Researchers abstaining from generative AI cite ethics and costs

Several researchers have announced they are abstaining from using generative AI tools.

• Main action: Several individual researchers (including Danielle Crowley at Bangor University, Hugh Possingham at the University of Queensland, Audrey Moores at McGill University, Tanisha Jowsey at Bond University, Juan Rocha at the Stockholm Resilience Center, Michaela Socolof at MIT and Elizabeth Wolkovich at the University of British Columbia) say they are purposefully abstaining from using generative AI because of concerns about copyright/consent, transparency of training data, accuracy/hallucinations, and environmental impacts; some have adopted explicit policies (for example, Wolkovich will not chair defences or join committees where students use genAI for writing).
• Background and evidence: Surveys and studies cited include a Nature survey of ~5,000 researchers showing high acceptance for AI editing but far lower use for text generation, an Elsevier survey of 3,234 researchers reporting 58% use of AI, and a Patterns study estimating 2025 global AI system footprints of 32.6–79.7 million tonnes CO2 and 312.5–764.6 billion litres of water; other factual details include specific examples of AI errors (hallucinated citations and incorrect molecular depictions) and published commentaries calling for restrictions on genAI in chemistry and qualitative research.