Are Many Reactive Agents Better Than a Few Deliberative Ones?

Problem solvers fa l l along a wide spectrum ranging f r om highly del iberat ive to h igh ly reactive. H igh ly del iberat ive systems are able to design op t ima l l y efficient solut ions to problems, bu t they require complete wor ld models and consume inord inate computa t iona l resources. Reactive systems move in real t ime but cannot guarantee efficient solut ions. They are also subject to looping behavior. One way to generate incremental ly more efficient solut ions is to be incremental ly more del iberat ive, e.g., to increase the amount of menta l search between actions. Th is paper presents an alternat ive method for generat ing more efficient solut ions: increasing the number of reactive agents simultaneously a t tack ing a given problem. Th is method provides a second, orthogonal degree of f reedom. We f ind tha t in many domains, increasing agents is dramat ica l ly superior to increasing single-agent del iberat iveness. Th is is because solut ion qual i ty improves rap id ly as more reactive agents are added, but search t ime only increases l inearly. Th is contrasts w i t h add ing more deliberativeness, which incurs exponent ia l ly increasing t ime costs. A m ple empir ical evidence is presented to support our conclusions. 1 I n t r o d u c t i o n Th is paper considers two aspects of computa t iona l prob-