Data centres, the US elections and artificial opinions: the dual pollution caused by AI
On the eve of the US elections, OpenAI is considering renaming its infrastructure ‘AI compute campuses’. Meanwhile, deepfakes, biased sources and automated responses are distorting public debate and influencing public opinion.
Before changing data centres – which are seen as ugly and polluting – we could try changing their name. Perhaps that would make them more likeable. It sounds like a joke, but it isn’t: OpenAI is reportedly preparing a public relations campaign for after the elections. Among the ideas is to rebrand them as ‘AI compute campuses’. The word ‘campus’ conjures up images of research and well-tended lawns, rather than expensive bills and an environmental footprint. The news, first reported by Axios, comes on the eve of tomorrow’s US mid-term elections on 3 November.
For the PORTS-Pike project in Ohio, OpenAI expects to create 35,000 construction jobs over six years of work and 2,500 long-term operational jobs. The project also involves new power stations, including gas-fired ones. A change of name does not solve the problem of how to distribute the benefits and costs.
Artificial intelligence thus appears to be increasingly at the centre of the political debate. It promises jobs, but raises fears about its impact on employment. It attracts investment, but requires electricity, space and water. Those living near the servers want to know what the benefit is of living close to the world’s AI powerhouses.
And then there is the actual electoral interference. In its briefing on the US elections on 26 October, NewsGuard reports on the case of James Talarico, a Democratic candidate for the Senate in Texas. An operation attributed to the Russian network Storm-1516 circulated AI-generated videos in which a fake former student accused him of abuse. The person whose identity was impersonated denied the allegations. However, in NewsGuard’s tests, Gemini and Copilot reported the accusation as if it had actually come from him.
NewsGuard has identified 1,181 political websites masquerading as independent local news outlets. When asked about the candidates and issues covered by these sites, the assistants cited them in 48 per cent of the cases examined. This percentage relates to that particular test, not to all the AI’s responses. The report does not measure how many votes were swayed. However, it demonstrates how propaganda can be given a new lease of life, repackaged with the neutral tone of a digital assistant.
