We are primarily a photo library working with real actors with learning disabilities, who are paid for their work. We travel to meet groups up and down the country, work with people's own ideas, and photograph them. Those photographs form the main part of our library.

None of that has stopped or been reduced because of AI. We're not looking to replace the people we've built relationships with over nearly 25 years, or to use AI to cut costs by stopping that work.

We do use AI, and I want to be open about where it fits.

A lot of the images in Photosymbols have always needed things added to them. Backgrounds, objects, scenery. Previously, we might have drawn those things, made them using 3D modelling, or combined several photographs.

AI gives us another way to make those supporting elements. Sometimes it gives us a much clearer result.

Take a building in the background of an image. Photographing a fairly neutral building isn't always straightforward. The sun might be in the wrong place. There might be reflections of trees in the windows, parked cars or other things that distract from what you're trying to show. Generating that background gives us more control over what's there, so we can keep the image clear.

Our symbol designer, Kat, assembles these images by hand. She makes careful decisions about what belongs in the picture and how it should look.

We also use AI to generate whole images where there are gaps in the library, particularly where people need information now and the photographs would be difficult to organise.

Palliative care is one example. A full photography project about palliative care is sensitive work that takes time to do properly. We knew people were already trying to make information about it, so we generated a set of images showing the environments, the support someone might receive and the staff they might meet.

Those images are clearly labelled as AI and sit in their own tab. They're an addition to the library. They don't replace our work with real people.

Some of our AI images also come from needs that customers identify themselves.

Our Easy Read Bundle includes PictureMaker, an app we've built for people who can't find the image they need in the library. Sometimes that need is very specific. It might be someone holding a report with particular words on the cover, a plumber fixing a copper pipe under a sink, or a series of pictures showing how to make a fruit cake.

These are people making real Easy Read information and coming up against a gap.

Our staff review images generated through the app. Where we think an image would be useful to other customers, we can approve it for the separate AI section of the library.

The same happens in EasyMaker, our Easy Read document creation app. EasyMaker automatically chooses Photosymbols images featuring real people. Users can change those choices and, if they still can't find what they need, generate an image. Those generated images also go through staff review before any are added to the shared library.

So the images aren't just appearing in the library because an AI system has churned them out. There's a need behind them and a person deciding whether they're useful. We've also worked carefully on how the images are generated, so they're representative and fit the Photosymbols style. They should sit comfortably alongside the other images in an Easy Read document.

Across all of this work, we're guided by five expert advisers who have learning disabilities, are autistic or disabled: Michaela, Hayley, Grace, Matt and Shaun. They each bring their own experience and expertise. Whether we're working with photographs, AI elements or fully generated images, they advise us on what images should show, the colours we use, whether a scene makes sense and whether anything is confusing.

They're paid every month for their work as consultants. No vouchers or gift cards. Real pay for real work. Their judgement is an important part of how we make Photosymbols.

We also want customers to understand how an image was made and have a choice about using AI-generated images.

We've already written a blog about how we label our image using badges. They show whether an image is AI-generated, AI-assisted or includes real people. Where there's something specific to explain, Kat can also add a creator's note about how it was made.

Opting out

If you don't want to see fully AI-generated images, you can switch them off on your account page. From November, we'll also ask about your preferences when you take out or renew a subscription, so you can choose whether those images are available as part of your organisation's subscription. If fully AI-generated images don't suit your organisation's policy, we can switch them off for everyone on your subscription.

AI-assisted images remain part of the main library. These are images assembled by our symbol designer using AI elements, usually alongside photographs of real people, with advice from our expert advisers. They haven't simply come straight from a prompt.

We know some people are uncomfortable with AI imagery. We want to be honest about our use of it and give people enough information to make their own decisions.

There's a wider question here, though, and it matters a lot to us.

We hear people say that AI images of people with learning disabilities take away opportunities for people to be photographed. I understand that concern. But our photography continues, and we're also working with some of our long-established actors to explore digitally created characters, with their consent and involvement.

We're helping people with learning disabilities learn how to use image generation, write prompts and decide what a character should look like. We're exploring how they can have control over those characters, how they look, and what they do.

People have different views about this. One thing people with learning disabilities have told us is that they don't like typical AI people because they look like Ken and Barbie dolls. Too perfect, unnatural, and not like the people they know.

That's a familiar problem. One of the reasons we started Photosymbols was that stock photography didn't reflect the people in the learning disability community. 

If we decide that AI images should only show non-disabled people, we’re creating that same exclusion again. People with learning disabilities would be missing from those images, and that decision would have been made for them. We need to ask people whether and how they want to be represented, and give them a say in how that happens.

There's something important about capturing ordinariness. Ordinary people have all kinds of body shapes, wear different styles of clothes and have wrinkles, spots and thinning hair. Everyone is unique. The learning disability community is very diverse. What do people with learning disabilities really look like? That question needs to include people with different syndromes and people with profound and multiple learning disabilities.

People with learning disabilities need to be involved in making these decisions. What should a character look like? Does it feel representative? Who is missing? What do people want to see?

We haven't found all the answers. This is new work, but our direction is the same as it has always been: making images that reflect the people who use Easy Read, and involving them in creating those images.

Moving with the times

For the past two years, Pete has been travelling around the country through our Newton Project, meeting self-advocacy groups to talk about Easy Read and AI. We're listening to how people want to receive information, what they think about the technology and how they want to be involved.

These questions are connected. If you don't have access to the technology, you have fewer options for how you receive information.

People with learning disabilities have been left behind by successive waves of technology. We saw it clearly during the pandemic, when people didn't have the tablets, smartphones or support they needed to keep in touch and find out what was happening.

We don't want that to happen again with AI. We want to help people understand it, have a say and take part if they choose to.

That choice applies to our actors too. When we do photo shoots, we ask whether people want to be part of AI images. We've built that into our consent process.

We believe we need to be involved in shaping what happens next, because it's now possible for one person to generate a huge collection of images without involving anybody with a learning disability. Nobody needs to be photographed, consulted or paid. And looking at the finished images, a customer or commissioner might find it increasingly difficult to tell whether any of that involvement happened.

That concerns me. The relationships, experience and paid involvement behind an image can be invisible in the finished picture, but they matter.

Photosymbols will continue photographing real actors with learning disabilities, paying people for their expertise and listening to what they tell us. That work remains central to how we make our images and how we're approaching AI.

Whether you're choosing an image library or commissioning Easy Read, I think there are three questions worth asking.