Design, Analogy, and Creativity Design, Analogy, and Creativity

ion over Analog and New Model New Model Figure 3: Model-Based Analogy in IDEAL Figure 3 illustrates a portion of IDEAL's computational process that pertains to the learning, reminding and use of GTMs in analogical design. A design problem to IDEAL speci es the functional requirements of the desired device together with behavioral and structural constraints (if any). For example, one design problem presented to IDEAL speci ed the function of an electronic ampli er as the desired function with the additional requirement that the uctuations in the output voltage be small. When a design problem is presented 16 to IDEAL, then, in the problem interpretation step, it rst uses its conceptual knowledge of generic components and substances to elaborate on the problem speci cation. Then, in the reminding step, it classi es the elaborated problem speci cation into its conceptually organized memory of design analogs, and retrieves the best matching analog. This classi cation is based on the functional similarity between the design problem and the stored analogs. In the example of the ampli er, if IDEAL knows of a design for an ampli er that allows large voltage uctuations, then it retrieves this design because of its functional similarity with the desired design. Next, in the design adaptation phase, IDEAL compares the function of the desired design and the function delivered by the retrieved design, and spawns adaptation goals in the form of di erences between the desired function and the function delivered by the retrieved design. In the ampli er example, IDEAL forms the adaptation goal of reducing the voltage uctuations in the retrieved design. The left side of Figure 4 schematically depicts the ampli er design retrieved by IDEAL. It also illustrates a small fraction of its SBF model for the design. In particular, it illustrates the top level of its understanding of the causal behavior of the design that results in the achievement of its function.4 To illustrate the use of analogies in IDEAL, let us consider what happens if the system does not initially know about control mechanisms such as feedback and feedforward. In this knowledge condition, the system searches the SBF model of the retrieved design but fails to localize the cause of large uctuations in the output voltage. Let us suppose that at this stage an (external) oracle provides IDEAL with the design for the desired ampli er along with its SBF model. The right side of Figure 4 illustrates this design and a fraction of a causal behavior in its SBF model. The system stores the new design and its model in its analog memory. But it also compares the SBF models of the old design (that allows large voltage uctuations) and the new design (that regulates the voltage uctuations). In particular, it inspects the structure of the causal behaviors in the 4Brie y, a causal behavior is expressed as a directed graph of behavioral states and state transitions. Each state is expressed as a schema that speci es a substance owing between components, its properties at a speci c component, and the property values. Each state transition too is expressed as a schema that speci es the causes for the transition (e.g., the function of a component) and relations between the properties values in the preceding and succeeding states. 17 i/p Vin o/p Rin + Vout i/p Vin o/p Rin + V’out Rf V(a) A simple amplifier ELECTRICITY loc: i/p voltage: Vin volts ELECTRICITY loc: o/p voltage: Vout volts (b) Behavior "Amplify Electricity" of the Simple Amplifier USING FUNCTION ALLOW electricity of Rin PARAMETER-RELATIONS V-o = f+(Vin) ELECTRICITY loc: Vvoltage: V-o volts USING FUNCTION ALLOW electricity of Op-Amp PARAMETER-RELATIONS Vout = f+(V-o) . . . GIVEN: state1