15 Amendments of Marie TOUSSAINT related to 2021/0106(COD)
Amendment 75 #
Proposal for a regulation
Recital 3 a (new)
Recital 3 a (new)
(3 a) In order to ensure the dual ecological and digital transition, secure the technological resilience of the Union and achieve the objectives of the new European Green Deal, sustainability should be at the core at the European AI framework and guarantee that the development of AI is compatible with sustainable environmental resources for current and future generations, at all stages of the lifecycle of AI products; sustainability of AI should encompass sustainable data sources, power supplies and infrastructures;”
Amendment 78 #
Proposal for a regulation
Recital 3 b (new)
Recital 3 b (new)
(3 b) Studies have shown that the training and use of AI has significant environmental impacts which, if they are left out of the equation, could threaten the EU’s objectives under the Green Deal. The environmental impact of AI encompasses both the critical raw material needed to design infrastructures and microprocessors, as well as the energy required by the development, training and use of an AI system. In particular, tuning, meaning the action of re-purposing or refining and AI model, was found to be even more environmentally costly than training a model in the first place. Best practices in AI sustainability should include a reasonable allocation of resources, consider potential shortages in key critical raw material and limit unnecessary data acquisition and processing.
Amendment 81 #
Proposal for a regulation
Recital 3 c (new)
Recital 3 c (new)
(3 c) To ensure sustainable AI, developers should report key environmental parameters such as training time and resource use, expected energy and data processing required by the use of the AI during its lifetime, and provide carbon emission reports to regulatory authorities in order to enable transparency and comparison between models. Tools for calculating emissions, like the Machine Learning Emission calculator, are already available on the market and should be built upon and systematically used as a matter of transparency requirement and reporting obligations.
Amendment 82 #
(3 d) Sustainable AI taskforces should be incorporated in national surveillance authorities and in Member State government and relevant national and European agencies, in order to maintain the sustainable development and use of AI.
Amendment 83 #
Proposal for a regulation
Recital 3 e (new)
Recital 3 e (new)
(3 e) In order to ensure the compatibility of AI development and sustainability goals, a “proportionality framework” should assess whether the training or the tuning of an AI model for a particular task is proportional to the carbon footprint and environmental impact it would have. Such a scheme should enable certain model training and development to be stopped in case the predicted environmental cost is deemed to exceed the social, environmental and economical benefit or if another non-AI solution with an equivalent level of success is available.
Amendment 104 #
Proposal for a regulation
Recital 27
Recital 27
(27) High-risk AI systems should only be placed on the Union market or put into service if they comply with certain mandatory requirements. Those requirements should ensure that high-risk AI systems available in the Union or whose output is otherwise used in the Union do not pose unacceptable risks to important Union climate priorities, environmental imperatives and public interests as recognised and protected by Union law. AI systems identified as high-risk should be limited to those that have a significant harmful impact on the health, safety, greenhouse gas emissions, crucial environmental parameters like biodiversity or soil pollution and fundamental rights of persons in the Union and such limitation minimises any potential restriction to international trade, if any.
Amendment 193 #
Proposal for a regulation
Article 13 – paragraph 3 – point b – point v a (new)
Article 13 – paragraph 3 – point b – point v a (new)
(v a) key environmental performances during the training and expected during the using phase in the form of a multicriteria life-cycle assessment report considering the material and energy impact of the AI system.
Amendment 197 #
Proposal for a regulation
Article 15 – paragraph 4 a (new)
Article 15 – paragraph 4 a (new)
4 a. High-risk AI shall be designed and trained according to sustainability standards with regard to their material and energy resource consumption. The Commission shall develop a “proportionality framework” assessing whether the training or the tuning of an AI model for a particular task is proportional to the carbon footprint and environmental impact it would have. Such a scheme shall enable model training and development to be stopped in case the predicted environmental cost is deemed to exceed the social, environmental and economic demonstrated benefit or if another non-AI solution with an equivalent level of success is available.
Amendment 206 #
Proposal for a regulation
Article 41 – paragraph 4 a (new)
Article 41 – paragraph 4 a (new)
4 a. The Commission shall develop sustainability standard requirement for AI systems and AI development practices after consultation of relevant stakeholders, including businesses, NGOs, AI and sustainability experts and academics.
Amendment 209 #
Proposal for a regulation
Article 43 – paragraph 4 a (new)
Article 43 – paragraph 4 a (new)
4 a. For the purpose of environmental conformity assessment, the provider shall perform a multicriteria life-cycle assessment reporting considering the material and energy impact of the all life stages of the AI system.
Amendment 223 #
Proposal for a regulation
Article 54 – paragraph 1 – point a – point iii
Article 54 – paragraph 1 – point a – point iii
(iii) a high level of protection and improvement of the quality of the environment, meaning the costs of developing the AI system shall not exceed the benefit of developing it for the purpose of protecting the environment;
Amendment 226 #
Proposal for a regulation
Article 54 – paragraph 1 – point a a (new)
Article 54 – paragraph 1 – point a a (new)
(a a) the principle of data minimisation should be upheld, and the data acquisition and processing shall be kept to what is strictly necessary for the purpose of the AI application;
Amendment 231 #
Proposal for a regulation
Article 57 – paragraph 1
Article 57 – paragraph 1
1. The Board shall be composed of the national supervisory authorities, who shall be represented by the head or equivalent high-level official of that authority, and the European Data Protection Supervisor. Other national authorities mayshall be invited to the meetings, where the issues discussed are of relevance for them. A sustainable AI taskforce comprised of independent digital sustainability experts shall be established within the board to ensure the systemic consideration and inclusion of the EU’s environmental imperatives within the regulation of AI.
Amendment 258 #
Proposal for a regulation
Annex III – paragraph 1 – point 4 a (new)
Annex III – paragraph 1 – point 4 a (new)
4 a. Environmental impact and energy use: According to the “proportionality framework” assessing whether the training or the tuning of an AI model for a particular task is proportional to the carbon footprint and environmental impact it would have: (a) AI systems whose predicted environmental costs exceed the social, environmental and economic demonstrated benefit; or (b) AI systems where another non-AI solution with an equivalent level of success is available.
Amendment 261 #
Proposal for a regulation
Annex III a (new)
Annex III a (new)