White paper
Frugal AI: an essential challenge for the development of artificial intelligence technologies
Edited by :
Vincent Six
Summary
This professional thesis, written by Vincent Six in 2022, was ahead of its time in addressing the emergence of generative AI.
Vincent Six, the author, describes himself as passionate about innovation and new technologies. He has worked for over 25 years in higher education, focusing on engineering training.
To support his research, he conducted interviews with around 15 professionals specializing in AI or sustainable IT practices.
The result of nine months of research, extensive reading, interviews, and analysis, this thesis provides key insights for a more responsible use of AI technologies.
It is structured as follows:
1. Environmental challenges
2. The environmental impact of digital technology
3. Artificial intelligence and its sustainability
4. Towards a culture of frugality
The editorial perspective
This thesis is an in-depth and rigorous piece of work. The author precisely describes the context and assesses the impact of AI specifically and digital technology more broadly.
In this regard, the thesis serves as a valuable reference for understanding the environmental footprint of digital technologies.
The third of four sections focuses specifically on AI, allowing for a detailed examination of AI’s technical challenges.
This section is well-researched, offering a clear understanding of AI-powered services, language model development methods, and the environmental footprint of AI.
The core argument of the thesis and its response to its main question lie in this third section.
Beyond discussing AI’s impact, the author explores multiple strategies to reduce AI’s energy consumption, covering various techniques to shrink model sizes, reduce complexity, and enhance frugality.
Given that the thesis was published in June 2022, it is especially impressive that it managed to compile such a comprehensive range of actionable solutions.
The final section, while interesting, only partially answers the research question. It spends too much time discussing corporate culture and the mindset shift needed for frugality (which seems to be conflated with the concept of "sobriety" in the thesis).
As a result, this final section somewhat overshadows practical solutions for achieving frugal AI.
In brief, the editorial perspective
The most
- Excellent anticipation of the subject, with the thesis published in June 2022
- Very good description of digital environmental impacts
- Enumerates and precisely describes different ways to achieve more frugal AI at the algorithmic level
- Very good explanations on different types of algorithms and technological layers involved in AI-based services
- Good summary table before the conclusion to assess AI projects globally for frugality issues
The least
- The core of the thesis does not sufficiently explore ways to make AI more frugal
- Some broad proposals in the last section on "frugal corporate culture" are vague, not specifically applied to digital issues, and lean more towards general sobriety than AI frugality
Publication date
June 2022
Available in
- French
License
Intellectual property of the author