Publication
AI Engineering for Trust by Design
André Meyer-Vitali
In: Survey of Tools for Software Engineering, Vol. 2024, No. 1, Pages 20-22, United Innovations, 3/2024.
Abstract
Some of the current problems related to a lack of trust in AI systems are a direct result of the massive use of black-box methods that depend solely on data. Instead, a new AI generation has its foundation built on hybrid AI systems (also known as neuro-symbolic or neuro-explicit). These hybrids do not rely solely on data-driven approaches, but on the full range of AI technologies, which includes symbolic AI methods for search, reasoning, planning, acting and other operations. “Trust by Design” is achieved through the combination of Machine Learning with symbolic conclusions, the explicit representation of knowledge and interaction among agents and humans in hybrid AI systems. Knowledge no longer needs to be learned when it is represented by semantic and other explicit models, which can also guide the learning process in a direction that improves generalisation, robustness, and interpretability. This hybrid approach is also known as the third wave of AI. The requirements are particularly strict when it comes to applications with significant physical, economic or social risk. The AI systems used in such applications are required – for example by the European AI Act – to have been validated and certified according to well-defined criteria.