In this first episode, we discussed trustworthy AI, human-machine collaboration, flexibility, trustworthiness and user acceptance of AI solutions.

In the context of the PERKS project, the first of a series of webinars, was hosted on 24 October 2024, about the topic ‘Human-centred and trustworthy AI’. The event brought together four experts in AI and human-centred design to discuss the challenges and opportunities of creating trustworthy and human-centred AI systems.

Asbjorn Folstad, from SINTEF, Ricardo Bessa, from INESC TEC, and Matteo Falsetta, from GFT Technologies, were the speakers for the three projects: THEMIS 5.0, AI4REALNET and HUMAINE, respectively. The webinar, led and moderated by Irene Celino, PERKS project coordinator, saw participants engaged in discussing about the importance of designing AI systems that align with user needs and expectations. A summary of the key aspects and insights coming from the discussion is as follows:

Understanding Human-centred AI: a truly human-centred AI focuses on empowering users rather than replacing them. As Ricardo Bessa noted, AI systems must be context-aware, adapting to specific user goals and operational environments to ensure relevance and minimize negative impacts. This requires co-designing systems with users, recognizing that their knowledge, preferences, and expertise are essential inputs throughout the AI lifecycle.

A recurring theme was then the concept of augmentation over automation. The speakers collectively define AI as a partner, enabling humans to delegate repetitive or low-value tasks, while focusing their efforts on strategic, high-impact activities. Matteo Falsetta from HUMAINE highlighted the potential of AI to improve the quality of life by enabling individuals to dedicate more time to creative and cognitive pursuits rather than mundane operations.

Finally, Asbjørn Følstad added that making AI human-centred involves identifying and mitigating unintentional consequences, which often require user involvement at every stage of development. This includes anticipating different perspectives in a given context.

The Role of Technology: innovative technologies emerged as enablers of human-AI collaboration: indeed, tools like knowledge graphs, explainable AI, and advanced learning paradigms were identified as useful in a human-machine interactive process.

The THEMIS 5.0 project uses, for example, frameworks that combine technical robustness with contextual adaptability, helping stakeholders understand how AI interacts within complex systems like those applied in healthcare. On the other hand, HUMAINE focuses on neuro-symbolic AI in order to provide dynamic, transparent decision-making support.

Finally, Ricardo Bessa also brought attention to innovative visualization techniques such as hypervision in AI4REALNET. All the speakers agree that all these innovative solutions are helpful toward designing AI technologies that are intuitive, inclusive, and accessible.

Trust in AI systems: trust was at the heart of the discussions, particularly the distinction between trust (a user’s perception) and trustworthiness. Speakers explored strategies to align these dimensions. The THEMIS 5.0 framework was cited for its contextual approach to trustworthiness, encompassing factors like privacy, safety, and fairness. The AI4REALNET project emphasizes the importance of transparency, proposing that AI systems should actively explain their limitations to the users. Such practices, coupled with measurable KPIs for resilience and robustness, provide a framework for managing expectations and enhancing reliability.

Moreover, both HUMAINE and THEMIS 5.0 highlighted the interpretability: the speakers agree about value of white-box models—AI systems designed to be transparent and explainable in their decision-making processes.

Ethical and legal aspects: the final key aspect was related to the ethical dimension of AI development. Legal frameworks like the EU AI Act were recognized not only as requirements but as tools for social and human rights. All the speakers highlighted the need for AI to respect privacy, avoid bias, and promote fairness.

The speakers reported some final considerations about the challenges and opportunities for human-centred and trustworthy AI:

  • User Involvement: encourage users to actively participate in the design and evaluation of AI systems. This involvement enhances both user confidence and security in the daily use of solutions.
  • Systemic Design: Integrate AI into workflows, data systems and business solutions.
  • Expectation Management: consider the need to manage the lack of trust in AI-based solutions by maintaining a realistic and transparent approach w.r.t. their capabilities and limitations.
  • Ethical aspects: establish ethics as a standard to be adopted in the design of AI-based solutions.

The speakers, representing different projects, show how collaborative efforts can lead to high-impact, human-friendly AI solutions. Moreover, the entire discussion and the covered topics confirm the value and relevance of the research and innovation actions carried out by PERKS in the context of procedural knowledge management in industry. The recording of this webinar is available at this link and the introductory slides are available at this link.

With the continuation of the series, the PERKS project aims to continue the discussion with other European research and innovation projects on the most advanced topics related to AI, data and robotics for industry optimization.