Human-Level Artificial Intelligence Must Be a Science

Abstract

Human-level artificial intelligence (HAI) surely is a special research endeavor in more than one way: The very nature of intelligence is in the first place not entirely clear, there are no criteria commonly agreed upon necessary or sufficient for the ascription of intelligence other than similarity to human performance, there is a lack of clarity concerning how to properly investigate artificial intelligence and how to proceed after the very first steps of implementing an artificially intelligent system, etc. These and similar observations have led some researchers to claim that HAI might not be a science in the normal sense and would require a different approach. Taking a recently published paper by Cassimatis as starting point, I oppose this view, giving arguments why HAI should (and even has to) conform to normal scientific standards and methods, using the approach of psychometric artificial intelligence as one of the main foundations of my position.