CheckMag | The planet is on the brink of a climate crisis, while Google gives us unwanted AI search results with every click

There's no arguing with the benefits that AI can provide humanity. Autonomous vehicles, facial recognition, medical diagnoses and treatment, pharmaceutical research, scientific discoveries, programming. The list of potential benefits is huge. Even on consumer facing products such as AI image and video generation, summarising a 120-page document in seconds or revolutionising education through fully personalised learning, there is no arguing with the positive impact that AI could have.

There wouldn't be any debate if training hundreds of models came at no expense beyond the silicon, but unfortunately the power requirements for running AI on a global scale is significant. At a time when we should be reducing our energy requirements, not quadrupling them overnight.

Currently, global demand for data centre capacity is around 60 gigawatts and has remained relatively static until the last few years. However, demand is expected to reach 200 to 300 gigawatts by 2030, of which 70% will likely be attributed to AI. Much of this power will come from renewable sources, or in Google's case, taking the nuclear option, literally. However, there is speculation that AI could keep us dependent on carbon-based fuels for decades.

On May 20th 2025, MIT released an article breaking down the energy costs associated with each query run through a range of AI models, including Large Language Models (LLM) and image and video generators (Diffusion).

Even if you exclude the 50 gigawatt-hours of electricity it took to train GPT-4, the smallest text-based model with 8 billion parameters uses 57 joules of energy per response or 114 joules when accounting for cooling. On a large model with 50 times more parameters, that number climbs to 3,353 joules (6706 with cooling) for each response. 

It would be counterintuitive to get into the maths here, as MIT do a far better job, likening each response to travelling 400 feet (122 meters) on an e-bike. Google processes around 158,500 searches every second. So, by MIT's maths, if we could capture the power associated with running Gemini for 1 second, a person could travel 19,337 kilometres on an e-bike, or roughly one and a quarter times around the planet. 

Google (and pretty much every AI model out there) admits "Gemini can make mistakes, including about people, so double-check it". Why would we want to waste precious energy on an AI-generated search response, which then has to be fact-checked through a traditional site anyway? Unfortunately, turning it off isn't as straightforward as you would think.

The first problem is that there isn't a consistent solution for turning it off on any device. Google support is littered with requests to turn it off, with most solutions being either non-existent, or having no effect. One solution suggests disabling it from "myactivity.google.com" and turning off Gemini activity, but no such setting exists. Another solution suggests going to your Google account, selecting "Data and Privacy" and turning off "Web and App Activity" also with no effect. Currently, the only way I am aware of disabling it is to run an ad blocker and add a very specific filter, although it is questionable whether this prevents it from running or just that the output cannot be seen in the browser.

The second issue is that there isn't a single location in which to control Gemini. If you want to turn it off on your phone, the settings are completely different to your PC's browser. If you want to go further and disable it in Gmail or Google Docs, there are different locations again. An article by NordVPN as far back as February suggests it isn't possible to turn it off at all, and this is echoed in some of the Google support threads.

Of course, there are plenty of people that value Gemini in their searches, but not giving people a choice about a technology as divisive as AI will ultimately push them elsewhere. As someone who has subscribed to Google One for close to 10 years, I should have the option as a paying customer to turn it off. But alas, only Enterprise and Education domain admins have that privilege, likely because Google knows businesses won't tolerate its covert data processing tactics.

The market has proven that no one wants an AI Pin and the Rabbit R1 has so far failed to reach a broad audience. Perhaps Sam Altman and Jony Ive will succeed where others have failed. However, AI has evolved to the point where the same Sam Altman who created ChatGPT has developed a tool to mitigate the problems caused OpenAI. A tool that works by offering every person on the planet a sum of cryptocurrency to have their irises scanned to prove they are human. The entire concept sounds like it's been pulled directly from a dystopian science fiction novel (available on Amazon).

The prolific integration of AI into every facet of companies' products is making the technology completely unavoidable for consumers. Google forcing Gemini to appear at the top of every search result only serves as a constant reminder of how damaging, wasteful and pointless AI can be in the wrong hands.