From Description to Curation: Reframing Audio Description as Interpretive Ethics

(Not Just What You See, But Why It Matters)

Reframing Audio Description for the Age of AI

Let’s get the obvious out of the way:

Yes, AI can now generate flawless scene-level audio description. It can tell you who’s on screen, what they’re doing, and how many freckles they have. Great. Gold star, robot.

But the future of audio description is narrative power.

Here’s the shift: In a world where machines can describe everything, the real question becomes:

Who decides what gets described, and how?

Welcome to the era of meaning design. Your job is to say what’s there, sure, but also to help people understand why it matters.

Audio Description Is Never Neutral

AD has always been an interpretive act, even when pretending it’s objective:

Do you describe her dress as “tight” or “tailored”?

Do you bring the intention as playful, nuanced sarcasm, straight up description, or with a wink of knowing?

• Do you name the wheelchair, or normalize it into the character’s presence?
• Do you mention race? Emotion? Ambiguity?

These are cultural values wearing a headset.

Enter: The Social Model of Description

Borrowing from disability theory:

Are you describing the scene as an isolated object, or are you framing it within a cultural, embodied, and politicized context?

Medical model AD:

“A woman with cerebral palsy struggles to walk.”

Social model AD:

“A woman with cerebral palsy moves through a hallway without ramps.”

One describes deficit and the other describes barriers.

Guess which one reflects a future we actually want?
And how do AD performers bring that to life (with condescension? The “inspiration” with a tone of pity? Or with “of course, this is the way it is” confidence and normalization?

What Human AD Pros Still Do Better Than AI

Even the best AI (at this point) can’t:

• Negotiate context: When is silence more powerful than clarity?
• Read the room: When is naming emotion supportive, and when is it intrusive?
• Hold space for ambiguity: Not every moment wants to be pinned down.
• Align with the audience’s frame: A queer, disabled, neurodivergent viewer may interpret a moment differently (and rightly so).

As machines describe, you curate meaning.

The Next Frontier: Curatorial Ethics

When AI can do the “what,” your value lies in the “how.”

That’s where curatorial ethics enters:

• Who gets centered in the description?
• What ideological frames are at work?
• Is the description merely functional, or does it invite identification and agency?

Human AD performers and AD writers must move upstream into framing, consultation, and cultural mediation. Out of the transcription booth, into the strategy room.

TL;DR

Move away from “Can AI describe the scene?”

And ask “Whose voice decides what matters in that description?“

And if the answer isn’t rooted in context, culture, and care, then we’re just … describing furniture. That’s not doing access.

---

PS some further thoughts:

AI Capacity and Prompt Engineering

With the right prompt and a short enough runtime, today’s models can generate surprisingly competent AD, sometimes even matching baseline human output, especially in well-lit, single-narrative contexts like trailers. I don’t dispute the capability ceiling is rising fast.

Where I’m pushing isn’t on whether LLMs can technically perform description, but whether they can meaningfully contextualize it, especially in emotionally charged, identity-laden, or narratively ambiguous spaces.

Prompting can encode ethics, but only if the prompter already carries that interpretive schema (most don’t, at least for right now that I’m aware of). And when models are fine-tuned, they often optimize for fluency or compliance, not ethical ambiguity.

Customization and prompt-crafting can go far. But that’s my argument, not a rebuttal. It just shifts the critical question from “Can AI do AD?” to “Who gets to shape what ‘good’ AD looks like in an AI-first world?”

That’s a human-level, ethics-driven curation problem, not a computational one. And it’s why I frame this work as signs-and-symbols-as-service; the problem isn’t technical, it’s interpretive gatekeeping.

Gait and the Medical/Social Model

My example wasn’t perfectly parallel. The second version focuses on architectural barriers, while the first conveys motion, but also embeds “struggle” as a subjective interpretation, which invokes ableist framing unless carefully contextualized.

Whether to include a character’s gait is context-sensitive … and that’s the point. The problem with “struggles to walk” is what it implies when left unexamined. What I’m advocating isn’t a blanket rule, but a curatorial awareness: what lens are we describing through? Who gets centered in the frame?

Rather than claiming the social model must always omit gait, I’d reframe the contrast as:
Medical model assumes deficit as salient.
Social model asks why the movement is salient, and to whom.
This reframing is critical, especially when the viewer isn’t part of the dominant norm, and thus may perceive “struggle” as a biased projection rather than narrative fact.

Of course we need to describe gait when it matters. But how we frame that gait (struggle, adaptation, grace, resistance) is the ethical pivot I’m highlighting.

Finally — oof on me! I goofed with some formatting issues with this original post. Nobody wants to read raw markdown when they’re expecting a finished piece. I’m a work in progress here, and thankful for your patience and understanding.

---