The hospital trust I work for has recently announced plans to widely implement AI ambient voice technology (AVT). Many of us have concerns from a clinical, educational and data governance perspective, but my biggest concern is the environmental impact. From what I can understand, no one knows the environmental impact, because companies don't release that information. But we know it is huge, with estimates (from the United Nationas University) is that "By 2030, the global data centres powering artificial intelligence are projected to consume 945 terawatt-hours of electricity. This is nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria—countries collectively home to more than 650 million people. Their associated water footprint will equal the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa.
I've also quickly realised that this is not just individual trusts or GP practices, this is being driven by national policy. But without understanding the environmental impact of it, how are we able to take account of this in the NHS's legal (and hopefully moral) obligation to reach net zero? And without this information, how can those of us required to use these tools (as it appears we may not have a choice in it) make an informed decision about how we use them?
I'm a clinician, and know nothing about AI, so apologies if these are silly questions, but I haven't been able to find the answer searching here so far. I would be really interested to hear if anyone:
- Has any concrete data on AI environmental impact for individual companies (eg Heidi, one of the most popular AVT tools)?
- Has any suggestions on how we can challenge AI roll outs- if we can't stop it, how can we ensure that the greenest options are chosen (especially if there is no actual data)?
- If you already have AVT/other widespread AI in your organisation, do you have any policies in place to ensure they are used in the greenest way possible? Or again is that impossible if no one can measure the impact?
- In view of the above, is there anything we can do to hold the AI companies to account and get them to release their emissions/water use etc data?
Thank you!
Thanks for raising this Jennifer - these are definitely not silly questions! I am a paediatric radiologist, and in recent weeks we have been trialing some new AI-powered MRI sequences which look like they are both faster and better quality than our previous images. This is potentially good news for patients, particularly very young patients as they may be able to avoid a general anaesthetic, both safer and avoiding the greenhouse gases of anaesthetic agents. However, to weigh up whether these potential benefits are not outweighed by the energy and water costs associated with the AI we need transparency, which at present is hard to come by. I was at a big radiology conference in Vienna earlier in the year (took the train!) and was asking all the AI salespeople if they could provide figures on carbon footprint or water consumption and didn't get very far. I am trying to raise awareness of this issue within radiology, but agree this needs to be raised at organisational level and across society more broadly (how much energy is being used for generative art and videos for purely recreational purposes?). Sorry not to actually answer any of your questions apart from number 4 - worth writing to your MP.
Best wishes,
Michael
Hi Jennifer! I'm one of the only Health Managers in Australia involved in the sustainability impacts of AI datacentre hyperscaling and I've been watching this closely. First world governments are rapidly moving to set policies for datacentres to build their own power (gas turbine and solar have been primarily purchased globally so far) and to contribute to the grid instead of taking from the grid, and some datacentre providers are also plugging into recycled water (grey water) to cycle evaporate cooling needs for APU and GPU chips instead of burning through the local drinking water. in Australia almost 3% of all power produced now goes to cloud centres and AI datacentres, with a projected 8% by 2030, with the biggest power draw being AI training rather than image generation or prompts. I understand Ireland is aparrently experiencing one of the biggest electricity-demand burdons from AI datacentres in the world over there.
Our solution in health for the monitoring and restriction of wasteful AI generative and iterative use is an enterprise AI portal on our computers and devices(not built yet), that gatekeeps all AI usage against rigorous usage policies, and measures the upstream/downstream impacts on cost and emissions. We cant restrict personal devices obviously, but its a huge step forward for annual reporting, waste restriction, and usage research in health.
Please log in or sign up to comment.