Made by humans, for humans

Dear Debriefers, 

I’m Peter, and I wrote this newsletter.

The Debrief is made by me, working with writers and illustrators from around the world to do reporting and tell stories through our own words and art. Our lived experiences shape how we see the world, how we write, and what we write about.

We don’t use Artificial Intelligence (AI) tools to write these articles, or to make the illustrations. We work hard – draft, redraft, edit, and draft again – to find authentic and creative ways to share them. 

But we’re not totally AI-free either. There are some points we use these tools, especially in translation and transcription. In this article I go through each area of the Debrief’s work – writing, illustrating, collaboration, making a media platform – and share how we approach AI in each.

The Debrief values independence, lived experience, accuracy, originality, and community. These require care and cultivation, and for the Debrief I believe that further use of current AI technologies would go against these principles.

I think the wider disability community needs to be cautious too. Sharing our lived experience is an essential part of our advocacy, and if we undermine it with reckless use of AI tools, we undermine our mantra of “nothing about us without us”.

About this edition

If you value human-made media and are able to contribute, please join those supporting the Debrief. Thanks to Allison, Erica and Louise for new contributions.

Peter Torres Fremlin is editor of Disability Debrief and is from the UK.

“This should be published”

When I was fifteen, I wrote a piece for school boldly titled “Being Disabled”. My English teacher, Mrs. Jarvis, wrote on it “This should be published”. Her feedback gave made me believe that I could write, and that it might be important. That confidence lasted years.

I went on to study English Literature at Cambridge university. Each week my essays were taken apart by teachers. When I later started work in international development, a manager sat me down and took out all the semi-colons I’d used in a report. Bureaucratese was a different style than a university essay.

I spent over a decade living abroad, most of which I was trying to learn one language or another.  When I moved back to the UK in the pandemic, I said that I was now going to “learn English”.

It wasn’t exactly a joke. Since then I’ve had several writing coaches and the Debrief itself is a testament to how my writing has changed through those years, especially evident in my poetry.

Writing a newsletter

Most of my work is done at my desk, or lying down in bed. 

I spend some days (or weeks) researching, reading, or calls, and then structuring notes. Ideas emerge slowly, or at the end of the day after I’ve stopped work.

When it’s time to write, I open a blank page and stare at it for a while. I write, browse social media, write some more. When that doesn’t work I turn on some TV show in the background. When I get into the flow, I write more fluidly, especially if the weekly deadline is looming. 

If there’s time, I send the draft to Celestine Fraser to review it. She tells me she likes it, she doesn’t have any many comments at all, and then I see ten minutes of voice-notes going through line-by-line with what would make the piece better.

I revise, proof-read, check the links, and then, nearly always with the feeling I must have missed something, I press send.

How I use AI when I write

There are some areas where I rely on AI tools, especially in translation and transcription.

I regularly use machine translation to navigate sources from different languages. If I quote something I check against my own knowledge of the language or by asking a native speaker. I prompt AI chatbots for alternative translations.

Automatic transcription is very helpful. For more serious things, I use rev, which combines machine with human transcription. And if I quote something, I check it against the audio.

While writing, I make very occasional and minor use of AI. For example, I recently asked ChatGPT for “space-related phrases beginning with P”, so I could make a title where all the words began with “P”. (No luck.)

I’ve also tried ChatGPT as the last proof-read in the moments before pressing send. But so far I’ve found it to be inconsistent, and not a replacement for going through it myself. And, after all, the time-old method for finding the last typo is to publish.

Picking news highlights by hand

I've basically followed the same method for curating news as I did for the Debrief's first edition in 2020.

Through the month I save disability-related links from browsing, google alerts, other newsletters and what readers send me.

Then I sit down, and over few days, go through hundreds of links. Many of them I can close quickly, some I skim, others I spend more time with. At the start it’s quite entertaining, but by the end my mind is overwhelmed, and I’m just clicking through.

I still do it myself because seeing the wide scope of the news shows me the trends. And curation comes from my intuition – shaped by over fifteen years working on disability issues – to pick out things that are interesting or I haven’t seen before.

And it’s from this informed view that I then make editorial decisions about what to commission on the Debrief.

Quoting from AI-generated material

One of the ways that AI has entered Debrief newsletters is through sources that are AI-generated material.

