AI is everywhere. It feels like I can’t go a week without someone telling me I must use it: it will save me time; it will make my work easier, and my writing better; it will help me think through problems, and act as a free therapist; it will create posters and images and flyers in a flash; it will help me find information and learn faster; it will save me having to think so much or remember things. This is not particularly appealing: I like thinking and designing, and writing for myself. And what I see of AI is that it is ruining everything. It is destroying institutions, destabilising economies, erasing job opportunities, creating a permanent underclass, eroding societal trust, strip mining the planet, pushing some people into psychosis, and making us all more stupid and less able to think critically.
A recent experience illustrated the problem perfectly.
We moved into our house in June 2021. The pandemic meant we couldn’t get any work done immediately, so we lived an a perfectly functional but aesthetically unappealing house for some time. The kitchen made me sad. In 2024 we handed over vast amounts of money and invited in a team of people who quietly and efficiently gutted and remade the house. We lived peripatetically for several months, grateful for the generosity of friends with spare rooms. By the end of this time we were exhausted, and had decision fatigue. I was not aware that there were different types of window vent and I had no opinions on them, and yet I was made to decide. I was asked for my preferences on everything: door handles, light switch fittings and various off-white shades of grout. When we moved back into our own (now much improved) home, we couldn’t face any further decisions or any further disruption. And so another few years went by, our garden remaining a collection of slabs and pot plants, and an out of control blackberry bush. Until in 2026 we felt psychologically ready to once again invite people with mallets and angle grinders into our lives and into our space.
Finding a company to execute home improvements is the 10th circle of hell. Personal recommendations are ideal, but as is so often the case, this job was either too big or too small for those recommended by neighbours and colleagues. And so I looked online and found a company who appeared competent, were local and who had experience in gardens of a similar size. Initial conversations went well. I sent a sketch of the small front and back gardens, to-scale with measurements scribbled alongside. I also supplied some photos and an outline of what we were looking for. At a home visit Neil appeared competent, reasonable and friendly. He took his own photos and measurements at the visit. So far, so good.
And then he returned and handed us what he called ‘the plans’. There are 16 meanings listed in OED’s entry for the noun plan. None of these would be appropriate to use in reference to the pictures Neil provided.
At this point it had not dawned on me that this was AI generated and so I was confused by the many obvious flaws in these ‘plans’. Neil kept talking about the designer’s ideas on which plants to select to create areas of colour all year round. But we were more concerned about the fact that the proposed placement of the bin tidy obstructed the front door, that the aerial views did not match up with the front-on views, and that the ‘plans’ did not match reality. “But do you like it?” What does it matter if I like it, if it’s not possible within our actual real garden space? We sent Neil away with a list of things that needed changing (in his words ‘tweaking’, in our words ‘fundamentally redesigning’) and waited for version 2.
Sadly further versions were no better. They either omitted permanent structures such as the outdoor tap, access drain and garden gate, or placed them in impossible locations. Neil attempted to reassure me that the designer was taking on board feedback and had “done his own calculations”.
And then we were sent a link. To ChatGPT. And it suddenly all made sense. A large language model has no concept of space, measurements or physical properties. And so it’s attempts to produce a garden design bore little relation to reality. It’s lack of understanding of space is not a minor error that can be worked around. It’s a fundamental flaw that renders the designs useless. ChatGPT can create an aesthetic, a vibe, but that is all it can do. “No matter how much context you give it, AI tends to produce design work that looks plausible, but is very generic and on further inspection full of (obvious to us humans) flaws.”
What was most frustrating was that Neil could not, or would not, see how problematic the ‘designs’ were. If a human had made such obvious design errors, I have no doubt they would have been made to start again. Or have been fired. But the lure of the LLM and the shininess of it’s outputs somehow blinds people so that they forgive all, are quick to make excuses, and strive to find workarounds for it’s failings.
And so it is September, and our garden remains a vision of weed-afflicted slabs and needy pot plants. I have managed to grow tomatoes and chard, but my dream of raised beds, a wisteria arch, and abundant home grown veg remains unrealised. The saddest part of this is that I think Neil and his team are probably competent and could have created a beautiful space for us. But in his blind faith in, and repeated defence of, ChatGPT he lost our trust, and lost our business.
My experience of garden planning is trivial. But it exemplifies what we are seeing everywhere. People are outsourcing their thinking to machines that are not capable of the task. I wanted to pay humans to use their experience and skills and human brains and bodies to design and create a beautiful space for biodiversity, crops, and relaxation. If we stop paying humans to do these tasks there is a risk that we will lose skills and knowledge forever. There are many examples of failure of LLMs, some with potentially dangerous outcomes, and yet none of these failures seem to have resulted in caution, or slowed the pace of adoption.
- Referencing fake legal cases in evidence to New York federal court
- Giving dangerous advice to people with eating disorders
- Advising small business owners to break the law
- Including poison in meal recipes
A man wishing to reduce salt intake replaced sodium chloride with sodium bromide on the advice of ChatGPT and developed psychosis due to bromide toxicity.
A farmer in China lost 25acres of crops after outsourcing his decision making on pesticide use to a LLM, killing not only the weeds he wished to target, but also his sesame seedlings.
Of course there are also wider implications of these technological systems. “These systems consume an unfathomable amount of data, land, energy, labour, and water. They are rooted in profoundly disturbing ideologies that seek to flatten the world into a “one size fits all” abstraction and to replace humans with machines.“
Despite what we’re repeatedly told this is not inevitable. AI is not a force of nature, it is a human made technology. Marc Andreesen stated “Any deceleration of AI will cost lives” in his 2023 Techno-Optimist Manifesto, suggesting AI resistance as not merely futile, but murder. Thankfully, there are already many people resisting this hyperbolic framing: “the rhetoric of inevitability functions as a political choice, one that forecloses debate, concentrates power, and insulates a small number of corporations from accountability.”
The AI Resist List includes legal challenges, worker organising, community campaigns, artistic interventions, and technical tools. It is built by a team of global researchers, journalists, and scholars supported by the Distributed AI Research Institute (DAIR), We and AI, and the Refugee Law Lab at York University. The initial mapping of the list was done by Wael Qarssifi, a journalist from Syria, who has reported on the human impacts of surveillance in Syria, Malaysia, and the EU; and Verónica Martínez, a reporter and photojournalist whose work has documented the effects of surveillance and militarisation along the US-Mexico border. The AI Resist List not only asks us to reject what we do not want in our communities and lives. It also invites us to “imagine better tech futures, rooted in justice and regeneration for people and the planet.” Permacomputing, oriented around issues of resilience and regenerativity in computer and network technology, is just one of these imagined futures, being built by choices people make today.
It is essential that we centre humans in our decision making around technology and that we resist the erosion of critical thinking, trust and creativity. You can’t vibe code a garden. And you can’t vibe code a flourishing society.
We write history with our feet and with our presence and our collective voice and vision… We can change the world because we have many times before.
— Rebecca Solnit, writer and activist, Hope in the Dark
drlj 