Development of Integrated Criminal Justice Expert System Applications


Kevin J. Lynch, M.S.,
Artificial Intelligence Group
Management Information Systems Department
University of Arizona
Tucson, Arizona

Frank J. Rodgers, B.S.
Latent Print Section Supervisor
Laboratory Bureau
Phoenix Police Department
Phoenix, Arizona


Introduction

In April of 1988, the Phoenix Police Department acquired twenty million dollars of bond funds to upgrade its information processing capabilities over the next five years. Enhancements include additional systems for computer-aided dispatch, computerized record management, and automated fingerprint identification. On a national and international level, the use of computer technology in the criminal justice system today is characterized by a large number of complex systems, large data stores with redundant and sometimes conflicting information of varying quality [13,14], little integration among systems, and poor support of the life-cycle from the initial crime report to final adjudication. "Expert system" technology has recently become available from the field of artificial intelligence that would enable `intelligent' access and use of existing data stores to greatly increase the effectiveness of the law enforcement community. Expert systems are being developed for burglary [16], murder [26,28], forensic science [27], and hostage-taking incidents [29]. The FBI considers artificial intelligence techniques to be extremely valuable for effective law enforcement through the next decade [3,30]. To fully utilize this new technology, an architecture is necessary to define an integrated, expandable structure that capitalizes on the significant investment in existing systems and technologies.

Existing Systems

The Phoenix Police Department is either using or installing a number of criminal justice information systems (CJIS). The summaries below give examples of the types of systems that might be integrated for substantial benefit to the law enforcement community:

CAPRI - Computer-Aided Police Records Index - manages criminal history information at a local and regional level. Queries can be initiated on-site, from remote terminals, and from in-vehicle mobile data terminals. Data includes criminal histories, incident report references, name indexes, fingerprint classifications, personal identification numbers, subject demographic information and violence potential [23].

CAPRI is the forerunner of PACE, described below.

PACE - Police Automated Computer Entry - evolved from the recognition that a growing burden of paperwork was threatening the efficiency and effectiveness of the Phoenix Police Department. PACE will automate the access and update of offense reports, supplements, arrest records, juvenile referrals, and field interrogation cards. PACE is anticipated to reduce patrol officers' report-writing time by 50%, and eliminate data redundancy existing in the current disjunct systems [24].

AFIS - Automated Fingerprint Identification System - provides automated searches of 10-print cards and latent fingerprints and generates a ranked candidate list. Unsolved latents can be entered and automatically searched against new 10-print cards [22].

WIN - Western Identification Network - is a regional network designed to access AFIS systems in eleven western states (Alaska, Arizona, California, Colorado, Idaho, Montana, Nevada, Oregon, Utah, Washington, Wyoming) [9].

NCIC - National Crime Information Center - a national information system which uses a telecommunications network to store and retrieve wanted person and and stolen item data (vehicles, guns, license plates, securities).

Expert System Technology

Historically, criminal justice computer automation has focused on tasks such as storage and retrieval of criminal histories, modus operandi, and fingerprint information - problems addressed by deterministic, algorithmic solutions. Much more difficult is the computerized solution of problems that require considerable human expertise, i.e., forensic investigation. Expert systems, frequently termed "knowledge-based" systems, address this important class of problems. Expert systems are computer programs that perform at the level of a human expert, typically in a `narrow' field. For instance, an expert system developed for the police department assists detectives in identifying possible burglary suspects [25]. One important feature of an expert system is its ability to provide users with an explanation of the reasoning employed to reach a conclusion. In addition, expert systems can work with incomplete or unavailable information, and can produce ranked results based on a weighting of the significance of the inputs.

An expert system is constructed by first capturing the expertise of one or more recognized experts in the application area. Typical experts apply assorted "heuristics," or rules-of-thumb, as well as an application-specific body of knowledge to the problem at hand. The task of embedding this knowledge in an expert system is known as "knowledge engineering." Knowledge engineering is an iterative process that allows the expert and knowledge engineer to continue to refine the rules and facts embedded in the expert system until it functions at the level of the expert. Unlike conventional computer programs, expert systems are designed to be readily modifiable to incorporate new, conflicting, or incomplete information.

Expert systems provide users with consultation on problems in the application domain. Often, user input is followed by system inquiries for information in order to further direct its search. Heuristics determine whether a possible conclusion or multiple conclusions are reachable. When a set of conclusions is reached, the expert system ranks them with respect to their certainty based on available information. The user can query the expert system at any time for the rationale used to solve a problem. This explanation capability provides significant assistance regardless of the user's degree of expertise, because it allows an objective evaluation of the schemes being used to reach a conclusion.

