TRUST-AI proposes to advance the frontier of Explainable AI

A new FET project started in October working on a new model of Artificial Intellingence to increase trust in AI outcomes

Artificial Intelligence’s massive potential will only be realized if humans are able to trust the outcome of AI models, and use their own expertise to validate and improve those models. This is virtually impossible with the current black-box models, such as neural networks or tree ensembles. These models are composed of many hundreds or even thousands of nodes, with finely calibrated weights, which cannot be explained.

 

The next generation of AI will therefore have to improve explainability. The approach of the FET project TRUST-AI, started in October 2020, consists in involving human intelligence in the discovery process, resulting in AI and humans working in concert to find better solutions (i.e. models that are effective, comprehensible and generalisable). This is made possible by employing ‘explainable-by-design’ symbolic models and learning algorithms, and by adopting a human-centric, ‘guided empirical’ learning process, using theory and insight from humans to guide the machine’s empirical search. By searching for the ‘right’ structure of the AI models, not just for their parameters, it should be possible to find comprehensible models and avoid unnecessary weights.

 

The resulting tool is TRUST ⎼ a Transparent, Reliable and Unbiased Smart Tool. By being fundamentally explainable, it will be transparent, reliable and able to prevent undesirable biases.  TRUST will be developed and demonstrated in three use cases: one in healthcare, for the treatment of paraganglioma, one in online retail, when selecting delivery time slots, and another one in energy, aiming at demand forecast of individual buildings. These use cases will validate TRUST, but its true potential should be exploited with further applications, in a wide range of sectors, including banking, insurance, manufacturing and public administration.

 

Background information

FET-Open and FET Proactive are now part of the Enhanced European Innovation Council (EIC) Pilot (specifically the Pathfinder), the new home for deep-tech research and innovation in Horizon 2020, the EU funding programme for research and innovation.

 

Photo by Hitesh Choudhary on Unsplash