As AI becomes increasingly woven into our everyday lives and working practices, the way we talk about it matters too. In this guest post, Angela Bradbury from Trellis asks us to take a closer look at some of the language we’ve started to accept without question. It’s easy to talk about these tools as though they are independent actors, which ‘think’ and ‘say’ things. But, as Angela argues, that framing can distract us from something much more important: the people behind them. She advises shifting the conversation away from simply asking what AI is doing, towards a more revealing question: what are the humans doing?

Something curious has happened with the way people talk about AI.

Here are just some of the verbatim things I’ve heard people say in the last week:

“The doctors don’t know what’s wrong with me. But AI thinks it’s perimenopause.”

“I don’t get any recognition or feedback from my boss. So I’m asking AI instead, and it’s saying I’m doing a good job.”

“AI is the best therapist I’ve ever had.”

Every time I hear something like this, a part of me winces. Not at the person saying it, but at the norm this has become. But AI doesn’t “think” anything. AI doesn’t “say” anything. AI isn’t a person, or an entity. And I think that matters.

Let’s look at the last example: AI as therapist.

First of all, it’s notable that people are comfortable calling a piece of software a therapist (rather than a therapeutic tool). I could be wrong, but I suspect this kind of language has become much more commonplace than it was a few years ago. I’ll come back to this in a second.

Second, what kind of AI tool is this person using? If it’s ChatGPT, or Gemini, or Claude – these are tools that have been developed to engage with as wide a range of contexts as possible. They have been optimised to serve the needs of the companies that built them: to maximise user engagement and revenue.

Third, how is this person engaging with the tool? Are they including detailed instructions in the prompt as to therapeutic techniques and guardrails, or just launching in with a question then having a back-and-forth? Again, these tools have been built to maximise engagement, not for safeguarding or holding clear boundaries in-session.

I want to be clear that I do not judge people who talk about AI in this way. I’ve caught myself doing it from time to time – because that’s what happens with language. But I’d like us all to be a bit more conscious about how we talk about AI – and make tiny shifts where we can.

At Trellis, we’ve been building a tool that records SEND support meetings involving learners, their families, teachers and other professionals. The staff member using the tool gets an LLM-powered draft of a support plan and meeting minutes, based entirely on the recorded discussion. They’re guided to review, edit and finalise those documents.

Our team has been working on this for a year and a half, with a laser focus on the goal of creating a tool that does not hallucinate, creates appropriate output, guides the user to provide important information (and nothing else), protects the privacy of everyone involved, keeps the humans at the centre of this process, and much, much more. It is not a trivial task, and we’re not done. Having Trellis listen to your meeting and provide minutes is not the same as having Copilot do the same – and yet, I’m sure many would refer to both interchangeably as “AI”.

As we continue developing and using AI systems, we need to stop talking about them as if AI is a nebulous, single entity acting of its own accord. Language matters, because it can change our perception, even if only by a small amount every day. We can end up either anthropomorphising and demonising AI, or absolving those who design and deploy the tools from responsibility.

We need to keep talking about the human decisions that shape these tools: Who’s building them? What development and design choices are they making? Who’s using the tools, and who else is affected? What training and support are they getting? What are the power dynamics at play?

AI isn’t anyone’s therapist. People are using AI-powered tools to meet their emotional needs – tools that have been designed by collectives of humans to maximise engagement and revenue.

So please, the next time you hear someone talking about what AI is doing, ask instead: “what are the humans doing?”

If you’d like to carry on the conversation with Angela, you can connect with her on LinkedIn, find her on Slack, and learn more about Trellis on the website: https://www.trellis.scot/