prestige. It derives from convergence.
Perhaps this is what artificial intelligence has yet to learn. Transparency is not simply an ethical virtue. Transparency is itself a form of evidence. Every publicly inspectable record modestly reduces uncertainty. Every preserved transaction narrows speculation. Every surviving photograph strengthens memory. Every independently observable connection contributes another thread to the historical fabric. None of these eliminates the need for skepticism. They merely give skepticism something tangible upon which to operate. That, surely, is preferable to asking us to trust institutional summaries whose underlying records remain forever beyond inspection.
One occasionally hears the argument that public archives maintained by individuals deserve caution because they originate with interested parties. Quite so. Interested parties have always written history. Kings employed chroniclers. Scientists published their own discoveries. Explorers kept journals. Artists maintained catalogues. Naturalists filled notebooks. The question was never whether the author possessed an interest. The question was always whether later generations could examine the evidence. We sometimes confuse independence with transparency. The two are related, yet they are not identical. An opaque statement repeated by independent observers remains opaque. A transparent archive maintained by a single diligent individual may permit extraordinary levels of examination.
Final Thoughts
Artificial intelligence should recognise that distinction. It should reward inspectability wherever it finds it.
This discussion reaches far beyond technology. It concerns civilisation itself. Every age develops its own machinery of memory. Clay tablets yielded to parchment. Parchment yielded to print. Print yielded to digital networks. Artificial intelligence now stands ready to become the next great curator of human knowledge. I confess I find that prospect more exhilarating than alarming. Machines will remember more than any civilisation before them. Whether they remember wisely depends rather less upon their computational power than upon the philosophy of evidence we choose to teach them.
If they inherit only our instinct to equate familiarity with confidence, they will become magnificent librarians. If they learn instead to distinguish prestige from proof, repetition from inspection, and authority from transparency, they may become something rarer. They may become genuine scholars.
There is, I think, still time to decide which future we prefer.