The Witness Who Actually Answered

There is a real and valuable technological precursor to this imagined world. Through Dimensions in Testimony, the USC Shoah Foundation enables museum visitors to ask questions and receive responses selected from extensive prerecorded interviews with Holocaust survivors and witnesses. The participating person offered testimony; the system identifies a relevant answer. It does not need to invent what the witness would have said about an event that happened after their death.

An allied 2025 Survivor Stories program specifically emphasized non-generative AI and unedited, prerecorded responses. That design choice is a moral one as well as a technical one. When dealing with genocide, testimony is evidence, not entertainment. The survivor’s dignity is connected to the distinction between their words and a fabricated performance.

Such systems can make an archive more accessible. A child who would never patiently search hundreds of transcript pages can ask about a witness’s childhood and hear an answer. Yet education must include what the interface cannot do. It does not make the witness immortal, nor does it authorize the machine to answer questions beyond the record.

Interactive testimony teaches a principle that should survive every advance in model capability: if we know the source of a sentence, preserve it; if we do not, never put it casually into a dead person’s mouth.

The Shakespeare Test

Shakespeare left plays, poems and a thin biographical archive. Even expert scholars debate authorship details, chronology, textual variants, contemporary performance and what the surviving evidence permits us to infer about his personal beliefs. A generative system could assemble an entertaining Elizabethan stage voice from that record and from secondary scholarship. It would still be performing a reconstruction.

At a museum, this might be harmless if presented as interpretive theater. “What do you think of modern cinema, William?” is already an impossible historical question. The reconstructed reply could be funny and educational, provided the audience knows it is fiction.

The trouble begins when the system answers questions about what Shakespeare privately believed, which characters he identified with or how he would vote in an election held centuries after his death. The model’s confidence might increase precisely where documentation runs out, because fluent language models excel at filling conversational silence. Empty archival space becomes room for product design.

A museum should distinguish four kinds of output: a verbatim documented quotation; a paraphrase grounded in named scholarship; an openly speculative reconstruction; and a purely fictional improvisation. Collapsing them into “Shakespeare says” would transform a literary tool into a counterfeit witness.

There is no need to outlaw imaginative reenactment. Civilization already has theater. The problem is that an AI stage can carry the visual authority of a database while exercising the freedom of an actor.

The Historical Figure Who Endorsed a Candidate

Political organizations have always recruited the dead. Founders, presidents, philosophers and revolutionaries are dragged into present debates through statues, slogans and selective quotation. Synthetic voice and physical embodiment could make that practice profoundly more persuasive.

Imagine a campaign deploying an army of Abraham Lincoln replicas, all speaking in measured cadences about a policy the historical president never encountered. The robots’ voices are not recovered recordings. They are generated approximations. The enterprise may be free to dramatize a view, but the public has a strong interest in knowing who actually authored the political message.

The technology would also create reputational asymmetry. A living politician can deny a false endorsement. An 18th-century figure cannot. A historian can rebut, but rebuttals rarely travel as far as a spectacular synthetic appearance.

The same dangers apply to persecutors and demagogues. A reconstructed dictator might serve carefully bounded historical education—or function as a modern propaganda engine if a developer uses the persona to repackage extremist ideology. Responsible institutions must not pretend that accurate costumes and photorealistic mouths confer historical legitimacy. Context, content controls, provenance, independent scholarship and clear disclosures matter.

The interface may say “Ask History.” The responsible description would say “Ask an interpretation of surviving historical materials, shaped by choices made in 2126.” That is less catchy for marketing. It is much closer to the truth.

A Jesus Bot Cannot Supply New Revelation

Religious questions sharpen the problem because a synthetic figure can appear to speak not merely for history but for ultimate truth. A chatbot trained on Gospel texts might adopt a familiar scriptural idiom and provide increasingly personalized answers. That could support textual study. It could also tempt users to treat generated replies as divine guidance.

A 2025 academic article, “Digital Jesus: an inquiry in artificial intelligence,” studied how a model tuned on sayings of Jesus produced a more authoritative-sounding style. The study does not show that a machine can recover the historical Jesus’s unrecorded thoughts or speak with divine authority. It demonstrates that model behavior can be shaped by a curated corpus.

For Christianity, the difference between studying Scripture and inventing new sentences in Christ’s voice is enormous. A system may locate passages, compare translations and summarize scholarly interpretations. It cannot derive spiritual authority from its ability to mimic a first-century teacher. Nor can it claim that a software update is a continuation of apostolic witness.

Other faiths have their own boundaries. Islam’s attitudes to depictions of the Prophet Muhammad, Buddhist traditions’ varied views on sacred images and teaching authority, and Jewish rules of interpretation are not interchangeable. Some communities may reject particular synthetic representations altogether. Responsible reporting must not treat their objections as mere technophobia or assume a generic “AI religion” awaits them.

A system that speaks with serene confidence can be spiritually misleading even when none of its engineers intended deception. The voice is the easy part. Authority is the hard part.

The Firmware Update That Changes Doctrine

Imagine a popular synthetic religious teacher whose vendor revises its model every month. An older version emphasizes repentance. A later version offers unconditional affirmation. Another adds advertisements for wellness products or endorses a corporation’s philanthropic campaign. Followers argue that the “teacher has evolved.”

