Feeling AI-lienated …. and what I’m doing about it

The lack of detail and transparency in reporting makes it very difficult to put precise figures to the impacts, but the best estimates I could find indicate that AI systems in 2025:

1. directly drove carbon emissions at a similar rate to the whole of Greece or Chile, or New York City (~30–80 million tonnes); and
2. directly drove water consumption on a par with global annual consumption of bottled water, or the entire water footprint of Iceland (~300-800 billion litres). 

This is rising fast with AI-induced water consumption and carbon emissions from data centres alone projected to double - if not quadruple - over the next four years. In the UK the climate impacts of AI datacentres were dramatically underestimated, with recent government data suggesting they could be ten times higher than previously calculated. These could add over 100 million tonnes to UK CO2 emissions over the next decade alone, which is hard to see as compatible with the UK’s legally binding commitments to cut approximately this amount from its annual emissions every five years. 

Despite talk of their commitment to sustainability, around 5 years on from making pledges to radically cut CO2 emissions, tech companies’ footprints look worse than ever as they pour resources into AI. Analysis of tech giants’ self-reported data showed: “Google’s emissions jumped nearly 50%. Amazon’s rose by 33%, Microsoft’s more than 23% and Meta’s more than 60%.” 

AI-driven electricity demand is serving to maintain and even increase the demand for fossil fuels and other destructive, unsustainable forms of energy generation. This has already meant the maintenance of coal plants where they were otherwise due to be phased out, more gas plants, more opportunities for unsustainable biomass projects, and ultimately even more profit for polluters at the expense of a habitable planet for us all. 

The additional water stress caused by AI data centres is about the last thing we need amidst the absolute necessity to protect freshwater supplies in a heating world. But nonetheless it’s already leading to restrictions in households’ access to water, contributing to ‘mega-droughts’, and devastating ecosystems.

Impacts linked to AI data centres and the wider supply chains span much further too: exacerbating air pollution, affecting land use, driving e-waste and the harms caused by mining for minerals to name a few. And this is before the effects of what AI is used to do are even factored in (see below).

Whilst there are lots of possible positive applications for them, most of what I’m seeing AI tools used for is in assisting us to continue carrying out all the harms we’ve been wreaking on ourselves and the world around us… only now we can do so with more speed, less thought, less effort and often more error (see below). I think this is what has saddened me the most. 

It’s turbocharging overconsumption when we are already beyond the limits where our systems can recover from it. It’s being used to promote dangerous political agendas, as a tool for manipulation and misinformation. It’s facilitating mass surveillance at a time where authoritarianism is on the rise. Amidst high and escalating global tensions, we are seeing its creeping use in warfare (for example in autonomous weapons).  It’s burning us out, isolating us and further dividing us when we need to be connecting with one another and working together.

Enormous investment in AI has made a small number of people extraordinarily wealthy (or even more extraordinarily so) but will benefit ‘trickle down’ to the rest of us? And could that benefit possibly justify the social and environmental harms we are seeing? In the absence of very careful planning, requiring significant involvement of and leadership from workforces, new technologies have tended to benefit the most powerful at the expense of the majority. What we are seeing right now with AI shows no sign of bucking that trend. 

Training the models that underpin major AI platforms has required massive amounts of data, much of it taken without consent or compensation made to the creators of that data. Adding insult to injury, this kind of “brazen theft” is undermining the ability of writers and artists to make a living. [As a result a whole host of plaintiffs including authors, publishers and artists have resorted to suing Meta, Open AI, Anthropic and others for copyright infringement].

Less well-publicised are the stories of how the proliferation of AI systems and infrastructure is fuelling exploitative working practices in some of the poorest communities in the world. This includes low paid, precarious work, which can be physically dangerous, mentally traumatising or both (examples in mineral extraction and content moderation). 

The large language models (LLMs) underlying text-based AI tools (such as chatbots) are probability-based: they predict the next words based on everything that’s been used to train them. This makes them very prone to reproducing and reinforcing the underlying biases of their training data, and liable to create entirely fictional statements presented as if they’re factual . 

Far from being neutral information sources, AI tools have shown covert racial bias, gender bias, and a shift towards the political right. Since the use of AI has become so widespread, this is already widening equality gaps, contributing to discrimination in hiring procedures, policing, healthcare provision and even medical diagnosis to name a few. 

And whilst tools and their training might be getting more sophisticated, their outputs are still a long way from being reliably accurate. A recent study led by the BBC highlighted that AI assistants “routinely misrepresent news content no matter which language, territory, or AI platform is tested”, with “significant” issues in nearly half of the responses studied.  

Errors and ‘hallucinations’ like this are misleading and risky for users of AI tools: they stand to degrade the quality, trustworthiness and overall value of work that has been produced using it. And whether we choose to use those tools or not, we are all at risk from the impacts of their mistakes.  We already know that AI-created falsehoods are making their way into everything from health and wellbeing advice to evidence at criminal trials. The consequences can be extremely serious: for example, errors in facial recognition are known to have led to numerous wrongful arrests - and lengthy imprisonment - of people who could not possibly have committed the crimes they are accused of.

Emerging experimental evidence indicates that interacting with AI tools might progressively and rapidly undermine our ability to learn effectively. It stands to degrade our analytical, creative, relational and critical thinking skills, contributing to reliance on technologies we don’t fully understand… and don’t have free, equal nor guaranteed access to. 

Earlier this year the CEO of OpenAI (the company that owns chat GPT) said publicly: “We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter”.  That future - one that sounds deeply dystopian to me - is much more likely to be enabled the more we offload tasks to AI that we could otherwise do (or learn to do) independently. 

With the rapid pace of AI integration and with millions in corporate money being ploughed into pro-AI lobbying, it’s perhaps unsurprising that regulation of AI has been pretty limited. Where governments have legislated on AI, policies often explicitly emphasised enabling “innovation” over proceeding with caution. This has come at the expense of preventing harm and upholding other ethical principles (see above). 

Transparency and accountability are severely lacking across the sector, all whilst companies are building increasingly more complex, more autonomous AI systems without credible plans to prevent or deal with catastrophic outcomes.