Summary

In this paper I argue that archaeology is foundationally a science based on cognition since its empirical domain is defined by material objects and relationships which have been affected by intelligent behavior. I also argue that the relationship between cognitive science and archaeology should be an active two way street. Archaeology has much that it can offer cognitive scientists and cognitive science has much that it can offer archaeology.

I then start exploring the kinds of issues that cognitive scientists consider. In particular, I stress the fact that they are interested in describing the processes, structures and mechanisms which underlie intelligent action. This leads to a description of the hierarchy of interrelated concepts that characterize Physical Symbol Systems, which are the systems that Newell and Simon claim underlie all cognition. I narrow the focus of PSSs to discuss information processing and information processing systems because they describe "where the action is" in intelligent systems.

After a quick description of effective procedures, I explore two research problems in cognitive science: visual perception as conceived by David Marr, and problem solving, loosely based on the research program of Newell and Simon. Finally I outline how some of the concepts and research of cognitive science I identified can be applied to problems in archaeology. I briefly discuss early Stone Age tools and the weights of Mohenjo-daro.

Introduction

Cognitive science is an interdisciplinary approach to studying mind, and in particular, intelligent thought and behavior. It is a relatively new academic discipline which is developing from a merger of interests among certain linguists, psychologists, philosophers, computer scientists, anthropologists, neuroscientists, and others (Norman, 1981). Historically, however, cognitive science has not generally been extended to include archaeology or its issues. From the perspective of this cognitive scientist, however, it is obvious that archaeology could become a core cognitive science. The study of material culture is an important domain with unique data and methods which can contribute to the general understanding of intelligence. Also archaeology can (and does) profit from data and methods developed in the other cognitive sciences.

In archaeology, a primary concern is what material culture tells us about the living culture that produced it. The artifacts and structures found at archaeological sites represent varying amounts of skill, knowledge, and social organization. By analyzing these objects in the context of their appearance one may be able to infer a great deal about their role in society and the intelligence that was necessary to create them and to use them. Research in cognitive science has shown that there are many constraints on how people solve problems and achieve other goals (Newell & Simon, 1972; Newell, 1981). Thus a cognitive archaeologist can study the objects and structures found at archaeological sites with an eye toward answering questions about the knowledges, purposes, practices, and skills of the people who produced them. Such research can broaden the data base and theoretical insights of cognitive science as well as add to the knowledge of the peoples which inhabited the archaeological sites.

To take a cognitive scientist's perspective on archaeology one has to think about the cognitive processes involved in producing material culture. An archaeologist who is a cognitive scientist would use the tools and concepts of cognitive science to help implement Lewis Binford's program to link "human activities (i.e. dynamics) to the consequences of those activities that may be apparent in material things (i.e. statics)" (Binford, 1983, p. 19). A cognitive scientist would consider in some detail what it is about the people and the society that can account for the components of the material culture. This would include both the goals and purposes of the members of the community which motivated the organization and sequence of behaviors that created the material culture, and the concomitant intelligence, knowledge, and skill necessary to produce it.

Archaeology is foundationally a discipline rooted in cognition. Material objects or relations among material objects become archaeological data only if they can be shown to exist in their current form or location as a direct or indirect consequence of intelligent behavior. If an object's appearance can be entirely explained by natural (for which read nonintellectual) processes, it is of no direct interest to an archaeologist. For example, some primitively shaped stone tools found in the Olduvai Gorge "might have been dismissed as natural were it not for the numbers that have been found". (Cottrell, 1960). An object however, could be of interest to archaeology even if its structure were entirely shaped by natural processes if it were believed to have been selected for use by humans for some intentional goal. Marcus (1990) found some smooth colored stones which were naturally carved which have archaeological significance. They were found in Olmec burial sites together with other burial objects many miles from where they originated. Although not created by intelligence they were acted upon by intelligence. This fact gives them their archaeological significance.