The Debrief uses a wide range of sources – reports, news articles, comments from readers – and now an increasing amount of such material has used AI in its generation.

Some of these I avoid, as excessive AI can appear inauthentic or boring (qualities which make me less likely to quote it, who or whatever wrote it). But some people use it also to make interesting or newsworthy points, which then make sense to quote. 

And either way, there are many times I can’t tell the difference. AI-generated writing has many of the qualities previously associated with good human writing. And humans are now in turn imitating AI writing in their style. Not to mention the many cases where there’s a bit of both.

Working with others

The Debrief has published articles by over twenty different writers. They range from reporting on floods in Brazil, insights from a special education teacher in Tanzania, to an interview with an activist in Myanmar.

Each article comes out of very close work together, over a series of zoom calls, and several rounds of editing drafts. It’s usually months and sometimes well over a year from an initial conversation to the published article.

These organic collaborations have become one of my favourite parts of the project. It’s an honour to think and laugh together. And often the work moves us to tears.

Collaborations based on mutual trust

Working together in this way counts on a good deal of mutual trust. And while I do sometimes work with old friends and colleagues, many contributors are people I only know through the Debrief itself.

I work to create a safe space where writers can explore sensitive and often painful subjects. And I trust that they are who they say they are, and that what they share is accurate to the best of their knowledge.

I look at someone’s previous writing, count on a mutual reference, or see how they use social media. And then, mostly, I rely on my intuition in the way we build a relationship through conversation and discussing the initial pitch of an article.

For over a year now, I’ve asked each contributor how and if they use AI in their writing. If they use it for help with grammar, I remind them that that is my job as editor. One of the joys I find in editing is to work with imperfect, messy drafts to help bring out a writer’s unique style.

This process is not fool-proof. Maybe not everyone is totally transparent about how they use AI. As I would with fact-checking and sourcing, I do some due-diligence. And, as with other areas, if I find out something wasn’t right, I’ll share a correction.

Using machine translation on drafts written in other languages

Several Debrief articles have been drafted in languages other than English. Articles on women with disabilities in Yemen, austerity in Argentina, or black people with disabilities in Brazil started, respectively, in Arabic, Spanish, and Portuguese.

In each case, the first translation was done using machine translation. I personally used Google translate for the Arabic and Portuguese; Eduardo Quiroga used Deepl.com for the Spanish. These tools were credited in the Acknowledgement section of each article.

And in each case, the articles and their translations were still checked and edited by me. So far I’ve only used this process in languages where I have a reading ability. (Like I said, I had many years of language learning before I started the Debrief.)

Illustration and photography

The Debrief commissions illustrations from disabled artists, and publishes photographs of disabled people. 

As with writers, I work carefully with illustrators, and these relationships are nurtured through several years. Artists’ own lived experience and artistic visions leads to gorgeous pieces that respond to the article material with soul and depth

As with my own writing, you can see, for example, how Kinanty Andini’s art has developed through her years of illustrating on the Debrief. For a sneak peak into how Kinanty does her work, see this time-lapse video of how one drawing was made.

As for the photographs we publish, many of these are editorial photos licensed from Getty Images, where it’s clear where they were taken and who took them. They’re selected by Celestine, who trawls through hundreds of photos from the last month. Her filmmaker eye is sensitive to selecting one that will challenge our expectations and show disability in new ways.

Making an independent media platform

Behind the scenes, I am also a simple person, carrying on in my pre-AI ways. I go to meetings and take my own notes (although I do use the auto-transcription for reference afterwards). 

I write my own social media posts, and reply to emails myself. Relationships are one of the most important things I can cultivate, and outsourcing them would be absurd. And, indeed, when I don’t reply to messages, that is also done by me, a human, feeling slightly guilty. 

The joy and challenge of starting something new is the range of things one has to be across: fundraising, admin, contracts, or legal issues. I might ask ChatGPT for advice in an area I don’t know.

One area that ChatGPT did help me with was in coding some Python scripts a couple of years ago. These scripts upload the database of curated links to the website, and getting AI advice was much easier for me than Googling, confused.

More regular work on the website is done by Cathy Sarisky. She's gradually using generative AI more in her coding, but it still needs a lot of careful review from her. 

As for the fundraising, so far I haven’t used AI in writing proposals for potential donors, but then again, I also haven’t been asked to do a twenty page application.