Expert systems are necessarily domain-specific applications. The heuristics used in one domain do not generalize easily to other domains, in most cases. Terminology and the body of knowledge applied in the area of expertise is often very specific, although it usually builds on much more general areas (i.e., criminalistics involves the application of the natural sciences in forensic investigations). A considerable amount of work is necessary to build an expert system that typically combines 500-1500 rules with what can be a very complex representation of the body of knowledge for the application domain. The actual cost of constructing expert systems continues to decline, however, as new, more powerful tools are introduced [5,19].

Expert System Applications

An expert system's matching and ranking capabilities, combined with the power of enlarging or narrowing the geographic scope of an inquiry, provide powerful tools for investigative analysis. In a crime scene scenario, for instance, with little or no evidence, an expert system can generate a demographic profile of a suspect based on variables such as type of crime, location, time of day, and method of entry (if applicable). The demographic profile can then be used to generate candidate lists while expanding or narrowing a geographic search based on similar criteria.

At a different level of analysis, the modus operandi (MO) of a particular crime can be used by the expert system to identify the set of solved and unsolved crimes with similar MOs. It is the expert system's ability to match the MO of crimes based on similarity that is so powerful, because it can take into account conflicting or incomplete information that conventional computer programs are hard-pressed to deal with effectively.

In the Phoenix Police Department, the CAPRI or PACE systems would be searched to identify crimes with similar MOs. Once these crimes are identified, a number of possibilities arise. The first involves aggregating the physical evidence at other similar crime scenes to narrow a candidate list based on the subset of physical evidence that seems particularly applicable. This, in many cases, may pare the candidate list by an order of magnitude simply by adding limiting parameters suggested by the aggregated evidence.

Another possibility involves a scenario in which a suspect is stopped and has a television in the back seat of his car. As the officer enters the serial number into the PACE computer through his mobile data terminal (MDT), across town another officer has just finished taking a stolen property report and begins entering the data into PACE. A match of the stolen TV is made and the suspect is arrested. Again, the expert system can be used to identify similar unsolved crimes and the information associated with those crimes, both circumstantial and physical, used to interview the apprehended suspect, in an attempt to solve additional crimes.

Architecture

Expert systems provide ideal "front-ends" to the wealth of information that currently resides in multiple, heterogeneous CJIS. If the data in these systems is made available to end-users through an intelligent interface that represents the collective expertise of experts in their respective fields, the use of this data can be much more effective and efficient. By integrating existing systems through a single interface specifically tailored to the task, intelligent searches are performed of only the necessary or pertinent systems, and are completely transparent to the user. Those searches will ultimately be more efficient than currently because expertise will be applied to the query, thereby limiting its use of computer resources by intelligently adding search-limiting parameters [11]. This approach saves considerable time by providing the end-user with a single, intelligent interface, specific to their task, capable of guiding them through their search for information. Implementation of this scheme requires an architecture which takes into account the multiplicity of existing heterogeneous systems as well as the domain-specificity of expert system applications. Such an architecture, using existing or proposed systems in the Phoenix Police Department, is graphically depicted in Figure 1.

The "CJIS Interface" provides a single, unified view of the existing CJIS to the expert systems. Any expert system needs only to specify what data is required, while the details of how to perform the search are the responsibility of the CJIS Interface. Due to the possibly redundant, conflicting, or absent information in a given set of CJIS, the CJIS Interface controls the search of appropriate systems, reconciles conflicting information, and notifies the expert system when data for its request does not exist in the supported CJIS [7]. Expert systems can be constructed for additional applications and have immediate access to all of the information in the existing CJIS. New or modified systems can be added as well, making the new information immediately available to the existing expert systems by updating the CJIS Interface.

The CJIS Interface provides a lower-cost alternative to complete integration of existing CJIS. The CJIS do not have to "talk to" each other in this architecture, nor do they need to know the other systems exist. The CJIS Interface has knowledge about the type of information that resides in the different systems, as well as knowledge about how to access it, effectively integrating the systems from the point of view of the user interface. The systems may be on different machines, with different operating systems, communication protocols, media, and file formats. That information is contained in the CJIS Interface, isolating both the user and the builder of the expert system from the procedural details of providing access to the data in the existing CJIS. In addition, this architecture exploits the inherent parallelism of the CJIS - different searches can be performed on two different CJIS at exactly the same time.

