Moving beyond the machine learning paradigm towards more flexible AI is likely to involve coupling the strengths of ML with the strengths of other approaches to AI. The aim here is to move towards the kind of intelligence displayed by human beings, where learning happens without vast data resources, without intensive training, at low computational cost.15 In addition, humans gain knowledge in a way that allows them to use “common sense”, and to transfer knowledge and experience between domains by representing data in compact hierarchical structures based on concepts and their relationships.16
As our survey results made clear, replicating human level intelligence (often referred to as strong AI, or Artificial General Intelligence (AGI)) remains a distant goal, but even small steps in this direction will open up a host of transformative applications. One approach with potential is Symbolic AI, which has the advantages of being adaptable to context, and having a degree of transparency, allowing us to understand, validate and live comfortably with AI-sourced decisions, whether in healthcare, the judicial system, workplace recruitment or other domains.
