Why Frugal AI ?
Generative AI, as the latest digital revolution, is transforming digital technology usage and highlighting issues such as responsibility, safety, and ethics. However, the environmental footprint of digital technology — already accounting for approximately 4% of global greenhouse gas emissions — is often overlooked. The rise of AI is expected to increase this footprint, intensifying pressure on vital resources like electricity and water. In response, the concept of frugal AI has emerged, questioning the tension between rapid AI development and planetary boundaries.
Historically, frugal innovation has been associated with low-income countries facing severe resource constraints. Today, it is recognized as relevant across all levels of development. The term ‘frugal’ is not just about doing more with less but about designing accessible and sustainable methods over the long term. Frugal innovation is defined as robust, scalable, functional, affordable, and local. It represents an opportunity to innovate cost-effectively and sustainably under resource scarcity.
What is Frugal AI?
Frugal AI is at the intersection of 4 domains: the economy, the technology, the society, and climate change.
Figure 1: Frugal AI is a multidisciplinary domain.
Digital sobriety is often used interchangeably with Frugal AI. Both concepts share a common goal of sustainability but differ in their scope. In general, digital sobriety aims to promote sustainable digital habits among users and organizations, aligning with broader environmental goals.
Frugal AI refers to the use of intelligent technologies to develop cost-effective, efficient, and resource-conscious solutions. It enables systems to learn, automate, and make decisions with minimal human intervention, focusing on accessibility, affordability, and adaptability in resource-constrained environments. Optimization techniques allow AI to function effectively within limited infrastructure, making it essential where conventional approaches may be impractical.
As AI technologies advance, the two key concepts – and – have emerged as critical considerations in both research and practical applications. Because these terms may seem similar, naturally it raises the question how it differs from what was already known? The answer is that they encapsulate different principles in the design and deployment of AI systems.
Efficiency refers to the optimal use of resources to achieve performance goals, focusing on computational, algorithmic, and operational efficiency.
Efficiency is a concept that emerged earlier than frugality but it is its key element.
Frugality mainly contrasts with efficiency by emphasizing affordability, simplicity, scalability and sustainability from the outset, rather than optimizing existing systems.
Frugality embodies a broader philosophy than efficiency, emphasizing minimalism, accessibility, sustainability, and context-aware development. It involves designing AI systems to operate under strict resource constraints, often resulting in novel, cost-effective, and sustainable solutions.
”The practice of engineers working with constraints is not new. But we argue the form of constraints has shifted in the current period: Away from pure efficiency or even technical precision, to a ‘mission-driven’ set of constraints.” Source: Bhatti, Yasser Ahmad and Ventresca, Marc, The Emerging Market for Frugal Innovation: Fad, Fashion, or Fit? (January 15, 2012).
In the literature, several terms are used interchangeably to describe aspects of resource management in AI.
Understanding these terms can help clarify the distinction between frugality and other approaches:
- Lean AI: Minimizing waste and complexity.
- Sustainable AI: Reducing environmental footprint.
- Green AI: Achieving results without increasing computational cost.
- Responsible AI: Developing and deploying AI safely and ethically.
The Key Principles of FAI technologies and solutions
- Efficiency: Prioritizing computational and energy efficiency.
Frugal AI aims to reduce environmental impacts by minimizing energy consumption and promoting the use of available resources wisely. This can lead to novel, context-specific approaches that differ from traditional, resource-intensive AI models.
- Affordability: Making AI accessible regardless of financial resources.
Frugality emphasizes creating AI solutions that are accessible in resource constrained environments. This is particularly important for applications in developing regions or for organizations with limited budgets.
- Simplicity: Favoring simple, user-friendly solutions.
To design a frugal innovative solution, the first question one needs to ask oneself is ‘are the needs of a targeted environment fully understood?’. Frugal AI systems are designed to function effectively, avoiding unnecessary complexity.
- Scalability: Designing adaptable, modular systems.
How Orange acts
Orange is actively translating the principles of frugal AI into everyday practices across research, engineering, and product teams, aligning with its Net Zero Carbon trajectory and ESG commitments.
