A carregar...
Publicação
Carbon footprint of artificial intelligence (AI) models: estimation and reduction approaches
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 677.41 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Artificial intelligence (AI) has become an integral part of our lives, with its models
increasingly used across various sectors. However, there is limited research on its
environmental and sustainability costs. With this study, we intend to advance the
understanding of how carbon emissions associated with AI models can be measured and
reduced. This is done through literature review, the analysis of real-world case studies,
and the elicitation of expert stakeholder’s perspectives. This combination of
methodologies enables a comprehensive evaluation of the practices currently used to
calculate the carbon emissions associated with the training and inference of AI models,
as well as of the strategies applied to mitigate their carbon footprint. Our results show
that: (i) current estimations of the carbon emissions of training and deploying AI models
are flawed due to the limited understanding of their complexity, unavailability of coherent
estimation frameworks and incomplete data availability; (ii) the adoption of standardized
emission’s reporting among tech companies is a necessary step towards more accurate
calculations; (iii) the implementation of carbon reduction techniques such as algorithm,
hardware and data centre optimization can serve as possible solutions to minimize the
carbon emissions of these models; (iv) all stakeholders involved in the AI model’s
lifecycles need to be publicly informed about the emission impact and actively engaged
in mitigation efforts. This thesis acknowledges the growing data and computational
resources that accompany the current advancement of AI models and discusses how this
trend may affect their long-term environmental sustainability. Future research and
research policies should focus on addressing major gaps in AI model carbon emission
calculation and on developing effective mitigation strategies and technological innovation,
which can support companies’ efforts towards a more sustainable AI.
Descrição
Palavras-chave
AI Models Carbon Emissions Sustainable AI Emission Calculators Carbon Reduction Frameworks Stakeholder Engagement
Contexto Educativo
Citação
Şark, Mina Vildan (2025). “Carbon footprint of artificial intelligence (AI) models: estimation and reduction approaches”. Dissertação de Mestrado. Universidade de Lisboa. Instituto Superior de Economia e Gestão
Editora
Instituto Superior de Economia e Gestão