Concepts in Cognitive Science

The cognitive sciences have not yet coalesced into a single discipline with a unitary perspective on intelligence, but there has been much cross-fertilization of ideas within the special sciences and a growing number of research teams containing members from several disciplines. One methodological principle which is close to universal among researchers who call themselves cognitive scientists is the detailed consideration of the underpinnings of any demonstrated knowledge or behavior. This differs foundationally from the positivistic and behavioristic approaches that many of the sciences previously accepted.

Researchers in cognitive science consider the structures, knowledges, and processes, which underlie the observed behaviors, as well as the behaviors themselves. Cognitive scientists often use these mental concepts in their explanation of these behaviors. The power of a cognitive science approach comes from the idea that intelligent beings have goals and intentions which are implemented by a complex hierarchical system of information transmission. Newell and Simon (1976; Newell, 1981) claim that research through the years gives empirical and theoretical support to the importance of an information processing analysis. They argue that all intelligent beings are Physical Symbol Systems, and being a PSS is necessary and sufficient for producing intelligent behavior.

Newell (1981) identifies a hierarchy of at least five descriptive levels within all PSSs: (a) the device level, (b) the circuit level, (c) the logic level (d) the program level, and (e) the PMS (Processor, Memory, Switch) level. Each of the levels has its own principles and characteristics which are only partially constrained at the other levels.

Researchers studying information processing, which includes many psychologists, computer scientists, and linguists, spend much of their time abstractly characterizing the information flow in the highest two levels of a PSS. The study of information processing systems, ie., systems that receive, store, modify, produce, send, and act upon information (Anderson, 1990), predates the identification of PSSs by a number of years. The mathematical concept of information was first proposed by R.V.L. Hartley in 1928 (Dretske, 1981). Studying information transmission and systems which deal with it gradually developed during the second world war and became well-known not long afterwards (Wiener, 1948; Shannon & Weaver, 1949; Broadbent, 1958). It is only after working with different information processing systems (IPSs) that Newell and Simon attempted to formalize a PSS. The representation of information in terms of structural relations among symbols, the changes in knowledge as computations upon those symbols, and actions as behaviors under the control of symbols is now a long standing tradition (cf. Fodor, 1975; Pylyshyn, 1984; Rapaport, 1990).

An IPS has several components including receptors, memories, processors, and effectors (Newell & Simon, 1972). Such systems receive information from the environment through their receptors. They then go through processes of transforming, storing, comparing and evaluating this information. For example, assume that you see a duck. What happens informationally? In order for you to know that you see a duck, or to be aware that it is a duck you see, you have to compare part of the visual input with some representation of a duck in memory. Processors must parse the visual stimulus into meaningful components in order to isolate the duck from its visual context, and to compare the resultant duck information to a memorial representation. The representation must include not only information concerning the visual appearance of a duck, but also information identifying the visual information to be that of a duck. In order to behave appropriately, therefore intelligently, such as to say "Oh, there's a duck." there has to be a link between your representation of the visual appearance and a representation of the verbal form "duck." In addition, this information has to tie to devices which control the effectors in your vocal apparatus.

The example above informally and globally identifies some of the processes that are involved in doing what is subjectively a simple task. Information processing is involved with describing the processes that are involved in solving problems, telling stories, perceiving objects, understanding sentences, and doing other cognitive things. It is also involved with the goal of understanding the cognitive events that occur. Describing the processes ultimately means specifying an effective procedure (and when truly successful, the effective procedure) which does the task. It should be noted that in many intelligent behaviors stating a procedure which is truly effective is an incredibly difficult task. Not only have many of the computational problems not been solved, but it is not often known what the problems are that need to be solved. Marr (1982) argues persuasively that clearly specifying the computational problems which underlie the intelligent behavior is an important level of analysis. After a computational problem is known one can identify a procedure to solve it.

