ULTIMATE

OBJECTIVES
Artificial Intelligence (AI) has entered the business mainstream, opening opportunities to boost productivity and innovation but suffer limitations hindering wider adoption in industrial settings. Both model and data-driven AI approaches naturally complement each other. However, hybrid AI does not fully address the issue of trustworthiness (validity, explainability and ethics). ULTIMATE will pioneer the development of industrial-grade hybrid AI based on 3 stages to ensure trustworthiness and promote the widespread adoption of hybrid AI in  industry.
Main objectives are the following:
· Develop data representation / visualisation models, and propose innovative architectures to construct and train hybrid AI algorithms.
· Design rigorous evaluation methodologies with appropriate properties (e.g. accuracy. robustness, safety) to strengthen their trustworthiness.
· Implement the developed hybrid AI algorithms under operational conditions (Robotics and Space) and fully assess them.
· Ensure the ethical compliance and trustworthiness through qualitative / quantitative approaches.

EXPECTED IMPACT
IMPACT 1: advance the current knowledge on the design, development, and deployment of production-grade hybrid AI and on rigorous  evaluation methodologies (e.g. confidence estimation methods) to significantly increase the trustworthiness.
IMPACT 2: go beyond some existing standards as a reference in AI solutions to meet industrial requirements (related to safety for instance) to  over AI systems trustworthiness more adequately including social and ethical issues.
IMPACT 3: ensure that AI development and implementation is human-centric and is a force for good in society whilst evaluating the  consequences taking into account the criteria of people (compliance with appropriate legal, ethical and societal foundations) and the  achine’s criteria.
IMPACT 4: support the creation of highquality jobs where humans are making informed decisions using AI outputs rather than simply  executing tasks they do not understand.