The Need for General Intelligent Systems
The original goal of artificial intelligence was the design and construction of computational artifacts that combined many cognitive abilities in an integrated system. These entities were intended to have the same intellectual capacity as humans and they were supposed to exhibit their intelligence in a general way across many different domains. I will refer to this research agenda as aimed at the creation of general intelligent systems.
Unfortunately, modern artificial intelligence has largely abandoned this objective, having instead divided into many distinct subfields that care little about generality, intelligence, or even systems. Subfields like computational linguistics, planning, and computer vision focus their attention on specific components that underlie intelligent behavior, but seldom show concern about how they might interact with each other. Subfields like knowledge representation and machine learning focus on idealized tasks like inheritance, classification, and reactive control that ignore the richness and complexity of human intelligence.
The fragmentation of artificial intelligence has taken energy away from efforts on general intelligent systems, but it has led to certain types of progress within each of its subfields. Despite this subdivision into distinct communities, the past decade has seen many applications of AI technology developed and fielded successfully. Yet these systems have a "niche" flavor that differs markedly from those originally envisioned by the field's early researchers. More broadly based applications, such as human-level tutoring systems, flexible and instructable household robots, and believable characters for interactive entertainment, will require that we develop truly integrated intelligent systems rather than continuing to focus on isolated components.
As Newell (1973) argued, "You can't play twenty questions with nature and win." At the time, he was critiquing the strategy of experimental cognitive psychologists, who studied isolated components of human cognition without considering their interaction. However, over the past decade, his statement has become an equally valid criticism of the fragmented nature of AI research. Newell proposed that we move beyond separate phenomena and capabilities to develop complete models of intelligent behavior. Moreover, he believed that we should demonstrate our systems' intelligence on the same range of domains and tasks as handled by humans, and that we should evaluate them in terms of generality and flexibility, rather than success on a single domain. He also viewed artificial intelligence and cognitive psychology as close allies with distinct yet related goals that could benefit greatly from working together. This proposal was linked closely to his notion of a cognitive architecture, an idea that I can best explain by contrasting it with alternative frameworks.
Three Architectural Paradigms
Artificial intelligence has explored three main avenues to the creation of general intelligent systems. Perhaps the most widely known is the multi-agent systems framework (Sycara 1998), which has much in common with traditional approaches to software engineering. In this scheme, one develops distinct modules for different facets of an intelligent system, which then communicate directly with each other. The architecture specifies the inputs/outputs of each module and the protocols for communicating among them, but places no constraints on how each component operates. Indeed, the ability to replace one large-scale module with another equivalent one is viewed as an advantage of this approach, since it lets teams develop them separately and eases their integration.
One disadvantage of the multi-agent systems framework is the need for modules to communicate directly with one another. Another paradigm addresses this issue by having modules read and alter a shared memory of beliefs, goals, and other short-term structures. …