Darryl Reeves

Symbolic Systems 205

Final Paper

Simulation and Behavior

When the names Herbert Simon and Allen Newell are mentioned among those in the field of Computer Science a few terms come to mind. These terms include pioneers, visionaries, Carnegie Mellon professors, great friends, as well as others. Simon’s background was in Political Science (he earned his PhD in the field from The University of Chicago). In addition to this training he also studied economics, advanced mathematics, symbolic logic, and mathematical statistics. These diverse fields fit into his desire to be a “mathematical social scientist.”ⁱ Newell’s path was different but it was equally as diverse. He received a B.S. in Physics from Stanford University in 1949. He went on to study Mathematics at Princeton and logistics and air defense organization at RAND. His studies culminated in a PhD in the field of Industrial Administration from the Carnegie Institute of Technology (now Carnegie Mellon).ⁱⁱ

Even more interesting than their educational backgrounds were their accomplishments. They are both widely viewed as the founders of the field of Artificial Intelligence. Together with JC Shaw, they proposed the idea that computers might be able to use heuristic search to solve problems as humans do. Simon also went on to pioneer the use of quantitative modeling of human behavior in a number of fields. He eventually won the Nobel Prize in Economics for his work on decision-making processes in economics organizations.ⁱⁱⁱ Newell could also boast of an impressive list of accomplishments. Along with his colleagues, he introduced the idea of means-end analysis and planning in problem solving. He collaborated with others to create SOAR, a well-known cognitive architecture. And he also worked towards the creation of a Unified Theory of Cognition, which is a single set of mechanisms that account for all of cognition.^iv

The first chapter of Simon’s The Sciences of the Artificial is titled “The Natural and Artificial Worlds.” In this chapter, Simon explains that the world is composed of natural phenomena and artifacts that are created to adapt the world to the needs of man. His view is that the “artificial” elements of the world are not distinct from the natural elements because the artificial exists as an adaptation to the natural. The functional aspect of artificial things can be described using three categories: the purpose or goal, the character of the artifact (inner environment), and the environment in which the artifact is used (outer environment). The outer environment dictates the design of the inner environment such that the artifact is able to attain a particular goal in the outer environment. Through these details, Simon gives us a better understanding of the role and structure of artifacts in our world.

Because our class is one concerned with simulation, the relation of Simon’s first chapter to the topic of simulation must be explained. The relation of these ideas to simulation is that the inner environment does not have to be designed in a uniform manner in two distinct objects. Distinct objects can perform the same objectives (ie, Simon’s discussion of planes and birds both being able to fly through the air despite having different internal structures) and, therefore, an artifact can be used to simulate some function of a natural object. Simulation of systems that we lack knowledge of is possible because performing a simulation allows us to create simplified models of the system and extract knowledge from the simulation when focusing on certain details of the system rather than the system in totality. This discussion of artificial and natural things is conveniently tied back into Simon’s field of expertise (Artificial Intelligence) when he brings computers and minds into the discussion. Simon states that computers can be used to simulate the mind and give us insights into human behavior before we have a complete understanding of the exact inner working of the brain. He goes on to state that physical symbol systems such as computers and the human mind and brain have the necessary and sufficient means for general intelligent action despite their differences in internal structure.

An interesting question that this discussion of artificial versus natural objects introduces is whether or not the human mind is a natural construct. If one believes the theory that we are all born with a blank slate where our minds are concerned and we attain knowledge throughout life, does our mind gain artificial properties through the process of knowledge acquisition? If we are born with a blank slate, the knowledge that we gain and the skills that we learn are adaptations for our survival and navigation of the world around us. This seems to be a valid interpretation of the mind given Simon’s description of artificial things. Given that the mind (as a physical entity) has not been definitively located, the possibility that it is an artificial construct according to Simon’s definition is quite possible.

Given Simon’s intimate attachment to the field of Artificial Intelligence, his statement that physical symbol systems are capable of general intelligent action just as human minds and brains are, perhaps sheds some lights on Simon’s intuitions about the limitations of replicating human intelligence in computers. Are we to interpret Simon’s belief that physical symbol systems have the necessary and sufficient means for general intelligent action to mean that physical symbolic systems are not necessarily intelligent systems but instead are able to generate intelligent behavior that may not be intelligent in the complete sense that humans are? Simon may be revealing that artificial systems are only capable of “faking” intelligent behavior but not ever achieving it in reality. He clearly states that artifacts only need to operate functionally similar to some aspect of the natural object that the artifacts are intended to simulate. Furthermore, Simon makes it clear that the underlying structure of the artificial object does not need to be the same as the object it is simulating (as flight in birds and airplanes demonstrates). If we take this a step further, not only do physical symbol systems not need to have the same internal structures as brains and minds but they also do not need to fully simulate the process undertaken by brains and minds to generate intelligent behavior. Therefore, it is possible that physical symbol systems do not need to think as humans do in order to simulate their behavior, which would put a serious constraint on AI if this is the best that symbol systems can do.

