The Decision Intelligence Handbook

Chapter 5. Decision Reasoning: The Decision Simulation Process

With AI there are two chasms to cross in turning data into useful insight. The first is whether we can trust machine learning (ML) systems, and that’s increasingly solved. The second challenge is how to go about turning ML predictions into actions—crossing the “knowledge-action” gap. This is why decision intelligence is a critical emerging discipline. DI is every bit as nuanced and challenging as ML was in the first place. It requires integrative approaches, systems thinking, human factors, user interface design, and often multiple ML pipelines working in concert, as well as advanced ML pipeline management, active learning, MLOps, uncertainty identification, AI ethics, and all the other hygiene factors that should always accompany mature AI.

For this reason, DI is hugely transdisciplinary; we don’t yet produce data scientists who are able to navigate the whole stack from data to decisions, and this is why at FDL.ai we always say this process is a “team sport.”

James Parr, Founder, Frontier Development Lab (FDL.ai), SpaceML, and Trillium Technologies

Using a computer simulation can help you to explore many combinations of lever choices and external assumptions to find the right set of actions. This is where “the rubber meets the road”: in addition to using LLMs (including ChatGPT) to help you to build the CDD in the first place, it’s a second place where DI becomes a true collaboration between computers and humans ...