April M. Beisaw, RPA

Zooarchaeology and Taphonomy Consulting

James C. Beisaw, Ph.D.

Institute for Systems Research, University of Maryland

Developments in the field of Artificial Intelligence bring the promise of educational web sites past their current hypertext links and towards a truly interactive learning environment. Currently, many archaeology web sites assume a level of knowledge of their likely visitors and cater their content toward that audience. Through intelligent tutoring systems and knowledge bases, the web can be used to guide users along a path of higher learning by storing a model of that user's current knowledge. Knowledge bases can also be used to provide a new form of searchable database that would allow for professionals to conduct "real research" using the Internet. This paper will explore the impact of these technologies on the value of the data that archaeological web sites can present to both the public and professional arenas.

Current forms of Web-based learning resemble electronic text books. Web pages simulate book pages with hyperlinks serving as the glossary and document search features replacing the index.

Choice of content presents a decision for the site designers and sponsors. Most commonly, sites are gauged to either beginner or advanced target audiences. The intermediate user often resorts to sifting through this beginner and advanced content, in search of relevant information. This form of unguided education can lead to misinformation and/or misunderstanding, limiting the educational value of web content. Equally important is the time wasted by a user on educational material which does not meet his/her needs.

Those web sites that offer information at a variety of target audience levels are better able to guide the user through an educational path. Interpretive graphics assist in holding the user's interest while simple electronic tests can help the user gauge progress. However, an incorrect response on one of these tests can prove to be more frustrating that motivational. These sites rarely contain a mechanism for the information that was misunderstood to be explained further. Also, once the information is read by a visitor, there is little reason for a user to return to that site.

Advanced or professional target audiences conducting "research" using the web often find only static data tables and accompanying interpretive text similar to that which is available in standard professional journals. The main benefit of web-based articles is the inherent search-ability and accessibility of electronic texts which speeds research but offers few additional benefits. A truly interactive web-based learning and research environment is possible through the use of technologies developed by the field of artificial intelligence.

Computer-assisted instruction (CAI), the use of computers to present educational material to students and test their understanding of that material, uses software which does little to adapt to individual users. An intelligent tutoring system (ITS), on the other hand, uses artificial intelligence techniques to enhance the educational experience by adapting to a particular learning situation. In particular, the software can adjust its behavior to individual users and track a particular user’s progress and provide the appropriate instructional material.

The adaptability of ITS typically is accomplished by separating knowledge of the subject matter, instructional techniques and individual students. The subject matter is represented in a knowledge base (KB) and the knowledge of a student in a type of KB is referred to as a student (or user) model. Instructional techniques are often represented as condition-action rules though other approaches are possible. This separation of knowledge sources supports a more flexible architecture which allows an ITS to interact with a student in a more appropriate, context-sensitive manner.

ITSs are one type of system which use KBs to represent their knowledge of a topic or subject. A KB can be contrasted to a database (DB) primarily by the conceptual level at which information is represented. A DB represents low-level facts which may be connected through relations between components of those facts (records). Although KBs come in many forms, they share the ability to represent abstract knowledge about a subject. These conceptual abstractions include properties of each record and relations between classes of records. While DBs allow complex and efficient searches on large volumes of data, a key characteristic of KBs is their ability to compactly represent conceptual knowledge and draw inferences from that knowledge.

In an effort to create a system that would serve as both a teaching and a research tool for faunal analysis, we designed and built a faunal knowledge base. Currently, this knowledge base is programmed with knowledge of the taxonomic hierarchy of mammals and the physiology on of a mammalian skeleton. Using the knowledge-based software, this information is then directly linked to descriptions and details of specific animal bones. This specific information can be augmented as required to create a complete catalog of information relevant to zooarchaeology including dental formulas, diets, and habitat ranges.

The power of the faunal knowledge base lies not in the data storage but in their ability to represent conceptual knowledge regarding animal taxonomy and physiology and draw inferences from that knowledge . Using data frames and a rule based system the knowledge base can guide the user to the data that is desired much like an complex flow chart. The frame system maintains the integrity of a parent-child relationship between data to ensure that each data evaluation decision made by the system narrows the possible outcomes until a suggested identification is made.