The concept of 'effective procedure' is one of the more important concepts in the symbolic sciences, and one which is needed at least on an informal basis in order to work within any cognitive science. "An effective procedure is a finite, unambiguous description of a finite set of operations. The operations must be effective in the sense that there is a strictly mechanical procedure for completing them" (Brainerd & Landweber, 1974, pp 1-2). One task for cognitive scientists is to identify intelligent behaviors, and to try to describe effective procedures which produce them. However, it is often not possible to identify exact detailed procedures which will produce the behavior. In these cases the cognitive scientist can try to identify a set of conceptual and behavioral components which when combined could account for the global intelligent behaviors.

Careful analyses of intelligent acts point to whether they are complex or not, and identify the knowledge and skill necessary to perform these acts. These analyses can be supported by several methods including careful study of the details found in the data, the evaluation of similar phenomena, computer simulation, and the attempt to duplicate the phenomena under experimental control and analysis. The more detailed and varied the information cognitive scientists have about intelligent acts, the greater the confidence in their explanation.

Topics in Cognitive Science

Cognitive science is often has a goal of specifying how different cognitive tasks are achieved. In order to do this one needs a clear description both of the end product and of the conditions under which the task occurs. Unless one considers the conditions under which the task occurs the task cannot be accurately evaluated. To make this point, in his Sciences of the Artificial (1969), Herbert Simon points out that the path of an ant over a somewhat rough surface is quite complicated. If, however, the ant's movement is evaluated in conjunction with constraints dictated by details of the surface traversed, it can be seen that the actual task solved by the ant is a relatively simple one. The complexity in this case is in the constraints, not in the information processing system. On the other hand, as can be seen from the description above, the "simple" task of verbally identifying a duck may require quite a complex cognitive system.

If we apply the principle of studying the task needed to be done in conjunction with constraints built into the system, ie. the nature of the visual stimulus and the properties of the human visual apparatus, object and scene perception seems to be a tractable problem. The Gestalt psychologists, Wertheimer, Kohler, and Koffka (Kohler, 1947; Koffka, 1935) were the first researchers who realized that human vision was constrained by something other than some general purpose learning program operating over the distribution of nerve firings as a function of variations of light energy entering the eyes. They identified organizational principles in the perceptual experience (e.g., common fate, good continuation) which were not directly represented by the punctate stimulation.

James Gibson (1950, 1966) noted that when we physically move through the world, although almost all of the local stimulation changes, there are a great many "higher order" invariances in the visual field. For example, contours conserve shape over translation, and objects in the direction of motion conserve shape as size expands. Similarly, there are invariances when rigid objects move, and there are regularities built into how patterns of ambient light are reflected from different kinds of objects. Gibson thought that all we need to do to perceive the world is to "pick up" the appropriate invariances in the visual stimuli. Although he greatly underestimated the difficulty building a perceptual apparatus to respond to these invariances and an information processing system to organize them, he moved a great part of the information processing problem into considering specific mechanisms to account for an identifiable structure in the environment.

Visual perception research now usually starts with constraints based upon regularities found in ambient light reflected off of the surfaces of real objects. No researcher currently believes that the visual system works by simply remembering arbitrary regularities found on the retina. Marr (1982) for example, has a relatively successful theory based on analyzing the computational problem in terms of an analysis of the set of objects we actually perceive in the visual world. He proposed algorithms which could solve the computations, and searched for ways in which the algorithms could be physically realized within the eye and the visual cortex.

Marr's (1982) theory begins with an attempt to identify some of the properties of the visible world which are accessible to the IPS. It tries to identify those visual properties we use to identify the visible properties of objects. It does this instead of attempting to build a visual system that could capture all of the details of the visual stimulus. Visibly, objects consist of surfaces which are seen to be spatially continuous with discontinuities at their boundaries. An important task of the visual system, then, is to identify the boundaries of surfaces. This had been seen to be a difficult computational task because the surfaces themselves often have different markings which are visually just as distinct, or even more distinct, than the differences between surfaces and these have to be overridden.

Marr noticed that surface markings are visible phenomena occurring within small localized regions of the visual field. A broader scope receptive field could ignore such local discontinuities. Marr's theory includes a computational system that incorporates a hierarchical analysis of the visible world. This is a major achievement of the theory. Visual processes are identified which can respond to cues leading to smooth contours on a larger scale while ignoring even large local variations in reflectance.

