Hello subscribers – I just wanted to share two new things on my website that I hope will be helpful to those interested in advocating for the environment. And some short thoughts on a related topic…

The first is a daily news update which sends an automated list of new and blogs that are (mainly) relevant to the Scottish environment each day. You can sign up here, and check tomorrow it doesn’t end up in your spam folder.

The second is a dashboard I’m putting together of Scottish environmental polling. This is far from comprehensive but should be a useful resource if you’re looking to quickly find some polling figures – you can find it here.

I built both of these using generative AI to produce the code. I’m being explicit about this because I am acutely aware of the huge environmental and ethical questions around AI, as well as the fact that huge amounts of inaccurate, dangerous, and frankly cringe AI-generated content is quickly making large parts of the internet pretty unusable.

I am hugely sceptical of the claims made by AI boosters about the value of the technology or the inevitably of it upending work, and society, as we know it. The obscene levels of capital investment in a speculative rush to build data centres is an obvious bubble. Tech firms are burning cash chasing this tech in the hope they will find a way to make it pay later, usually explained by the inevitability of discovering “artificial general intelligence” where their chatbots will become more useful than human workers and either replace us all or drive incredible productivity gains.

It’s all a bit silly. Anybody who has spent 5 minutes with an AI chatbot on a subject they understand well will have quickly discovered their limitations and shortcomings. I have found using LLMs for research like being confronted with a bright but overconfident teenager – they might draw together some relevant sources but it is almost always wrong in a subtle but important way.

Big tech are betting they can overcome these limitations through “scale” (building absurd levels of data centre capacity). I think the opposite will happen. LLMs are trained on the internet, and as the internet becomes more and more dominated by AI written or edited content, I think it is likely that LLMs will in fact get worse over time by cannibalising their own output.

So why use these at all? On the whole, I wouldn’t – certainly in a professional capacity I’d view using AI outputs as a way to make me less effective. Even if LLMs could do the core functions of my job “better” than I can, I wouldn’t use them, just as I would never choose to read an AI-written novel or listen to AI-generated music even if these were “good”. Some thing are supposed to be human.

But I’m not close-minded to the idea that this new tech can have some genuine uses. And the ability of LLMs to translate natural human language into computer code is now at the point of being genuinely astonishing. It is also flawed. I’ve used Claude on the tools referenced above as well as some personal projects. It does run into problems, and without specific prompts there is a tendency to guess what features the user is looking for, which I find frustrating. But I can also see that it’s a huge breakhthrough and allows those without programming skills (like myself) to create things they would not have otherwise been able to. I don’t think it will replace software engineers for the same reason it won’t replace doctors or lawyers – it’s fine to let the computer do these things up until something goes wrong.

In a saner world we would be approaching this technology with caution, planning a rollout of infrastructure that was genuinely consistent with environmental limits, and targeting LLMs towards their genuine strengths instead of forcing them down users throats. The impact of the war on Iran on energy prices may well bring us closer to that world sooner than many think.