Skip to main content
FIELD STATION· an action lab
DISPATCH· ENTRY 59

Just enough AI in Bristol

Notes from a workshop on smaller, less energy-hungry AI, run by Friends of the Earth and the University of Bristol, with the slides I presented.

Doug Belshaw · 16 Nov 2026 ·ai, society
# ai# society
Title slide reading 'Pick a task. Find the right AI.', Just Enough AI, Pervasive Media Studio, Dr Doug Belshaw, Bristol, 11 Nov 2026, beside a photograph of Watershed's brick frontage
FIG 1 The title slide of Pick a task. Find the right AI. Photo of Watershed: Simon Cobb, CC0, Wikimedia Commons

Last Wednesday I was at the Pervasive Media Studio in Bristol for a workshop called Just Enough AI. Friends of the Earth and the University of Bristol's Centre for Sociodigital Futures co-hosted it, and Mary Stevens and Matt Dowse organised it.

"Just enough" is Rachel Coldicutt's phrase. Rachel wasn't there, but her phrase gave the project its name: use AI that's the right size for the job. Most of the time, that means smaller models that use far less energy than the big ones.

Mary wrote an insight brief for everyone before the day. It covers what small and local models are, what they're good and bad at, and why so few charities and campaigning groups use them. The barriers it describes are mostly about confidence and skills. The brief cites Harnessing AI for Environmental Justice, the report Mary and I wrote with Laura Hilliger in early 2025. In it, we recommended moving from Big AI to local models.

Two decks

As ever, my first slide deck for the day was too conceptual. The organisers asked for something more practical, so I made a second one.

The part that got people talking was splitting an AI tool into three parts. There's the model, which is the engine. There's the harness, which is the app around it. And there's the context, which is what you give it. Once you see these as separate, the chatbot built into your office software becomes one choice among dozens.

The people

What I enjoyed most was meeting people who are thoughtfully exploring, or already using, smaller models in their own organisations.

I helped two people from Somerset Wildlife Trust think through their processes and where small models might help. The timing was good. FIELD STATION is starting Data in Practice, a three-year programme with the Royal Society of Wildlife Trusts, the network Somerset and the other trusts belong to.

A diagram that travelled

Reading Mary's brief, I drew a system dynamics diagram as a backup slide, in case anyone asked why everyone defaults to Big AI. It ended up in the event's introduction slides.

Causal loop diagram titled 'Why does everyone default to Big AI?'. At the centre is 'Use Big AI by default'. A convenience loop runs through 'Big Tech grows' and 'Built into everyday tools and cheap, for now'. A confidence loop runs through 'Hands-on experience' and 'Alternatives feel daunting', marked Bearing. A backlash loop runs from 'Big Tech grows' through 'Backlash grows' to 'Organisations try alternatives', marked SWARM, RELAY and RACK
FIG 2 Why does everyone default to Big AI? Drawn by FIELD STATION from Mary Stevens' insight brief. Solid arrows: more leads to more. Dashed arrows: more leads to less.

Two loops keep the default in place. The first is convenience: Big AI is built into the tools we already use, and it's cheap for now, so we use it more. The second is confidence: if you always use the default, you never get hands-on with anything else, so the alternatives keep feeling hard. A third loop pushes the other way. As Big Tech grows, so does the backlash, and some organisations start trying alternatives.

I've since developed a fuller version of the diagram and built a new deck around it: Why do people default to Big AI? I'm presenting it to an AI and sustainability network this week.

All in all, a worthwhile trip. Thanks to Mary and Matt for having me. If your organisation is working out what "just enough" looks like for you, get in touch.