We can only do this because you support us

I can spend days going through disability links, writing newsletters, or editing other people’s work, because this is my job. And I can commission the work from writers and illustrators because I have the funding to do so.

If it weren’t my job, maybe it would make sense for me to use AI to highlight the latest disability news. Or if I didn’t have the time to edit and mentor someone to develop their writing, then maybe an AI tool could provide some assistance.

The Debrief position on avoiding AI in its writing and illustration isn’t meant as a judgement on how others are or aren’t using AI in their work. Work in other areas has different constraints and opportunities.

The Debrief shares the lived experience of disabled people, and finds creative new ways to tell disability stories, thanks to the readers and organizations who support this work.

The rising tides of AI-slop

The care we can take with our work seems to be going against prevailing currents, of how journalism and independent thought are being undermined. Seeing how quickly AI writing is spreading often makes me despair. It can be hard to keep faith in the hard work and where it gets us.

There is a lot of AI-slop in my inbox, on social media, and in online writing (increasingly offline writing too). It’s dispiriting to see how our language can be hijacked, and I hate the way it makes me read texts wondering who or what wrote it.

Indeed the question about whether a text is AI-generated has added another layer of distance between us. I get that it might be difficult to acknowledge, but it is sad, and adds to the alienation, when people are not forthcoming about the fact they used AI to write something.

Technology and independence

There are further ethical issues around the use of AI, including questions of intellectual property, and the social and environmental damage these technologies are causing. These are covered elsewhere.

One overall concern I do have is that of independence. Outsourcing our ideas to products owned by our technological overlords would abandon the independent view that the Debrief has fought hard to establish. I publish the Debrief on Ghost, an open source platform, rather than venture-capital funded platforms whose interests will ultimately diverge from those of their users.

Assistive technology and Authenticity

From a disability view, there are a few further issues worth touching on. 

AI can be an assistive tool. It can generate captions or describe images. Some neurodivergent people say that it helps organize their thoughts.

And tools that help more people write fluently are especially useful to, say, those with learning difficulties or people writing in a second language. Not to mention that work and life are draining, especially for disabled people, and someone might not have the energy left over to craft a post on social media.

But there are also vital questions of voice and authenticity. What does “nothing about us without us” mean if our posts are written by machines? AI tools can generate posts about lived experience whether or not the experience was, in fact, lived.

Disability influencers are among the adopters of these tools. Of course not everyone posting online was honest in pre-AI days either, but it is now exponentially easier to make false, fabricated or even exploitative stories.

Perhaps there are ways to use AI tools to generate writing in more authentic ways. For example, I imagine someone could work with AI with the same thoroughness that one might work with a ghost writer.

We need to be careful, as individuals and as a community. Establishing trust takes many years. And inauthentic use of AI can break that quickly.

Doing things our way

The Debrief is made in community, through human relationships with readers, writers and artists spread across the world. The reason we do this work is to make a place on the internet where we can do things our way, and tell our own stories in our own words.

Our disability lens comes from combining our lived experiences of disability with reporting, analysis and creative writing. We are working hard to tell disability stories in original ways. Unquestioning adoption of AI tools would corrupt these goals, not further them. 

Maybe the boundaries I set with AI will be stricter or looser in the future. Either way I’ll let you know. I write this piece in the interest of transparency, and also to know how you readers see this rapidly changing technology.

I have spent decades learning how to write. I don’t intend to stop now. It is the most fulfilling work of my life to create a newsletter where we can exchange and learn together, and where I can share my skills with others, as I in turn learn from them.

I'm sitting at my desk as I press send. It's exciting and vulnerable to share how I work and what I believe. I wonder what I’ve missed, and feel a thrill in my stomach.

Peter 

Outro

Help us do this work. The Debrief is published through a pay-what-you-can model.

For more on disability and AI, see the Debrief library of curated links on AI, or Ria Andriani's essay on how AI opens up visual access but closes down social change.

Let your friends know. Sharing the newsletter is how people find it!

Connect. Get in touch. You can find me on Linkedin and Bluesky.

Acknowledgements

Thanks to Catrina and Damien Demolder for the photograph and being such fun to work with!

Thanks to Celestine Fraser for helping conceptualise and thoroughly revising this piece. And to Anne, Kinanty and Sonaksha for having a look at previous drafts.

And of course to the readers and organisations whose support lets us do this work.