Generality is a precondition to what we mean when we say that an automated problem-solving method is “routine.” Once the generality of a method is established, “routineness” means that relatively little human effort is required to get the method to successfully handle new problems within a particular domain and to successfully handle new problems from a different domain. The ease of making the transition to new problems lies at the heart of what we mean by “routine.”
Referring to the discussion of the Deep Blue chess-playing system and the Chinook checker-playing system, what fraction of Deep Blue’s and Chinook’s highly specialized software, hardware, databases, and evaluation techniques can be brought to bear on different games? For example, can Deep Blue’s massive parallel state-space search or Chinook’s three-way decomposition be gainfully applied to a game, such as Go, with a significantly larger number of possible alternative moves at each point in the game? What fraction of these systems can be applied to a game of incomplete information, such as bridge? What more broadly applicable principles are embodied in these two systems? For example, what fraction of these methodologies can be applied to the problem of getting a robot to mop the floor of an obstacle-laden room? Correctly recognizing images or patterns? Devising an algorithm to solve a mathematical problem? Automatically synthesizing a complex structure?
A problem-solving method cannot be considered routine if its executional steps must be substantially augmented, deleted, rearranged, reworked, or customized by the human user for each new problem.
The 2003 book Genetic Programming IV: Routine Human-Competitive Machine Intelligence demonstrates the generality of genetic programming by solving illustrative problems from several fields, including
· analog electrical circuits (including six post-2000 patented circuits),
· placement and routing of circuits,
· genetic networks, and
· metabolic pathways.
Our previous publications (and previous publications by others) additionally demonstrate that genetic programming is capable of solving problems in numerous other areas.
The bright line distinction between that which is delivered by genetic programming and that which is supplied by the intelligent human user (made in connection with the definition of the “AI ratio” and what we mean by “high-return”) helps make it clear that genetic programming is a systematic general problem-solving method.
As will be seen in examining our previously published papers and books (and the work of others), relatively little effort is required to make the transition to new problems within a particular domain or to new problems from an entirely different domain.
For example, after discussing the first problem of automatically synthesizing both the topology and tuning of a controller in chapter 3 of the 2003 book Genetic Programming IV: Routine Human-Competitive Machine Intelligence, the transition to each subsequent problem of controller synthesis in that chapter mainly involves providing genetic programming with a different specification of what needs to be done¾that is, a different fitness measure. Because virtually all controllers are built from the same primitive ingredients (e.g., integrators, differentiators, gains, adders, subtractors, and signals representing the plant output and the reference signal), additional problems of controller synthesis can be handled merely by changing the statement of what needs to be done.
Similarly, after discussing the first problem of automatically synthesizing both the topology and sizing of an analog electrical circuit in chapter 4 of the 2003 book Genetic Programming IV: Routine Human-Competitive Machine Intelligence, the transition to each subsequent problem of circuit design in that chapter mainly involves providing genetic programming with a different specification of what needs to be done.
The routineness of the transition from problem to problem is especially clear in chapter 15 of of the 2003 book Genetic Programming IV: Routine Human-Competitive Machine Intelligence involving six circuits that were patented after January 1, 2000. All six problems were run consecutively over a period of about two months intentionally using the very same computer, the very same software, and the very same settings of the minor control parameters. All six circuits were composed of the workhorse ingredients of present-day electronics (i.e., resistors, capacitors, and transistors). As we move from one problem to the next in that chapter, the only substantial change is the specification of what needs to be done. This specification is based on each inventor’s statement of performance of each patented circuit. As stated in the 1992 book Genetic Programming: On the Programming of Computers by Means of Natural Selection, “Structure arises from fitness.”
The transition from one problem domain to another becomes especially clear by comparing the work concerning the automatic synthesis of controllers, analog electrical circuits, antennas, genetic networks, and networks of chemical reactions in the 2003 book Genetic Programming IV: Routine Human-Competitive Machine Intelligence. For example, in making the transition from problems of automatic synthesis of controllers to problems of automatic synthesis of circuits, the primitive ingredients change from integrators, differentiators, gains, adders, subtractors, and the like to transistors, resistors, capacitors, and the like. The fitness measure changes from one that minimizes a controller’s integral of time-weighted absolute error, minimizes overshoot, and maximizes disturbance rejection to one that is based on electrical considerations such as the circuit’s amplification, elimination of distortion, suppression or passage of a signal, and the like. In making the transition from problems of automatic synthesis of circuits to problems of automatic synthesis of networks of chemical reactions (metabolic pathways), the primitive ingredients change to functions that represent chemical reactions that consume chemical substrates (inputs to chemical reactions) and produce reaction products (outputs), at certain rates, in the presence of certain catalysts (enzymes). The fitness measure compares the quantity of product that is produced by a candidate network to the observed data.
Of course, although the preparatory steps change from one problem to another and from one domain to another, the main executional steps of genetic programming (i.e., the flowchart of genetic programming) remain unchanged.
· The home page of Genetic Programming Inc. at www.genetic-programming.com.
· For information about the field of genetic programming in general, visit www.genetic-programming.org
· Information about the 1992 book Genetic Programming: On the Programming of Computers by Means of Natural Selection, the 1994 book Genetic Programming II: Automatic Discovery of Reusable Programs, the 1999 book Genetic Programming III: Darwinian Invention and Problem Solving, and the 2003 book Genetic Programming IV: Routine Human-Competitive Machine Intelligence. Click here to read chapter 1 of Genetic Programming IV book in PDF format.
· For information on 3,198 papers (many on-line) on genetic programming (as of June 27, 2003) by over 900 authors, see William Langdon’s bibliography on genetic programming.
· For information on the Genetic Programming and Evolvable Machines journal published by Kluwer Academic Publishers
· For information on the Genetic Programming book series from Kluwer Academic Publishers, see the Call For Book Proposals
· For information about the annual Genetic and Evolutionary Computation (GECCO) conference (which includes the annual GP conference) to be held on June 26–30, 2004 (Saturday – Wednesday) in Seattle and its sponsoring organization, the International Society for Genetic and Evolutionary Computation (ISGEC). For information about the annual NASA/DoD Conference on Evolvable Hardware Conference (EH) to be held on June 24-26 (Thursday-Saturday), 2004 in Seattle. For information about the annual Euro-Genetic-Programming Conference to be held on April 5-7, 2004 (Monday – Wednesday) at the University of Coimbra in Coimbra Portugal.
Last updated on August 27, 2003