The rise of artificial intelligence is not without major environmental challenges. The environmental impact of AI is colossal and probably still underestimated, as our Data & AI experts reveal.
Contents
AI: a new revolution that raises questions
DeepBlue in 1997, Siri in 2011 and ChatGPT in 2022. Artificial intelligence has experienced many revolutions. Today, with ChatGPT, OpenAI has democratized and proven the effectiveness and interest of a new model (the LLM). AI, and in particular Generative AI, is seen as the solution of all company's problems.
The adoption and multiplication of AI use cases requires many resources, whether during training or use, which raises the question of the environmental impact of AI.
The environmental impact of AI continues to grow
The rebound effect principle to explain the environmental impact of AI
Some studies quantifying the carbon footprint of our digital uses put forward a surprising idea. Generative AI would be more eco-responsible than traditional tools or human work without technology.
Why is that? Their speed of execution is said to compensate for their energy-hungry nature. But this reasoning overlooks a key aspect: the rebound effect.
Easy, rapid access to AI has led to an explosion in the number of uses and creations... some of them useless, cancelling out the expected environmental gains.
It's not just the tool that needs to be analyzed, but above all the way we use it. This is essential thinking if we are to adopt a more sustainable approach to these technologies!
Energy costs
The training of BLOOM (Hugging Face's latest LLM) emitted around 50 tons of CO2 equivalent, the equating to 60 flights from New York to London. What is more alarming is the fact that the model was trained on supercomputers in France, where the energy mix allows for lower emissions thanks to nuclear power.
Another example: OpenAI's GPT3 and Meta's OPT have emitted over 500 and 75 tons eCO2 respectively. We fear that this figure will increase as training data and model complexity expand.
The use of generative AI also has an impact: generating an image consumes as much energy as it takes to recharge a smartphone, while generating text is certainly more sober, but 100 requests to chatGPT consume the equivalent of the energy needed to light a 50W bulb for 1 hour.
Water consumption of data centers
Towards a reasonable and reasoned use of AI
It is therefore imperative to consider whether or not it is necessary to use and deploy an AI use case within companies. AI is not a miracle or default solution, and it carries risks. Many "AI" use cases are simply automation requirements, which can be described by rules that are easier to maintain and understand than the latest LLM.
Not all AI use cases require the most powerful, energy-intensive models. More lightweight, customized solutions, using models pre-trained on company data (RAG, fine tuning), can often suffice, and are much more economical in terms of energy consumption. These strategies not only reduce costs, but also optimize resources, but are also tailored to actual project needs.
Companies need to see AI not as a one-size-fits-all solution, but as a tool to be deployed judiciously.
How to choose your AI use case (and limit its environmental impact)?
To determine a data or AI use case, companies need to understand and translate their business needs into operational solutions. Few companies today have the maturity and skills to do this without support..
Numerous strategies can be put in place to meet these challenges:
- Evaluate and monitor the environmental impact using KPIs (energy and water consumption, etc.), particularly in the context of the CSRD.
- Reduce the impact of the use of Generative AI by training the different departments people in prompt engineering so as to use a minimum of queries and iterations.
- Raising awareness of environmental issues
- Develop a "sustainable AI by design" approach (take resources and needs into account when choosing algorithms, choose the most optimized model for the use case and not the best according to the press, optimize the AI life cycle by reusing pre-trained models to train other models).
All these elements require support and guidance to move towards "sustainable AI" (responsible AI). That's why at iQo, as a B Corp certified consultancy, and expert in Data & AI issueswe offer a range of support services to frame the implementation of "sustainable AI by design".
AI training: explaining and demystifying before taking the plunge
Training employees in AI is not just about responding to a current trend, it's about supporting the transformation of our businesses. All businesses are
AI Readiness: is your CIO ready for Generative AI?
AI Readiness or how do you know if your company and the various business lines are ready to take advantage of the potential of AI? We have defined
Machine Learning: definition, algorithms and use cases
If you're asking yourself "What is Machine Learning?" or "What are the key concepts behind Machine Learning?", then
