Until today, an AI that is at the same time efficient, intelligible, and explainable seems to be hardly conceivable. We would thus still be faced with the dilemma that appeared when AI was founded in 1956:

• opt for an intelligible Symbolic AI (white-box expert system), but not very efficient because its manually created decision-making rules are subject to the limitation of human understanding…
• or opt for a Connectionist AI, which is efficient because it uses machine learning but is not intelligible (black-box neural networks).

This original schism has been overcome. Since 2003, I have proposed and defended Augmented Fuzzy AI (AFAI). It’s a heterodox AI hybridizing continuous fuzzy logic, the theory of Fuzzy Relations of order N [Zalila 1993], and proprietary automatic fuzzy induction algorithms reproducing the human reasoning used in experimental sciences. This AFAI automatically discovers, from reference data, non-linear predictive models defined by IF…THEN rules with continuous logic.

The main advantage of AFAI is that it is both robust and intelligible: the induced models, with high predictive performance, are thus auditable by the business expert and certifiable by the regulator.

AFAI succeeds in achieving what would seem paradoxical: explaining the unexplainable, tacit or implicit knowledge, sensory perceptions or expert decisions taken by instinct or intuition, and even the hidden behavior of black-box models. Direct applications for the design of virtual experts: the sensory evaluation of products, the understanding of the appetite of target customers, the discovery of the optimal formulation of a new product, the early diagnosis of cancers from an epigenetic analysis (25,000 interacting variables), the piloting of autonomous machines, the identification of malicious behavior, etc.

Once validated and certified, you deploy such virtual experts to derive real-time decisions 24/7 in a deterministic and rational manner.

This AFAI, implemented in the xtractis software robots, already validates all the requirements of the 2021 European regulation on Trustworthy AI, which will govern the deployment of critical AI applications. This way, companies & organizations will avoid using black-box models, which would be discriminatory in hiring or granting credit, or whose unstable hidden behavior would damage or injure people or facilities.

In addition, the sovereignty of xtractis would guarantee the digital independence of Europe in a world where the possession of an efficient Trustworthy AI will undoubtedly be a significant strategic asset.