A standardized CJIS Interface, common to all law enforcement agencies, facilitates the proliferation of expert system technology regardless of the CJIS actually in use at a particular site. Because all data access is performed through the CJIS Interface, the expert system can be transplanted to another site with little difficulty. And the interface allows different systems and data to be applied to the problem-solving task, using the same expert system user interface. A common interface provides extensibility to the architecture, as systems can be added on either side of the interface and the increased information made quickly available to the end-user. This standardization applies at local, regional, national, and international levels, enabling geographical expansion of inquiries across dissimilar systems. This would enable, for instance, international policing organizations to identify similar terrorist incidents to aid in the tracking of world-wide terrorist organizations.

Linking multiple, heterogeneous systems through a single interface is a complex task, although implementations [8, 12, 20] and standards [2] exist. Other interfaces have already been constructed that facilitate rapid prototyping and evolutionary support of complex applications [18], allowing features to be added or changed easily with no adverse impact on existing functionality. The CJIS Interface incorporates a layered approach to isolate different machine dependencies while taking advantage of a common query and data representation format.

Additional complexity in interface construction is introduced when considering the efficient retrieval of textual information from the large data stores that comprise some CJIS. Even though the search would be limited to those pieces of text that met the limiting qualifications discussed earlier, the efficient processing of the retrieved text must still be addressed. Much of the information necessary to compare MOs, for instance, is stored in the form of text. This information is not indexed in the currently available CJIS to facilitate efficient processing. The time necessary to manually index the existing information comprehensively makes this approach unattractive, while automated full-text indexing techniques suffer problems when advanced compensating techniques are not used [21]. `Smart' indexing of the textual information can be performed by automated indexing tools outfitted with domain-specific indexing knowledge [6]. These indexing techniques, however, require additional work to be completed on the target CJIS. Fortunately, the CJIS Interface takes the available indices in the existing CJIS into account when planning a retrieval strategy [17], thereby performing a high-level optimization of the retrieval request [4] that partially eliminates text processing inefficiencies.

Ideally, the law enforcement community can mandate the design of and adherence to a CJIS Interface standard, as has been done for fingerprint identification data interchange [1], and can include index standardization. This benefits CJIS vendors, expert system vendors, and the law enforcement community itself. CJIS vendors can then focus on developing their base technologies, rather than user interfaces. Expert system vendors are guaranteed a uniform interface for data access which allows them to concentrate on their core technology. End-users realize benefits in increased system compatibility and functionality. One way to enforce a proposed interface standard is to require adherence to the standard in future vendor contracts.

In the absence of a formal standard and vendor participation, an interface can still be constructed for a particular site. An interface designed with only a small subset of systems in mind, however, is sure to suffer portability and extensibility problems. Still, for a single site, this is a lower-cost method for integrating existing systems and providing in-house extensibility of systems than a full-scale system-level integration effort.

Cost Savings

Police departments are labor-intensive organizations. The single highest recurring cost is salaries. The advent of expert systems in the criminal justice system community will provide for the more efficient use of human resources to focus on crime prevention. Expert systems, coupled with integrated data stores, significantly reduce personnel costs by saving time in collecting, evaluating, and searching for information [15]. Quicker suspect identifications saves investigative staff-hours and reduces serial offenses.

The expert system "front-ends" operate in a stand-alone fashion for training purposes, continuously available for consultation. In this capacity, the expert systems perform as "on-line mentors" for the novice at a fraction of the cost of providing a human expert. The steep learning curve associated with becoming technically proficient in a given specialty is overcome more quickly and cost-effectively, because the systems not only advise `what' to do but `why.'

Conclusion

Enormous sums are invested in current CJIS, while equally large amounts are annually invested in law enforcement personnel salaries. As expert system technology becomes available in the commercial marketplace, it enables the law enforcement community to significantly leverage its computer and human resources. Considerable additional benefits are gained by standardizing an interface to use this new technology. Doing so will preserve the commitment in existing resources, greatly enhance the abilities of law enforcement personnel through integration and intelligent use of systems, and provide a mechanism for system portability and growth at the local, regional, national, and international levels.

References

[1] ANSI/NBS - ICST 1-1986, American National Standard for Information Systems - Fingerprint Identification - Data Format for Information Interchange, American National Standards Institute, New York, NY.

[2] Bachman, C., Ross, R., "Toward a More Complete Reference Model of Computer-Based Information Systems," Computer Networks, Vol. 6, 1982.

[3] Bayse, W. A., Morris, C. G., "FBI Automation Strategy: Developing AI Applications for National Investigative Programs," in Signal, Vol. 41, No. 9, May 1987.

