FrameSAI: a Three-Layer Framework to Create Symbiotic AI Systems

As Artificial Intelligence (AI) becomes increasingly widespread in numerous domains of everyday life, humans must maintain their agency, valuing their expertise and judgment and preserving their privacy and safety. The research presented in this article aims at laying the groundwork for a standardized approach for creating systems, that allow a symbiotic relationship between the human and the machine. Specifically, we present a framework, called FRAMework to creatE SAI (FrameSAI), which guides practitioners in developing high-quality SAI systems while complying with law. It is composed of three layers (i.e., Principles, Guidelines, and Success Criteria), which become increasingly operational indications as one moves from one level to the next. A preliminary version of FrameSAI is presented, illustrating a first set of guidelines and success criteria mapped to the four levels of risks of the AI Act, the European Union (EU)’s regulation for the design, development, and deployment of AI systems