Frames represent concepts and stereotyped situations while frame instances represent specific objects and situations encountered by the user. More precisely, a frame represents a prototype, the typical member of a class (set) of data, by specifying the properties common to instances of the prototype. All instances of the prototype assume (inherit) its properties.

Frames and their instances consist of a set of slots while each slot consists of a set of facets each of which may have a value. Slots of a frame can be viewed as attribute-value pairs where each pair corresponds to a slot of the frame with each attribute specifying the slot name and its value the slot filler.

The production-system architecture is divided into a control system, production-rule memory and data memory. The control system is a set of mechanisms for coordinating the activities of the system. The production-rule memory contains procedural knowledge consisting of rules for solving problems and performing tasks. The data or working memory is a global database or blackboard containing declarative knowledge consisting of symbolic expressions representing facts or assertions about the current problem.

The production-rule memory contains an unordered collection of rules for solving problems or performing tasks which operate on data memory or interact with the external environment. A typical condition-action rule consists of a goal, some tests to determine if the rule is applicable to the goal and an action to perform. Due to the global nature of data memory, each of these rules has access to all the expressions stored in data memory.

In its role of coordinating the activities of the system, the control system repeatedly executes a basic cycle of three operations. This cycle terminates when a test for some condition of data memory is satisfied. The first operation, match-rules, determines the rules satisfied by the current contents of data memory, producing the conflict set. The second operation, select-rule, uses some criteria to select one rule instance from the conflict set–a process known as conflict resolution. The third operation, apply-rule, applies the selected rule instance to data memory performing the actions indicated. This operation is sometimes referred to as executing or firing the rule. Rule application may change the state of data memory so a new set of rule instances matches on subsequent cycles with the process continuing until the termination condition is reached.

Example 1: Dog Humerus ID - How the knowledge base works with screen captures (3 minutes)

Currently the system is text based, which restricts its usefulness to educating novices about faunal identification; however, the system will ultimately possess a graphical user interface and a photo/line-drawing library that will help guide novices in their pursuits. The taxonomic and skeletal portions of the knowledge base are also independently useful as they can be queried for the complete taxonomic name of a squirrel or used to assist a confused student in locating just which skeletal element contains the Olecranon fossa, for instance.

Once the user interface and additional support software is complete, the zooarchaeology knowledge base can be run remotely via the internet. Combined with a standard "electronic textbook" style of a static web site, users can complete interactive exercises at a variety of experience levels.

Using intelligent tutoring software, an interactive framework can be built around the knowledge base that will guide the user through information based on that user's specific past knowledge. This feature will eliminate the inherent educational boundaries created by static target-audience-specific content.

Additionally, accessing the system remotely, advanced users can conduct research using this system to assist in faunal identifications. Similar knowledge bases can be constructed which will allow for a new breed of artifact and library cataloging. With a knowledge-based system, artifacts can be quickly inventoried on the lowest level of detail and will automatically inherit the general knowledge that goes with the identification.

Example 2: Cataloging a box turtle carapace inherits all information relevant to a box turtle. (3 minutes)

Artifact catalogs and research data can then be queried remotely by other users to test provided hypotheses or collaborate on related research.

The future of web-based education lies in the interactivity and adaptability of the learning process. Using technologies being developed in cognitive sciences such as artificial intelligence, the internet can be used to guide students and teachers alike.

Once our system is fully functional over the internet, there are several AI techniques which could greatly enhance its usefulness as an identification tool. Unfortunately, the resulting automation may detract from its educational value. In particular, the use of scanning and/or computer vision technology (2D and/or 3D) together with neural networks trained to recognize a library of bones could, in principle, automate much of the identification process.

Archaeology enjoys a place of popularity on the internet. A variety of educational archaeology sites are created and maintained by professionals, educators, amateurs, conspiracy theorists, and looters alike. Unfortunately, professional and educational sites are not overly popular among the public for they lack the sensationalism of the "Chariots of the Gods"-like sites and the interactivity of the "how to find and excavate your own treasure-trove" sites.

To bring the excitement of professional archaeology to the public, we must move beyond static text, hyperlinks, and slow downloading graphics. Creating an interactive educational environment, the sense of discovery that lies at the popularity of archaeology can be conveyed to users at a variety of educational levels.