Several groups mainly tackling climate change and disinformation said the Biden administration’s approach to artificial intelligence regulation does not respond to AI’s impact on the climate and how it can be used to spread disinformation.
The groups sent a letter to the White House urging the Biden administration, including national climate advisor Ali Zaidi, to add more climate-centered policies in its AI executive order released earlier in October. Seventeen groups signed the letter, including Accountable Tech, Amazon Employees for Climate Justice, Climate Action Against Disinformation, Greenpeace USA, and the Tech Oversight Project.
The letter accuses the administration of neglecting AI’s impact on spreading disinformation on climate and the huge energy cost of training large language models.
“The creation and use of AI large language models heighten the risks associated with climate change on two fronts. First, LLMs can more quickly, cheaply, and covertly produce and distribute targeted climate disinformation, which threatens to perpetuate climate denialism and slow efforts to fight climate change. Second, due to their enormous energy requirements and the carbon footprint associated with their growth and proliferation, the widespread use of LLMs can increase carbon emissions,” the letter states.
Biden’s executive order does mention climate, as it directs government agencies to look into the possibility of using AI to combat climate change by providing insight into planning and improving the national grid and using AI to review environmental studies.
The group urged the government to require companies to report energy use and emissions throughout the entire AI life cycle, from training to inference. They also said companies should provide public explanations on how models produce information, the accuracy of climate change measures, and sources of evidence. The letter requests the government study the effect AI has on the spread of climate disinformation and implement safeguards around the mass production of disinformation.
The Biden executive order directs the secretaries of commerce, state, defense, and energy and the director of national intelligence to develop technical reporting standards for large AI developers.
The Verge reported in June that a study found tweets generated using GPT-3, now considered a less advanced model from OpenAI’s GPT-4, were more convincing than human-written posts. This raises the specter that AI can propel more misinformation on various issues. And a recent pre-print study highlighted the energy-hungriness of general-purpose AI models.
