Toward Human-Level Machine Intelligence
Lotfi A. Zadeh
Achievement of human-level machine intelligence has profound implications for modern society--a society which is becoming increasingly infocentric in its quest for efficiency, convenience and enhancement of quality of life.
Humans have many remarkable capabilities. Among them a capability that stands out in importance is the human ability to perform a wide variety of physical and mental tasks without any measurements and any computations, based on perceptions of distance, speed, direction, intent, likelihood and other attributes of physical and mental objects. A familiar example is driving a car in city traffic. Mechanization of this ability is a challenging objective of machine intelligence.
Science deals not with reality but with models of reality. In large measure, models of reality in scientific theories are based on classical, Aristotelian, bivalent logic. The brilliant successes of science are visible to all. But when we take a closer look, what we see is that alongside the brilliant successes there are areas where achievement of human-level machine intelligence is still a distant objective. We cannot write programs that can summarize a book. We cannot automate driving a car in heavy city traffic. And we are far from being able to construct systems which can understand natural language.
Why is the achievement of human-level machine intelligence a distant objective? What is widely unrecognized is that one of the principal reasons is the fundamental conflict between the precision of bivalent logic and imprecision of the real world.
In the world of bivalent logic, every proposition is either true or false, with no shades of truth allowed. In the real world, as perceived by humans, most propositions are true to a degree. Humans have a remarkable capability to reason and make rational decisions in an environment of imprecision, uncertainty, incompleteness of information and partiality of truth. It is this capability that is beyond the reach of bivalent logic--a logic which is intolerant of imprecision and partial truth.
A much better fit to the real world is fuzzy logic. In fuzzy logic, everything is or is allowed to be graduated, that is, be a matter of degree or, equivalently, fuzzy. Furthermore, in fuzzy logic everything is or is allowed to be granulated, with a granule being a clump of elements drawn together by indistinguishability, similarity, proximity or functionality. Graduation and granulation play key roles in the ways in which humans deal with complexity and imprecision. In this connection, it should be noted that, in large measure, fuzzy logic is inspired by the ways in which humans deal with complexity, imprecision and partiality of truth. It is in this sense that fuzzy logic is human-centric.
In coming years, fuzzy logic and fuzzy-logic-based methods are likely to play increasingly important roles in achievement of human-level machine intelligence. In addition, soft computing is certain to grow in visibility and importance. Basically, soft computing is a coalition of methodologies which in one way or another are directed at the development of better models of reality, human reasoning, risk assessment and decision making. This is the primary motivation for soft computing--a coalition of fuzzy logic, neurocomputing, evolutionary computing, probabilistic computing and machine learning. The guiding principle of soft computing is that, in general, better results can be achieved through the use of constituent methodologies of soft computing in combination rather than in a stand-alone mode.