Exercising a Native Intelligence Metric on an Autonomous On-Road Driving System
The intelligence of artificial systems is well quantified by the amount of specified complexity inherent in the system representation, provided we have tools to measure it. Some may generally agree with this claim, but argue that it is simply intractable to successfully and accurately measure the specified complexity of any system, no matter how it was represented. We respond to this important and substantive criticism by performing a computation required by our intelligence metric on an example problem. We have chosen autonomous on-road driving, a problem that has already been solved by “systems” that are known to be both complex and specified, namely, humans. We will begin with a concise statement of the scope of the problem and a summary description of an appropriate system representation approach. We describe generally how to apply a previously published Native Intelligence Metric (NIM) to measure the specification inherent in that representation. We claim that with an appropriate intelligence metric and an appropriate system representation, we can establish an equivalency between 1) the “state of the world” conditions, forming the input to the system, that the system can respond to successfully, 2) the system representation, and 3) the system performance. This equivalency is a potentially powerful result of both the intelligence metric and the system representation approach described in this paper.