Wednesday, December 11, 2013
I've been hearing from thought leaders all year about how virtualization, machine learning, and M2M traffic are going to solve the enterprise problems that the cloud and Big Data have not yet solved. That's just great. Those things sound terrific in sales pitches to prospects in large convention center halls. I think these thinking machines need to learn some basic vocabulary before they start speaking to us.
MIT's ConceptNet is an attempt at making machines autodidactic. The difference between humans and AIs is that we carbon-based life forms assemble experiential knowledge into frames of reference called common sense. Machines have not yet generated common sense because they don't experience the world the way we do, with five senses collecting rich content. MIT allows humans to input their own interpretations of common sense into the Open Mind Common Sense Project that feeds ConceptNet.
Contextual knowledge is good and adaptive thinking from experience is even better. That's what makes us human. The thing that makes us most human is our ability to reason morally. Humans can discover and apply first principles from a variety of philosophical traditions even if those traditions are irreconcilable. Machines can't do that yet, but giving them the ability to reason adaptively from common sense carries the risk they they will make uncontrolled decisions. The Laws of Robotics are more useful now than ever as a moral foundation for machine learning.
AIs can generate the data used in social network analysis but I do not believe they can actually perform the final analysis themselves. Human analysts must complete the task. I have blogged before that high-level strategic decisions must be left to humans and cannot be automated away with decision management rules. Introducing AIs into routine operations is good for business if humans update the heuristics and rule engines that circumscribe their operations. Humans need training in knowledge management, decision management, Six Sigma, and project management before they launch AIs. Those AIs need diplomas from machine learning schools with common sense as the core curriculum. Train the humans first so they know how to govern the machines. Program the thinking machines with the Laws of Robotics before they begin their machine learning.