Herbert A. Simon's concepts of bounded rationality and satisficing served as the basis for cognitive architecture. This cognitive architecture is known as FORR (FOr the Right Reasons), which is to learn and solve problems. 

In the early 1990s, FORR was first at the City University of New York. It has many problem-solving tasks, including playing games, robot pathfinding, designing recreational parks, using spoken conversation systems, and tackling NP-hard constraint satisfaction issues.

Furthermore, Gameplay, robot pathfinding, constraint satisfaction issues, park design, and spoken dialogue systems have all seen the application of FORR.

Decision-making process

FORR learns through experience rather than having a complete understanding of how to solve a problem. Intelligent beings don't always make the best choices; instead, they rely on a limited number of plausible justifications and relevant information. Even yet, we can still see these agents as reasonable. Herbert A. Simon, who together with Allen Newell laid the groundwork for the study of cognitive architectures and served as an inspiration for early systems like Soar and ACT-R, proposed the concept of bounded rationality.

Benefits

The foundation of FORR is that there are various justifications or rationales for taking specific actions to solve a problem. These justifications may always be valid (in chess, it is always correct to play a move that will checkmate the opponent) or merely occasionally valid. However, the majority of justifications are not necessarily true. When playing a game, for instance, one good purpose might be to capture pieces, while another good reason might be to control a specific section of the board. These negative factors are known as Advisors in FORR.

Any prospective justification, including probabilistic, logical, or perceptual ones, can be used with the tiered advisor system as long as it provides guidance on which course of action it prefers. Furthermore, FORR can be a connectionist architecture due to its reliance on a group of independent agents (the Advisors).

Architecture and implementation

A FORR architecture consists of three elements: a series of descriptives describing the problem's state, a tiered set of Advisors consulted to determine what action to take, and a behavioural script that queries the Advisors and executes the action they recommend.

A problem domain is a collection of related problems known as problem classes. For example, Tic-tac-toe is a problem class, and a particular tic-tac-toe game is a problem if the problem domain includes playing simple board games. If maze navigation is the problem domain, then a specific maze is the class, and a single attempt to navigate it is an instance. After identifying the problem domain, the development of a FORR architecture for that domain consists of two basic steps: identifying potential appropriate reasons (the Advisors) and determining their weights for a specific class.

Conclusion

FORR is a realistic picture of how human expertise works. Instead of limiting learning to a single method or type of knowledge, the architecture requires that multiple heuristic agents with different kinds of knowledge work together to make decisions. A FORR-based program learns from working with an expert model outside and doing things in its field. 

Furthermore, Forr talks about a way to play games that raises interesting questions about how to organize and change different knowledge and what role experience plays in this kind of learning. FORR's main strengths are its ability to smoothly combine different types of expertise, its ability to learn in many different ways, its ability to handle mistakes made by humans and machines, its ability to degrade gracefully, its openness, and its support for a developmental paradigm.