Own voices and owning a voice — Anne Rooney

 

AI's version of a Triassic scene. That none of these is
Triassic is the least of its problems. AI deosn't know it's wrong.
 It doesn't know not to mix images of bones and of fleshed-up
animals. And what is the skeleton with a hair-do?

Do you remember the #OwnVoices movement a few years ago? It was based on the idea that there is something uniquely distinct about being, say, Black, or deaf, or gay that is inaccessible to writers who are not Black, deaf or gay. Whether or not someone is well equipped to write from these viewpoints, we are all equipped to read writing from them. Just as we can read the plays of Sophocles and recognise the human experiences at their core, despite some surface alienness, because we are all in this together. 'This' being the messiness of embodied life.

Whoever you are, you share so much experience with people in all other times, places and conditions. A headache is the same. Getting wet in the rain is the same. Dreading a bad thing happening or eagerly anticipating a happy event are not dependent on who or where or when you are. There is a core of human experience that we can draw on, confident that readers will instantly know what it's like, and what we mean. And this brings us to AI (again). 

A large language model (LLM), like ChatGPT or Claude, doesn't share a bank of experiences with us. It doesn't have any experiences at all. It can cobble together some words that describe being wet in the rain, but it has no knowledge or understanding of what it feels like, or how it can feel different in different circumstances — how you might rush outside to dance in a warm summer thunderstorm but be frozen and wretched in a winter downpour. It understands nothing, feels nothing. It can pick out the patterns of words writers have used to describe being wet but it has never been wet, nor can it ever be wet. An LLM might produce coherent text, but it never produces wise or insightful text because there is no experience behind the words. Nor are the words chosen with awareness of their connotations, their cultural weight, or even their sound when read aloud. 

The ultimate 'own voice' is human. Whether it is Black, white, Brown, deaf, blind, sick, neurodivergent, gay, trans, working class, privately educated, orphaned.... it is human. I might not be able to write convincingly as though I had grown up immensely wealthy, or in sub-Saharan Africa, or without hands. But I can imagine something of it, which an LLM can't. I share enough humanity with people in different circumstances that I have some insight, even if it's only enough to say I can't imagine what it's like. An AI can trawl its training for how people have expressed those conditions, but it has no body that feels and no mind that thinks. Pain and injustice and delight are equally alien to it. And that is why I don't want to read a book produced with AI. 

Our own voices are human voices, and even if they are sometimes inarticulate or stumbling, they hold meaning and experience and shared humanity. They are not empty tokens. (Funny how access to AI is traded in tokens. Are they having yet another laugh at our expense there?)

And this all applies to pictures, too. Obviously. 

Anne Rooney

website; bluesky; insta 

Out now: Space Annual 2027, from Farshore/Science Museum, 2026