- Measure to Act: A comprehensive carbon management tool is being implemented internally to guide and streamline decision-making related to integrating AI into products and services.
- Promote Frugal AI Within the Company: Fostering responsible AI literacy and encouraging mindful use of digital technologies to minimize environmental impact.
- Optimize Systematically: Employing techniques such as model compression (quantization, pruning, distillation), data-efficient learning, and carbon-aware deployment strategies (e.g., right-sizing, edge/on-device inference where appropriate).
- Operate Responsibly: Targeting energy-efficient hardware and low-carbon regions or time windows for workloads; ensuring infrastructure is appropriately scaled and decommissioning legacy resources to reduce energy consumption.
- Share and Strengthen Governance: Maintaining transparent governance through our Trustworthy AI strategy, supported by the DATA and AI charter and the ethical council.
These actions complement Orange’s broader initiatives to reduce energy consumption, digital waste, and electronic pollution, while ensuring AI remains trustworthy, ethical, and sustainable.
Challenges linked to Frugal Artificial Intelligence (FAI)
The development and deployment of AI models require extensive data collection, preprocessing, and considerable energy consumption for model computation and running. It means AI advancements generally come with significant computational and environmental costs, including CO₂ emissions.
The concept of Frugal AI ( ) emerged as the framework aimed at reducing reliance on resources, while maintaining AI effectiveness.
Two primary constraints are associated with FAI:
- The resources involved in equipment manufacturing and maintenance
- The usage of resources during AI operations, such as minimizing data volume for training and inference, and reducing memory, processing power, and battery consumption.
Frugal AI emphasizes long-term sustainability by design, aiming to minimize environmental impact when deployed at scale.
To understand, adopt, and implement Frugal AI principles, society needs to challenge existing habits and limit its needs within the planet’s ecological boundaries.
Strategies for FAI design
In order to achieve frugality with AI, several perspectives can be explored.
Eco-Design Principles: When embraced at the early stage of projects, approaching problems with the eco-design mindset can significantly impact the overall outcome. It is important to select only the essential functionalities and develop methods that are the adequate ones to the current needs. One should avoid to follow the hype (for example for using Gen AI even if not necessarily justified in a particular use-case) and instead boost measuring of different aspects of potential impact throughout the lifecycle. These strategies build the awareness of potential risks and assure that even on later project stages appropriate adaptations can be taken to compensate for erroneous initial assumptions.
Resource-Aware Design: Adapting AI systems to available resources.
Reuse and Fine-Tuning: Using pre-trained models to reduce training costs.
AI Optimizations: All the methods known from the efficiency standpoint can be used:
- Model Compression Techniques
- Hardware Optimization Techniques
- Algorithmic Optimization Techniques
- Deployment Optimization Techniques
- Data efficiency methods
More detailed description of each group of techniques and particular methods which are part of them, are to be found in the Orange Research Paper on Frugal AI [1].
Key Success Factors for Frugal Innovation Adoption
Govindan’s study identifies understanding the concept of AI and the level of AI investment as the most influential factors for successful adoption of cost-effective, scalable and sustainable frugal innovation.
Structured training and fostering AI literacy and digital sobriety are recommended to enhance organizational understanding and decision-making.
Conclusion
Frugal AI represents a paradigm shift towards designing AI systems that are inherently resource-conscious, accessible, and sustainable. While efficiency seeks to optimize processes, frugality seeks to rethink and redesign them. As AI continues to expand, the principles and practices of frugal AI will be critical for ensuring that technological progress aligns with environmental and societal needs.
To design frugal AI, it is advisable to:
- Understand AI’s impact on society and the environment,
- Incorporate and apply eco-design principles,
- Explore alternative setups with limited resources,
- Develop AI tailored to current needs,
- Comply with recommendations, specifications, and regulations.
Sources :
[1] Document d’étude sur l’IA frugale – https://www.researchgate.net/publication/390920260_Frugal_AI_Introduction_Concepts_Development_and_Open_Questions
[2] [How Artificial Intelligence Drives Sustainable Frugal Innovation: A Multitheoretical Perspective | IEEE Journals & Magazine | IEEE Xplore]