Using these constraints as an important input into his theory, Marr developed a theory of object perception which has become central to much current research. He showed, for example, that different receptors respond to different sized units so both finer and more global aspects of the visual stimulus can be identified. He suggests that the visual system treats "objects" of approximately the same size equivalently regardless of the internal visual details, and things of the same size are often grouped together to form larger units. Thus a detailed analysis of our perceptual experience accompanied by an analysis of the structure of the input into the perceptual system led to a meaningful informational explanation of how we perceive our visual environment. Without an analysis of the structure of the stimulus input in relation to the perceptual interpretation of that input understanding the cognitive processes may be impossible.

An information processing analysis of problem solving begins with an analysis of the scope of the problem to be solved. What is the knowledge of the organism and what is his situation prior to attempting the problem (the initial state)? What is the knowledge of the organism and what is his situation after the problem has been solved (goal state)? What are the differences between these two conditions? The goal of the analysis is to identify the knowledge and skill required to execute a set of effective procedures that moves the world from the initial state to the goal state.

An analysis of problem solving has at least four components:

Problem solving incorporates the basic components of intelligent behavior. It is motivated goal directed behavior. In order to be able to solve a problem a person must have a goal in mind, and be able to identify and coordinate a sequence of actions which lead towards that goal. This requires being able to organize and use a data base consisting of representational and procedural knowledge. If a successful problem solver does not have such knowledge or skill she must have the ability to attain it from the social and physical environment.

Applications to Archaeology

Early man made these "chopping tools." Making them requires a sequence of activities that is far too complicated to have been achieved by random activity thousands of times. Thus they were created intentionally. Since one needs a tool to make the chopping tool, there must have been a state of technical knowledge between no tools and the shaped chopping tool. That state most likely is the direct use of naturally found objects as tools. It is highly unlikely that a rock would be first used as a tool to make another tool. The idea of using a tool is a cognitive state prior to the idea of using a tool to make a tool. Thus paleolithic man undoubtedly used unshaped natural materials as tools for more direct needs before he used them to shape the chopping tools, and this knowledge must have been communicated, either directly or indirectly.

Making a tool is a problem which was solved within primitive cultures. What cognition did they have in order to solve it? What is the initial state for a paleolithic toolmaker? The individuals who made tools had to know what a chopping tool is and what its properties are. This knowledge logically either has to be creatively invented from observing the value of naturally occurring broken stones or it has to be known within the community. It is highly unlikely that these tools were invented more than a very few times. Thus the community had to have the knowledge and had to communicate it to the toolmaker in one way or another.

One intermediate state is the presence of an appropriate stone to shape in the toolmaker's hands. To reach that state the toolmaker needed to know where to find the stone and how to select one of the right size, and how to determine that it is made of the right material. Again, this could logically be done by trial and error, in which case the toolmaker would have to show a great deal of diligence since only a very few of the tools made would be functional.

A second intermediate state consists of bringing the unshaped stone together with a "hammerstone" (which also had to be selected) to shape it with a plan to make the chopping tool. These tools are made by repeatedly hitting one stone with another. The toolmaker had to learn how to hit them together to produce the cutting edge. This latter is a skill which has to be learned. Whether toolmaking was a common skill learned and practiced by many in the community, or a rare one learned only by a few, it is evidence of intelligent behavior.

Different tools developed over many generations gradually became more regularized and efficient. Some archaeologists have used experimental methods to investigate the difficulty of reaching the different intermediate states in making these tools. One can also do research to find out what these artifacts may be used for. Structural analyses, experiments, and finding similar artifacts in current use among some peoples can help understand their role in the society (Bordaz, 1959). These efforts combined with data on the kinds, quantity, and distribution of artifacts may lead to a detailed account of some of the knowledge different prehistoric peoples attained.

