Share

ExpressIF® 3: CEA-List's sovereign explainable AI now possesses planning capabilities

employe-d-une-start-up-examinant-des-graphiques-d-affaires-a-l-aide-d-un-logiciel-d-intelligence-artificielle.jpg
ExpressIF® simulates human reasoning, offers greater transparency and now incorporates intelligent planning to guide strategic decisions. ©DC Studio on Freepik

The third release of ExpressIF® is packed with key features that push business intelligence modeling to new heights. Multi-agent planning is the star of this March 2025 release. Human reasoning in complex and uncertain environments can now be replicated in ExpressIF®.

Smart planning the cornerstone of ExpressIF® 3

ExpressIF®, which could already replicate knowledge-based deductive and combinatorial reasoning, now includes a multi-agent planning engine capable of handling problems that were impossible before, opening up a number of new possibilities. ExpressIF® 3 can orchestrate the management of extremely complex situations to optimize industrial robots’ movements or coordinate emergency response operations involving land vehicles, helicopters, search dogs, and—more recently—autonomous robots, for example.

CEA-List’s symbolic AI is still explainable, but now it can also use reasoning to factor production line speed, energy limitations, and the management of areas shared by robots and human operators to coordinate the cooperation of robots in a factory.

Fuzzy reasoning and active learning

Approximate reasoning is another major update in this third release: It enables ExpressIF® to draw relevant conclusions from imprecise information. The tool can now adapt, reasoning “like an expert” even when the data—be it from sensors or a human report—does not line up. An uncertainty can be modeled and converted into plausible value ranges so that the AI can continue to make robust decisions.

The new release also features active learning. When information is incomplete, the AI requests new experiments to refine its own predictive models—without prompting from the user. This “augmented curiosity” recommends relevant questions and prevents the AI from generating partial answers.

Reorganization around a common core for modular, pragmatic AI

These new features are just the beginning of what ExpressIF® now offers. The new release’s code has been restructured to isolate pure AI components from mathematical and signal processing components. This new organization is reflected in ExpressIF® Foundation, a common core now shared with two specialized versions of the software: ExpressIF® Sensors, for smart sensors, and ExpressIF® Materials, for the design of innovative materials. The active learning capabilities will be particularly useful here.

Among the new features of ExpressIF® Foundation is a graph algorithm library, crucial for applications like routing and molecular representations. With these new features, ExpressIF® 3 is even closer to sovereign, explainable, and modular AI—qualities that are more needed than ever in today’s world.

And, just as with previous releases, ExpressIF® 3 can be integrated into industrial software, embedded on a PCB, or accessed remotely via a web service.

See also

Technological advances

October 21, 2024 | ExpressIF® Materials, a virtual assistant for materials science labs

The current version of ExpressIF® has a new active learning algorithm that even recommends the next experiment
Read more
Success Story

B.Partners, artificial intelligence for traditional small businesses

Startup B.Partners has developed a smart online assistant that helps traditional small businesses and tradespeople grow. Just a few months after the release, users were won over. The secret to the too...
Read more
Use Cases

Smart maintenance

Smart maintenance continuously and automatically monitors industrial processes and equipment so that potential malfunctions can be predicted and prevented.
Read more