Traditionally computers have not been very good at making decisions when faced with incomplete information, risk, or incredible complexity. Instead we often rely on human intuition to inform these decisions when there is no clear answer and in some cases this can actually be an incredibly effective way to make decisions.
Often you will hear a story of a fire fighter knowing the way to go based on a gut feeling, a batter knowing when to swing, or even a chef knowing exactly what to add to a recipe. You may ask yourself “how do they do that?” In all three of these cases people are relying on heuristics to inform their decision. A heuristic is a basic rule of thumb that lets us take a large complex and seemingly impossible problem and substitute it for a much easier problem that we can answer. As humans we rely on many different heuristics to inform decisions we make and we are often not aware of them and just attribute it to “intuition”. As we learn more and gain experience we can hone our heuristics and become better decision makers when facing specific types of problems. The way we decide to consciously or subconsciously adjust our heuristics is called a meta-heuristic (applying a heuristic to our heuristics).
So what happens if we try to make a machine that is inspired by the way humans make their decisions when the problem is to big? Well we apply a heuristic algorithm of course! Heuristic algorithms are often applied to tricky problems to make them a little easier, but they traditionally can only be applied to certain types of problems leading a to a robot brain that can handle a big problem, but only big problems it has seen before. However, if we were to combine fast heuristic algorithms and calculations with slow strategic processes into a structure that selects the right type of approach for the problem, then a new better robot brain would emerge. This robot brain can handle uncertainty, risk, and complexity incredibly efficiently and while it won’t always make the best decision, it is capable of making really good decisions under the most challenging conditions. This approach is inspired by human cognition and the modeling of how our brains function which is called computational cognition. A robot brain that applies this approach is called a computationally cognitive agent or a comp cog agent for short.
In my dissertation titled “Autonomous Decision Making Facing Uncertainty, Risk, and Complexity” I explore this problem and then develop a comp cog agent capable of making incredible decisions under challenging circumstances. You can read my dissertation by clicking the link below.