"Key findings
• General Purpose Artificial Intelligence (GPAI) models are predictive tools that produce outputs on the basis of probability distribution. Contrary to the dominant narrative they are not ‘autonomous' nor do they have real ‘agency'. The central role played by humans at all levels of their development, training, implementation and use should not be underestimated.
• The governance framework of GPAI models is still ‘under construction'. The European Union's (EU) 2024 Artificial Intelligence (AI) Act provides a general framework, but more governance measures are planned: harmonised standards to be laid down by the European standards organisations, implementing and delegated acts adopted by the European Commission, and the evaluation of systemic risks and the drawing up of a GPAI Code of Practice by the European Artificial Intelligence Office. This will provide an extra layer of governance that will change the nature of legal obligations, as will rulings by courts and supervisory authorities.
• The electricity consumption, water usage and carbon footprint of GPAI models need to be addressed if the sector is to evolve towards more environmental sustainability. Legally obliging providers to inform users about the environmental impact of individual requests, as a way to help people and companies make better informed decisions, would be a crucial first step. Beyond that, other initiatives are needed to ensure that the entire generative AI ecosystem becomes more sustainable and socially responsible. The governance interventions referred to above can contribute to this."

Support: Digital

ISBN (PDF) : 2031-8782