Simon’s third chapter is entitled “The Psychology of Thinking.” In this chapter, he begins by putting forth the hypothesis that man as a behaving system is simple, and his behavior only appears complex because his outer environment is complex. In order to support this hypothesis, he limits his analysis to cognition and treats it separately from general behavior and includes memory not as part of the man but as part of his environment. One example of humans’ simple behavior is his introduction of a cryptarithmetic problem that is simple for a computer to solve using exhaustive search but a human using the technique of exhaustive search would take years to solve the problem. But when a human uses a technique that combines search and “reason” the problem becomes much more manageable.

The inability for a human to efficiently use exhaustive search to solve the problem leads Simon into a discussion on the limits of human mental abilities with corresponding examples. These limits include the speed of concept attainment, memory capacity, short-term memory, and time necessary to transfer information from short-term to long-term memory. He presents the hypothesis that memory is organized in list structures which allows for the processing of rote learning concept attainment, and natural language. This list structure is effectively used in many tasks that humans undertake. Simon concludes by stating that the commonalities regarding the organization of the human mind exhibited in the psychology experiments demonstrates the simplicity of the system.

Simon’s argument that human behavior is complex as a result of our outer environment, when in fact our behavior is in fact simple, seems problematic. If a man’s behavior seems complex based on his outer environment, doesn’t this make the behavior complex because it is an adaptation to this outer environment (based off of Simon’s first chapter)? It would appear that complex behavior resulting from an adaptation to a complex environment is in fact complex and not simple as Simon’s argument states. If the behavior was not sufficiently complex, it does not seem that it would be able to adapt to the complexities of the outer environment. Furthermore, the decision to extract memory out of human behavior to include it as part of the outer environment does not support the argument because he uses the concept of memory to show how simple human behavior is. This only serves to confuse the argument not strengthen it.

Newell’s “You can't play 20 questions with nature and win: Projective comments on the papers of this symposium” is a critical piece with its focus on the contemporary field of Psychology. Newell feels that psychology is too concerned with isolating certain phenomena in human behavior and refining observations about the phenomena in such a way that issues are never settled. Newell would like to see an effort to draw on previous work in order to combine the ideas to create a unified theory of cognition (which is the title of a book that he would later write). He was concerned with a lack of rigor in psychology in regards to quantitative explanations for phenomena. From Newell’s perspective Psychology’s current path may not lead it to being a mature science on the level of other sciences such as Physics or Chemistry.

But unlike some critics who do not help their cause by failing to offer solutions, Newell does. His first recommendation is to create a control structure within experiments. This is a process where, given a goal and task environment, researchers generate a group of methods that the subjects are likely to use to perform the task and then through the design of the experiment and/or the subsequent analysis figure out definitively what method was actually used. Secondly, more work should be put into putting experimental results together towards a unified theory; too much effort is put into attacking a feature of a certain theory rather than building on the work of others. Lastly, researches should analyze a complex task based on a comprehensive theory with control structures such as mental multiplication or chess or, alternatively, use small experimental tasks but devise a single system to test on the tasks (the methodology behind cognitive architectures).

From my limited experience with Psychology, it appears that Newell’s warnings for the most part have not been heeded. The bulk of the work that I have seen from my Psychology and research experience shows a bias towards isolating phenomena and little work towards unifying the work that exists with new discoveries. Perhaps, it demonstrates the true difficulty of the task that Newell outlines. Even with regards to cognitive architectures, these programs are a positive step in the right direction but still their breadth of discovery seems to be limited by their focus on very specific tasks. It may still be too early to tell if Newell’s premonitions will turn out to be true.

What these readings contribute when taken as a whole is a motivation to study cognitive science/artificial intelligence using simulation. In some ways the knowledge taken away from the work of these two authors is a bit contradictory. Simon puts forth the idea that in order to replicate the human mind we do not need to know all of the details underlying how our mind works and, furthermore, our behavior is not as complex as it would appear. Therefore, simulating human behavior should not be too difficult. On the other hand, Newell is calling for a more rigorous effort towards understanding the inner workings of our behavior and this implies that the details of this behavior are important and should be explored. In both cases, the use of simulation is well suited to support their claims and causes.