The Story Around the Text: Why AI Needs a Frame of Accountability

What if the most important part of a story is not what happens inside it, but who is standing outside, telling us how to interpret it?

A law can be drafted by a machine, debated by humans, and approved by a public institution. A film can contain several stories, yet the audience experiences them through an outer narrative that gives the inner stories meaning. These may appear to belong to unrelated worlds: one concerns legislation and artificial intelligence, the other concerns narrative structure. But both expose the same hidden problem:

Before people judge a message, they judge the frame that tells them what the message is, who is responsible for it, and why it deserves attention.

This is why the question “Was it written by AI?” is less simple than it sounds. It is also why a frame story is more than a clever storytelling technique. Both are systems for managing trust.

The Story Outside the Story

A frame story is an outer narrative that contains one or more inner narratives. A traveler recounts an adventure. A detective reconstructs a crime. An old person remembers a childhood. The events inside the frame may be dramatic, but the frame determines how the audience receives them.

Consider a ghost story told around a fire. The apparition is the inner story. The people gathered around the fire, the skeptical listener, the trembling witness, and the question of whether the storyteller is reliable form the frame. The same supernatural event feels different depending on whether it is presented as a confession, a joke, a police report, or a memory distorted by age.

The frame does not merely surround content. It assigns the content a status. It tells us whether to treat something as evidence, entertainment, testimony, propaganda, speculation, or truth.

Public institutions use frames in the same way. A proposal becomes a bill when it enters a legislative process. A set of words becomes a law when it passes through recognized procedures, receives institutional authorization, and becomes enforceable. The text matters, but the surrounding process matters too.

This helps explain the intuition behind the claim that it would be unfair to risk rejecting a public project simply because artificial intelligence helped write it. The concern is not that authorship is irrelevant. It is that authorship may be only one part of legitimacy. If the proposal is examined, debated, amended, and approved through accountable human institutions, its origin should not automatically determine its value.

The machine may have generated sentences. It did not, by that fact alone, create the public meaning of the law. That meaning emerged from the frame around the text: the representatives who introduced it, the citizens affected by it, the arguments made in response, and the procedure that gave it force.

A message does not become legitimate because of where its words originated. It becomes legitimate through the frame that makes its use accountable.

The Authorship Trap

Modern culture often treats authorship as a shortcut for trust. We ask who wrote something before asking what it says. This is usually sensible. A medical diagnosis from a qualified physician carries a different weight from a guess posted anonymously online. A budget written by an accountable public official is different from a forged document.

But authorship can become a trap when it replaces evaluation. “A human wrote it” is not the same as “it is accurate, wise, or fair.” Human beings produce brilliant constitutions and terrible laws, careful research and fabricated evidence, compassionate stories and manipulative propaganda. Biological origin is not a guarantee of intellectual quality.

Artificial intelligence makes this shortcut visible because it separates generation from responsibility. A system may produce a paragraph, but it cannot by itself represent constituents, stand before a court, explain a decision to an affected community, or accept moral and legal consequences. The central issue is therefore not simply whether a machine touched the text. The issue is whether a responsible person or institution can stand behind the result.

This distinction can be expressed through a simple model:

Generation asks: Who or what produced the words?

Verification asks: Who checked whether the words are accurate and coherent?

Deliberation asks: Who considered competing interests and possible harms?

Authorization asks: Who had the legitimate power to approve the result?

Accountability asks: Who must answer when the result causes damage?

These five stages are often collapsed into one word, authorship. That collapse was manageable when writing tools were passive and the writer was usually the obvious source of every sentence. It becomes dangerous when a machine can generate polished language at scale.

A law drafted with machine assistance may be legitimate if its verification, deliberation, authorization, and accountability remain visible and real. A law written entirely by a celebrated human may be illegitimate if it was rushed through without scrutiny, concealed from the public, or designed to evade responsibility.

The important question is not “Did a machine write this?” It is “What kind of frame surrounds this text?”

Frames Can Clarify, Distort, or Launder Responsibility

Frames are not automatically virtuous. They can create trust, but they can also manufacture it.

A documentary may frame a selective set of facts as an objective investigation. A political speech may frame a policy choice as an unavoidable necessity. A company may frame layoffs as “efficiency improvements,” shifting attention from human consequences to managerial language. In each case, the outer story influences how the inner facts are interpreted.

The same danger applies to artificial intelligence. Saying that AI “wrote” a law can create a frame of technological inevitability: the machine appears to be an independent political actor. Saying that AI was merely a “tool” can create the opposite frame, one that hides the extent to which human judgment was delegated.

Both descriptions may be misleading.

A better frame distinguishes between instrumental assistance and decision making. A calculator can help determine the cost of a public program, but it does not choose whether the program is just. A language model can propose clauses, summarize arguments, or identify contradictions, but it does not possess a constituency. It does not experience the consequences of a regulation. It cannot provide democratic consent.

This is why transparency matters. Transparency is not valuable because the public must worship or condemn the machine. It is valuable because people need to understand where judgment entered the process and where it may have been absent.