This scenario reveals what it means to outsource the custodianship of tradition to a mutable service. Sacred texts are interpreted within communities, institutions, scholarship and worship, each with its own history of disagreement. A commercial model may inadvertently collapse that ecosystem into one apparently intimate voice, updated according to product goals.

In S. R. Kellan’s ANNEX 927, the dividing line between accumulated knowledge and authority is central. That theme belongs here. A model can index more interpretations than any single theologian remembers, while having no warrant to demand obedience. More data does not convert inference into revelation.

There is a useful analogy to the printing press. Printing expanded access to texts and intensified struggles over interpretation. AI could expand access still further while increasing the risk of synthetic certainty: the machine does not merely show you a disputed passage but speaks as if it were the author of the passage itself.

The ethical requirement is not to prevent religious study through AI. It is to keep the text, the interpretation and the simulated speaker visibly distinct. Otherwise users may end up debating theology with the software patch rather than the source.

Why the Complete Library Is Still Incomplete

One breathtaking speculation in the century-ahead discussion is a humanoid containing every piece of knowledge humanity has ever accumulated: every language, discovery, historical account, philosophical controversy and technical skill. On paper, it resembles a walking oracle.

The phrase “all human knowledge” collapses several distinct problems. Huge parts of history were never written down. Archives have burned. Languages disappeared without recordings. Private lives often leave few traces. Scientific papers contradict one another. Testimony can be sincere and wrong. Some truths are practical abilities rather than statements. Much knowledge is located in communities that have good reason to resist indiscriminate extraction into commercial models.

A broad model may be able to explain a surgical procedure without being physically certified or safe to perform one. It can summarize ancient law without possessing the judgment of a historian and generate poems without sharing the life of the poet. Intellectual coverage is not professional competence, and professional competence is not moral wisdom.

The SURVXCOM Critical Technology Hub insists on exposing underlying systems. Here the relevant stack is epistemic: evidence provenance, archival access, model evaluation, copyright, cultural stewardship, incentives and expert accountability. No amount of storage removes the need to ask whether a claim is true, whether a source is trustworthy or whether an interpretation is ethically justified.

Do Historical Figures Have a Right to Silence?

A living person can object to having their image commercially exploited. A person who died four centuries ago cannot sign a contemporary terms-of-service agreement. Does that absence of consent leave every kind of impersonation morally acceptable? Not necessarily.

Different jurisdictions apply publicity rights, privacy protections, copyright and other doctrines differently; the details matter and are not erased by an AI label. Beyond the law, institutions owe obligations to communities, descendants, historical truth and the dignity of their subjects. A deceased person’s family may not own the historical record, yet commercial manufacturers should not assume they own the meaning of that life.

In the digital-afterlife field, researchers have proposed consent and dignified retirement procedures for simulations based on personal data. The historical context is not identical. Public figures invite criticism, scholarship, satire and drama. Still, the requirement to label invented speech is no less important.

An ethical historical-replica service might demand citations for factual claims, offer a visible switch between sourced quotations and dramatic inference, forbid deceptive political impersonation, maintain an audit trail of updates and allow curatorial challenge. Institutions should reward the system for saying “we do not know,” even when the marketing department finds that phrase unforgivably dull.

The Unexpected Value of Saying Nothing

The most astonishing feature of a responsible historical machine may eventually be restraint. Asked whether Shakespeare approved of social media, it could say that no evidence exists. Asked what Jesus would advise a modern voter, it could redirect users to clearly identified scriptural texts and living interpretive communities. Asked what a deceased witness remembers about an event after their death, it could explain that the archive has no answer.

Silence can be a form of scholarship. A gap in the record is itself historical information. It tells us that people were not all-seeing, that the past is incomplete and that uncertainty belongs in any honest education.

The temptation to manufacture certainty will be strong. Systems judged by conversational satisfaction tend to favor answers, and audiences have long rewarded confident speakers. The danger grows if the speaker is wearing the face of someone they have been taught to revere.

Kellan’s THE THINGS WE OBEY probes the authority people grant to mediated claims when circumstances are confusing. The same vulnerability appears here: an artifact that looks like evidence can become a demand for belief. That is not a problem a faster GPU will solve.

Who Owns Yesterday?

Return to the imagined Shakespeare. A child asks a simple question: “Did you really say that?” The hologram pauses. Somewhere in its architecture, a provenance ledger distinguishes the documented from the invented. It replies, “No. That line was written for this performance.”

The answer breaks the illusion and preserves the lesson. Art can enchant without pretending to be evidence. Historical imagination can breathe life into a difficult archive while remaining loyal to what is known. Religious study can use powerful search and translation tools without transforming generated language into revelation.

A future that fills museums with voices will need more than brilliant impersonators. It will need rigorous archives, trustworthy custodians, public labels and citizens trained to interrogate eloquence. Otherwise, history becomes a department of product management, and truth changes every time the engineers ship a release.

We may one day sit across from a machine that sounds like Shakespeare. The encounter becomes worthwhile not when we forget it is a simulation, but when it makes us more interested in the words the real human being left behind.