Robustness and safety

Explainability sounds like a good idea, but designing the right explanation for the right purpose, the right audience and at the right time requires a deeper understanding of the drivers of explainability, as well as human-centred design methodology used in HCI.

As uncomfortable as this may sound, no algorithmic system can satisfy all definitions of fairness. Choosing the right fairness parameters, and justifying trade-offs such as decreases in accuracy created by debiasing, require multidisciplinary discussions and governance,

AI systems, particularly those deployed by states, must maximise public interest objectives. These objectives include protection of fundamental rights, but may also involve ways for AI to make citizens’ lives better and to protect the planet.

We want AI systems to have high level of accuracy, robustness and cybersecurity. Behind these terms lie a host of complex trade-offs.

The European AI Act requires comprehensive risk management systems for high-risk AI systems, involving data governance, risk assessments, technical documentation, human control and ongoing audits.

Clarifying the roles of actors and power structures in the AI ecosystem