Imagine two identical bills. In the first case, an elected representative discloses that a language model generated an initial draft, legal experts checked every clause, affected communities reviewed the proposal, and the representative accepted responsibility for the final text. In the second case, officials quietly submit machine generated language, cannot explain its origin, overlook errors, and blame the system when challenged.

The words might be identical. The frames are not. The first creates a chain of responsibility. The second creates an accountability vacuum.

This offers a practical test for any AI assisted work:

If the outer process cannot answer who checked the work, who made the hard choices, and who bears the consequences, the frame is incomplete.

Why Institutions Are Narrative Machines

We often think of institutions as procedural machines. They receive proposals, apply rules, and produce decisions. But institutions are also narrative machines. They create frames that tell a population what an action means.

A court frames a dispute as a question of law. A university frames knowledge through curricula, credentials, and peer review. A newsroom frames events through selection, headlines, sourcing, and editorial standards. A legislature frames social conflict as a matter for collective decision.

The institution does not simply process content. It changes the category of the content. A private suggestion can become a public commitment. A scattered complaint can become a policy agenda. A paragraph can become a rule that reorganizes people’s lives.

This is why the question of whether artificial intelligence can help produce a law is ultimately a question about institutional boundaries. If the institution remains capable of examining, revising, and owning the result, machine assistance may expand its capacity. If the institution treats machine output as authoritative because it sounds fluent, the frame has been captured by the tool it was supposed to govern.

Fluency is particularly dangerous because it resembles competence. A polished sentence feels as though it has already passed through thought. But language models can produce the appearance of settled judgment before any genuine judgment has occurred.

The frame must therefore do more work in an age of generative systems. It must reveal not only the final text, but the path by which the text gained authority. A responsible process should make it possible to ask:

• What problem was the text intended to solve?
• What alternatives were considered?
• Which claims were independently checked?
• Which groups could be harmed?
• What did the machine contribute, and what did humans change?
• Who has the authority to revise or repeal the outcome?

These questions are not bureaucratic decoration. They are the institutional equivalent of a narrator whose reliability can be examined.

A New Principle: Judge the Frame, Then the Text

The usual debate about AI generated work is framed as a contest between human creativity and machine creativity. That debate is too narrow. It assumes that the central unit of value is the sentence, when in many consequential settings the central unit is the decision system around the sentence.

A useful way to evaluate any AI assisted artifact is to examine three layers.

1. The artifact

What does the text, image, code, or proposal actually contain? Is it accurate, clear, useful, and suited to its purpose?

2. The process

How was it produced? What prompts, data, reviews, revisions, and tests shaped it? Were the relevant experts and affected people involved?

3. The mandate

Why is this person or institution entitled to make the decision? Who authorized the work, and who can challenge it?

Most public controversy focuses on the first layer while pretending the other two are invisible. Critics may reject an artifact because a machine helped create it. Enthusiasts may accept it because the output is efficient. Both responses confuse production with legitimacy.

The deeper standard is traceable judgment. A trustworthy artifact should carry enough information about its process and mandate that another person can understand how it came to be and where responsibility rests.

This principle applies far beyond law. A company using AI to screen job candidates should be able to explain how its criteria were chosen, tested, and appealed. A hospital using AI to recommend treatment should preserve a clear line of clinical responsibility. A student using AI to develop an essay should be able to distinguish borrowed language from personal reasoning. A filmmaker using generated material should decide whether the audience is meant to see the work as personal expression, collaborative construction, or deliberate experiment.

The frame is not an apology attached after the fact. It is part of the work itself.

Key Takeaways

• Separate generation from responsibility. Ask who produced the words, who verified them, who made the consequential choices, and who must answer for the result.
• Make the process visible. When using AI in consequential work, document its role, the human revisions, the sources checked, and the points where judgment was exercised.
• Do not confuse fluency with validity. Polished language can conceal factual errors, weak reasoning, or missing perspectives. Evaluate claims independently.
• Design an accountability chain. Every AI assisted decision should have a named human or institution with the authority and obligation to explain, correct, or reverse it.
• Treat the frame as part of the artifact. The surrounding context determines whether people encounter a result as evidence, advice, art, policy, or manipulation.

The arrival of generative AI does not eliminate authorship. It forces us to stop treating authorship as a single moment in which one person creates one object. Creation is becoming distributed across prompts, models, editors, institutions, audiences, and procedures.

That distribution can weaken responsibility if no one owns the outcome. But it can also improve responsibility if the process is designed with explicit checkpoints, transparent roles, and meaningful review.

A frame story teaches us that the outer narrative changes the meaning of the inner one. The politics of artificial intelligence teaches us that the origin of language is not enough to determine whether language deserves authority. Together, they suggest a more demanding definition of trust.

The future will not be decided by whether machines can write our stories. It will be decided by whether humans build frames capable of questioning, governing, and taking responsibility for what machines write.

The most important text may still be the one on the page. But in an age of machine generated language, the most important story may be the one that explains how that text earned the right to matter.