Building trust to integrate AI into Air Liquide’s critical businesses

Building trust to integrate AI into Air Liquide’s critical businesses

13 Dec 2024

At Air Liquide, the adoption of tools and systems based on artificial intelligence requires a high level of trust. From AI risk management to regulatory compliance in sensitive sectors such as home healthcare, from user acculturation to consideration of societal and environmental issues, the Confiance.ai program has helped lay the foundations for controlled and responsible adoption.

 

AI adoption issues specific to Air Liquide

At Air Liquide, the adoption of AI requires a high level of trust before any deployment:

  • AI risk management with an appropriate strategy taking into account data protection, uncertainties in predictions, robustness, transparency
  • Consideration of compliance and regulatory issues for sensitive businesses such as home healthcare, and issues for critical systems such as air separation plants.
  • Promote acceptability in the business, with the necessary level of user acculturation and consideration of societal and environmental issues.

 

Products and services affected by the Confiance.ai program

The results of Confiance.ai will eventually have an impact on all Air Liquide’s businesses, in particular :

  • Primary production, through better management of our plants, optimizing operations and maintenance. The aim is to increase safety while guaranteeing correct operation of AI-based systems
  • Energy transition, with trustworthy AI tools for monitoring energy efficiency to prevent over-consumption, and forecasting models for optimal sourcing of renewable energy to help decarbonize our plants
  • Customer service with a supply chain streamlined by AI tools to optimize customer delivery rounds, prevent customer attrition, prevent risks to chronic patients in our homecare programs.

 

Confiance.ai’s contribution to Air Liquide’s strategy and challenges

Confiance.ai has contributed to the implementation of a tool-based and holistic methodology taking into account not only the technological dimension, but also the means of verification and validation of AI tools. This methodology, together with the trustworthy AI technological building blocks, forms an essential foundation for Air Liquide in the design, implementation, deployment and adoption of AI-based tools and systems. The results of Confiance.ai also constitute an important gas pedal for digital transformation at Air Liquide and the compliance of our AI systems with the new AI act regulations. Finally, The Body Of Knowledge resulting from Confiance.ai’s work has been a valuable source for developing the Air Liquide Group’s AI governance policy.

 

Competitive advantages developed with Confiance.ai

While it’s difficult in the short term to accurately assess the impact and benefits of Confiance.ai for Air Liquide, we see this program as a precursor that will boost the deployment and adoption of AI in our businesses. Here are just a few of the competitive advantages that can be activated in our industry thanks to the deployment of safe and reliable AI systems:

  • Improved maintenance, steering and asset management with better predictions enabling cost-optimized and effective decision-making
  • Optimization of the supply chain from production sites to storage and distribution at customer sites, improving customer experience, operational performance and carbon footprint
  • Improved customer and patient knowledge for a better understanding of needs, behaviors and parameters of consumption and health protocols, which can facilitate the prediction and improvement of customer services and better patient care.

Challenges and conditions for success for the next phase of the Confiance.ai program

It is important to continue this program in order to :

  • Consolidate results, make technological building blocks sustainable and industrialize them
  • Develop new trust tools for new generative AI technologies.

 

Example of current or future implementation/deployment

To optimize the supply chain, our R&D teams have developed a computer vision-based AI solution that automatically counts the bottles on a truck entering and leaving a packaging center. The system consists of a camera attached to a gantry crane to capture video footage of trucks passing by, an AI for bottle detection and counting, and a visualization interface. Four technological bricks have been implemented as part of the Confiance.ai program and tested to enable:

  • Provide guarantees on counting uncertainty
  • Monitor and alert in the event of deviations
  • Make the AI counting system robust to adverse conditions, e.g. rain, day/night, localize areas to be recounted using interpretability techniques. This solution, deployed in two gas cylinder filling centers, is currently being tested at Air Liquide Brazil and US.

Accelerating AI Act compliance with the Confiance.ai program

Some of our activities, such as home healthcare and production asset management, may require compliance with the AI Act. Technological building blocks such as the explicability of AI models will serve to bring greater transparency. Likewise, anonymization solutions will provide greater guarantees for the protection of people, assets…