It brings agility and flexibility to the projects as well as to the issues addressed: from a few weeks to a maximum of 6 months for a subject, managed in a dynamic way (by a "Think Big, Start Small & Fail fast" approach).
Based on an innovative collaborative model, the Data Innovation Lab began with an incubation period which helped strengthen the methodology and practices offered. The Lab worked on use cases with the Group's business units. Each year, through the projects associated with it, the DIL defines new subjects and new challenges, such as work on new machine learning methods (deep learning, reinforcement learning, transfer learning, etc.) which are promising in AI.
Transcription
Artificial Intelligence at the Heart of the Data Innovation Lab!
Energy consumption, ethics and sovereignty: how can we address the challenges raised by AI? In this video, discover the Data Innovation Lab, where data scientists and business experts use AI technologies to tackle these issues.
(Duration: 2:56)
Your browser does not support javascript.
To enable you to access the information, we suggest you view the video Artificial Intelligence at the Heart of the Data Innovation Lab! in a new tab.
[In this report, Didier Roustan, Programme Director at EDF R&D, presents the work carried out by the Data Innovation Lab on artificial intelligence. He discusses the potential applications of generative AI, its limitations, energy consumption and sovereignty challenges.]
Didier Roustan: There is no need to point out that advances in artificial intelligence continue to amaze us every single day. New open-source models are being released daily. Our role at EDF R&D is to test them and determine which use cases they are best suited for.
Here at the Data Innovation Lab, we bring together data scientists and business experts to tackle problems that can be solved using AI.
Big Data!
Didier Roustan: At EDF R&D, we use a wide range of artificial intelligence techniques. It is true that generative AI is receiving a great deal of attention at the moment. Its ability to generate natural-language text, images and videos is impressive. Business units have embraced these solutions because they are accessible and easy to implement.
We assess open-source solutions hosted on our own servers. Sensitive documents must never be sent for translation or summarisation to a solution whose origin and hosting environment are unknown.
Are generative AI systems energy-intensive?
Didier Roustan: The digital sector is expected to experience very strong growth in the coming years. RTE (the French electricity transmission system operator) forecasts a medium-term increase of 15 terawatt-hours in the sector's electricity consumption, equivalent to the output of two nuclear reactor units. In the United States, figures as high as 200 terawatt-hours are being discussed, which is considerable.
Why are we seeing such astronomical figures? ChatGPT's energy consumption is estimated at 500 megawatt-hours per day, equivalent to the consumption of a city of 40,000 people. And that is nothing compared with the energy required to train ChatGPT in the first place, since these solutions need vast amounts of data to learn.
These energy requirements, along with ethical and sovereignty issues, are major concerns for us. We need to develop energy-efficient digital solutions.
Can generative AI be trusted?
Didier Roustan: In fact, we tend to talk about generative AI hallucinations. These systems have a tendency to make up answers that have no basis in reality. Beyond the humorous side of it, this raises questions about how generative AI is used, as it can produce answers that seem credible while citing sources that do not exist.
A Canadian airline was ordered to refund a ticket to a customer, despite the customer service department's refusal, simply because the chatbot had previously agreed to the refund.
EDF R&D is working on tools and methods to monitor and control AI behaviour. In short, AI systems that monitor other AI systems.
Find out more at edf.fr
The DIL's players in EDF R&D
-
R&D Data Scientists
- Data science expertise
- Machine Learning
- Deep Learning
- Data visualisation…
-
Research Apprentice
- Background in mathematics and/or statistics
-
Partners (academia, start-ups, etc.)
- Specific expertise or resources complementing those available at R&D
-
Information Systems Department
- Access to data or business platforms
-
Business Lead
- Defining business needs
- Business expertise
- Business model
The Data Innovation Lab provides methods, tools and expertise to researchers looking to leverage data science to unlock the value of all the data generated by the company.