[4] Date, C. J., An Introduction to Database Systems, Vol. 1, 4th Edition, Addison-Wesley Publishing Co., Reading, Mass., 1986, p. 339.

[5] Harmon, Paul, King, David, Expert Systems - Artificial Intelligence in Business, John Wiley & Sons, Inc., NY, 1985.

[6] Humphrey, Suzanne, "Illustrated Description of an Interactive Knowledge Based Indexing System," in Proceedings of the Tenth Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, New Orleans, June, 1987. Yu, C. T., Van Rijsbergen, C. J., Eds., ACM Press, New York, New York.

[7] Jakobson, G., Lafond, C., Nyberg, E., Piatetsky-Shapiro, G., "An Intelligent Database Assistant," in IEEE Expert, Vol. 1, No. 2, 1986.

[8] Jakobson, G., Piatetsky-Shapiro, G., Lafond, C., Rajinikanth, M., Hernandez, J., "A Knowledge-Based System for Integrating Multiple Heterogeneous Databases," in Proceedings of the Third International Conference on Data and Knowledge Bases; Improving Usability and Responsiveness, Jerusalem, Israel, June 1988.

[9] Jones, A. D., Western Identification Network Request for Purchase (RFP), 1988.

[10] Kellogg, Charles, O'Hare, Anthony, Travis, Larry, "Optimizing the Rule-Data Interface in a KMS (Knowledge Management System)," in Proceedings of the Twelfth International Conference on Very Large Databases, August, 1986.

[11] King, J. J., Query Optimization by Semantic Reasoning UMI Research Press, Ann Arbor, Michigan, 1984.

[12] Landers, T., Rosenberg, R., "An Overview of Multibase," in Proceedings of the Second International Symposium on Distributed Databases, September, 1982.

[13] Laudon, Kenneth C., "Data Quality and Due Process in Large Interorganizational Record Systems," in Communications of the ACM, Vol. 29, No. 1, January, 1986.

[14] Laudon, Kenneth C., Dossier Society: Value Choices in the Design of National Information Systems, Columbia University Press, 1986.

[15] "Artificial Intelligence is Brainchild for Feds," in Law Enforcement Technology, September/October 1988.

[16] Lucas, R., "An Expert System to Detect Burglars Using a Logic Language and a Relational Database," in Proceedings of the Fifth British National Conference on Databases, Kent, England, July 1986.

[17] Lynch, K. J., Hoopes, L. M., "A Generic Query Processor for Retrieval," in Proceedings of the Fall Ingres Users Association Conference, Tampa Bay, Florida, Nov. 1987.

[18] Lynch, K. J., Hoopes, L. M., "An Interface for Rapid Prototyping and Evolutionary Support of Database-Intensive Applications," to appear in Proceedings of the IEEE International Phoenix Conference on Computers and Communications, March, 1989.

[19] Martorelli, William P., "PC-Based Expert Systems Arrive," in Datamation, Vol. 34, No. 7, April, 1988.

[20] Martin, James A., Oxman, Steven, Building Expert Systems, Prentice-Hall, Englewood Cliffs, NJ, 1988.

[21] McHenry, W. K., Lynch, K. J., Goodman, S. E., "Handling Textual Information in a GDSS Database: Experience with the Arizona Analyst Information System," in Proceedings of the Twenty-first Annual Hawaii International Conference on System Sciences, Jan. 1988.

[22] Moses, Kenneth R., "A Consumer's Guide to Fingerprint Computers," in Identification News, June 1986.

[23] Phoenix Police Department CAPRI Users Manual, 1976.

[24] Phoenix Police Department PACE System Overview, September, 1988.

[25] Ratledge, Edward, University of Delaware, Newark, NJ, Personal Communication, October 19, 1988.

[26] Ryan, Alan J., "Expert System Stalks Killers," in Computer world, Vol. 21, No. 28, July 1987.

[27] Sadasivam, V., Subramanian, R. K., "An Intelligent Computer System for Forensic Science," in Proceedings of TENCON 87: 1987 IEEE Region 10 Conference 'Computers and Communications Technology Toward 2000,' Seoul, South Korea, August, 1987.

[28] "Text Processors Will Help in Murder Enquiries," in Data Processing, Vol. 27, No. 4, May 1985.

[29] Vedder, R. G., Mason, R. O., "An Expert System Application for Decision Support in Law Enforcement," in Decision Sciences, Vol. 18, No. 3, Summer, 1987.

[30] Wilkins, Bryan, "FBI Moving to Employ AI Software as Crime Fighter," in Computerworld, Vol. 19, No. 5, February, 1985.