Renfrew makes several assertions from these cubes about cognition of the culture producing them including:

It is unlikely that this culture had the means to measure and cut the different lengths to the second or third decimal point, so the "weight maker" may have been a skilled and knowledgeable artisan who could cut and size the stones to a high degree of precision from experience. He would need to select stones of the right size for shaping and he would have to be skilful in that process. It would take an expert with many thousands of hours of experience to become skilful enough to recognize the right stones and to shape them the right size without a detailed supporting technology1 (cf. Norman, 1982; Frake, 1985). If the stones all came from the same quarry it is possible that a less skilful person could bring back a large number of stones and the artisan could select from among them. Or the artisan may have gone to the quarry to select the appropriate stones either to shape them at the quarry or at a "factory" elsewhere.

There are other possibilities in finding the appropriate stones for shaping. They may have brought a set of weights (standards) to match the stones against so that they were of approximately the right size for shaping and then used the standards to compare the linear dimensions against. They might have developed a set of rulers marked with the appropriate lengths to mark them off on the stones to be shaped into cubes. Either of these techniques require a high degree of planning and preparation. In any of these cases including that of the skilled artisan, if the cubes were really accurate to the third decimal, there would probably have to be a final testing and refining of the cubes by balancing them against one another to get them to be exactly the right size.

From an information processing analysis, we can conclude that at least some individuals in the culture had to spend a great deal of time in the manufacture of the cubes. These individuals had to be highly familiar with the relative linear sizes of the cubes needed as well as of their weight. They had to be skilled at judging the relative weight of "raw" stones. They needed to know what tools to use and how to use them to cut the stones roughly and then to refine the weights by some fine-tuned chipping or sanding process. They had to organize and control the cutting process in order to end up with a object of the right size and shape. Each of the surfaces were close to planar and abutting surfaces were close to being perpendicular to one another. They had to know how to check the weights, probably by balancing one against another on a balance scale, and to refine them when they were not accurate.

The artisans and the community obviously knew that two of one size was equal in weight to one of the next, so they had to be familiar with the concept of doubling. If there were no 3u, 5u, and 7u, etc. weights, we cannot be sure that they understood the process of counting sequentially. Renfrew (1982) suggests that there were 320u and 1600u cubes. Doubling would not give multiples of ten; I do not know the process that would lead to these weights. Without further evidence one probably should not conclude that the larger cubes were derived from the same system.

These people understood that quantity was based on proportional mass rather than proportional linear dimensions, and used that to establish relative value. Processes involving proportional weights imply the concept of a unit of mass. They had to know that two of size u equals one of size 2u, etc., in order to construct the sequence of weights.

It is very likely that these weights played an important role in some exchange system. This exchange system would have to be based on the quantity of substances rather than (or in addition to) the number of items. The system must have been fairly important since the society had to commit significant resources to its production. I do not know what status the weight artisans had, but they had to be highly trained and skilled individuals. They probably attained their degree of mastery by some kind of apprentice program, because they had a lot to learn in order to produce accurate weights. The fine finished form of the weights strongly suggests that accuracy was considered very important. An untrained individual could not simply go out and make the weights.

Conclusion

There is evidence of the mind of man in every archaeological site. Just about every artifact and every edifice, was intentionally created. Almost all of them required detailed knowledge accompanied by a modicum of skill. Studying them in detail from the point of view of what was the cognition that underlay the construction and use of these objects informs us about what the peoples of these communities knew, thought, and did. To the extent that there is minimal variability in a set of equivalent artifacts, such as tools with very similar shapes and sizes, or temples oriented in exactly the same direction, there is evidence of both planning and skill. Cognitive science has been developing concepts and methods that allow us to identify and specify many of the intentions, and cognitive and behavioral processes that went into the planning, design and construction of these objects. These are part of the conceptual world of the ancient peoples, and they can help us understand some of the cognitions and motivations involved in the use of these objects after they were constructed.

Since intentional actions are motivated actions, to the extent that a project required a great deal of time and effort there must have been very important reasons to sustain the effort.

References