Faced with the growing imprint of AI, the frugal approach is a lasting solution for a responsible digital future, by favoring sobriety and efficiency.
Artificial intelligence is undoubtedly the most powerful transformation engine of our time. The actors of technological innovation are at the forefront of this revolution. However, beyond the legitimate enthusiasm, special attention must be paid to the growing imprint of AI on our resources and our environment. This is why a more pragmatic and reasoned approach, called frugal AI, is not only desirable, but essential to build a lasting digital future.
Beyond the buzz: AI, a story much richer than the LLM
Since the advent of models like Chatgpt in 2022, the general public and part of the industry seem to reduce AI to the main models of language (LLM) and to generative AI. However, AI is a discipline that is over forty, punctuated by major advances in computer vision, natural language processing (Speech-to-Text, text-to-speech), predictive analytics and automatic learning. These technologies, often less publicized, are based on proven algorithms and, paradoxically, often less energy -consuming than current LLM.
Admittedly, the versatility and ease of use of LLM are undeniable. But their operation requires colossal IT resources, resulting in astronomical electrical consumption. When Jensen Huang, CEO of Nvidia, evokes the need to multiply by 100 the computing power available to meet future demand, it is imperative to become aware of the real cost of the AI, not only financial, but above all environmental.
The myth of “green” AI: a dangerous illusion
A frantic race for computing power is launched, where some digital giants are invested massively in electricity production, including nuclear, to supply their data centers. This approach, often presented as a quest for “green” energy, actually masks an ambition for disproportionate development and hegemony. The bet is that of an ever more powerful AI, and therefore, always more energy -consuming: fleas, servers, cables, cooling systems … without forgetting the building materials of these infrastructures.
Although LLM paradoxically becomes cheaper for use as technology is progressing, this increased accessibility encourages ever more intensive use. The speech of the “Green AI” tries to minimize an environmental impact which is very real and unbearable, with CO2 emissions from the digital sector increasing by 6 to 8% per year.
A fundamental question arises: should we really put AI everywhere? The answer is a categorical no. The responsible company cannot ignore the implications of such an exponential growth.
“Do better with less”: a guiding principle
For a software developer, systematically turning to an LLM, regardless of the problem to be resolved, often likes a barrel to kill a fly. A more reasoned and frugal approach to AI is advocated. It is a commitment in terms of responsibility: whenever possible, it is a question of “doing better with less”. This means recognizing the limits of a system that has been raw infinite too long. Eco -responsibility is a cardinal value which must guide the decisions of design, development and use of products, including in the face of generative AI.
The pillars of the frugal AI
A frugal AI approach begins with an in -depth questioning of use cases. It is essential to inventory and prioritize needs according to their usefulness and their concrete added value for users. The question of the “how” technical “only intervenes second, respecting clear principles:
- Do not limit yourself to LLM: it is crucial to systematically explore the alternatives. For a specific need, a specifically driven algorithm can surpass an LLM in performance and energy sobriety.
- Favor open source: this allows better control of tools, processes and, especially data, both in the test phase than in production.
- Work locally: deploy technologies on local servers, rather than in distant datacenters, strengthens the sovereignty and data security, while reducing the carbon footprint linked to the transport of information.
The application of these principles pushes to favor the most sober artificial intelligence technologies. The “good” technology, the optimal solution, is that which responds in the most frugal way possible to each case of identified use. For example, for image recognition, and in particular thanks to the help of neural networks, higher results can be obtained with a tailor -made model, for an energy expenditure a hundred times less than with a multimodal LLM. That’s it, “do better with less”.
However, the frugal approach is not a rejection of the LLM and the generative AI. It is a question of integrating them in a thoughtful way. Collaboration with national companies offering LLM accommodated locally makes it possible to take advantage of these technologies while respecting the principles of sovereignty and sobriety. These solutions can enrich the functionality and boost the performance of existing systems, provided they are used in a targeted and reasoned manner.
Ultimately, the frugal AI activates all the power of AI to deliver a concrete value to users, focusing on real uses. Whether it is optimizing complex processes, simplifying daily tasks, or improving communication, the objective is to create impacting and lasting solutions. Starting from uses, it is possible to develop an AI which is both efficient, responsible and lasting. It is a vision of innovation for a smarter and more respectful world.
