DANIEL H. LUFKIN*
*Staff Consultant to the Committee
The American Justice Institute of Sacramento, California, working under a grant from the National Institute of Mental Health, completed in 1972 a six volume report1 of a three-year study of "the utilization of advanced information system technology as a means of improving the correctional decision-making process." The aim of the study was to design a system to enable managers of correctional institutions to make completely objective decisions about the treatment and disposition of criminal offenders. The study was the work of the Institute's Correctional Decisions Information Project (CDIP), whose epigraph is inscribed on the second cover of Volume I of the report:
"TODAY AN INFORMATION SYSTEM HOLDS FOR CORRECTIONS THE SAME BREAKTHROUGH POTENTIAL AS DID THE MICROSCOPE FOR BIOLOGICAL SCIENCES YESTERYEAR."
It must in no way demean the dedicated and intelligent effort of the CDIP staff to point out that any project that aims to create an automated personal data system to monitor and control the popula tion of a prison efficiently necessarily creates a system with all the earmarks of the worst surveillance data bank any civil libertarian could imagine. CDIP has completed much of the work needed to reduce 1984 to practice. Simple substitutions of the words "governmental" for correctional2 and "citizen" for offender3 in the following excerpts from the CDIP report transforms serious and humane objectives for prisoners into a nightmare for citizens.
(C)orrectional administrators ...must be able to ...determine the ability of each operational program to assist various types of offenders toward correctional goal attainment. Such an ability is totally dependent upon information. Thus, information is power to withstand irrational, unjustified onslaughts. Information is power to confirm constructive policy decisions. information is power to provide leadership for a rational approach to an improved correctional process. (Vol. 1, pp. 1-2).
The Correctional Information System portrayed in these documents is for that breed of managers which strives for an increasingly effective efficient, and responsive approach to rational, humane control and reintegration of offenders. Vol. 1, p. 6)
The recycling approach, or Correctionetic concept of successive approximations to desired goal attainment, is not limited to the management of corrections. It applies equally well to individual offenders as they strive to achieve their objectives on any of a number of dimensions of personal adjustments, e.g., vocational, marital, leisure time/social, or academic. (Vol. 1, p. 7)
This type of decision-making assistance is possible for correctional managers as they perform the following basic functions which constitute the management process:
1. Goal Definition
3. Operations Control
4. Achievement Assessment
5. Effectiveness Evaluation (Vol. 1, p. 8)
The last paragraph betrays the weakness of the transformation: if we assume that the CDIP system could apply as well to a nation as to a prison, we are also assuming that "management" and "government" are interchangeable. In fact, however, the idea of the social contract, of authority derived from the consent of the governed (rather than from the managed), is precisely what differentiates a nation from a prison.
Valuable as a more thorough exploration of that differentiation might be, we shall forego it here in order to concentrate on the practicalities of the CDIP approach in the context of a special micro-society into which the problems of citizens' rights and privacy do not immediately intrude. Nevertheless, the conditions of prisoners and of free citizens are not diametrically opposed: prison and 20th-century America are not the end points on any scale of social values. Prisoners do have rights and privacy just as ordinary citizens have restrictions and intrusions. Correctionetics includes data on an offender's religion and sexual practices, but none on the contents of his letters or his conversations with his lawyer. Correctionetics is thoroughly benevolent, and efficient benevolence is precisely the characteristic that seems to lie at the root of our suspicions of the computerized state.
At the heart of Correctionetics is the capability for what CDIP calls demand reporting: the capability of producing from a generalized data base a report whose content and structure fit the needs of one particular decision. In such a system, the capability of generating routine reports with fixed content is implicit. In the correctional context, for example, the manager may request a report listing the total number and names of offenders in a particular institution who have been convicted of a certain crime, who have served their minimum sentence, are in a given range of age, and who have a particular occupation or skill. Such a report would simplify matching offenders eligible for release with known job possibilities outside. In the civil context, a demand request might be for a list of the blond males in their late thirties who drive blue Pontiacs with the last license digit of seven. The motive of the request is not the offer of a job, but rather the search for a bank robber.
The capability of a data system to perform such a search clearly rests on two features: a comprehensive file of personal characteristics, and the logical ability to compare the file contents with the terms of the request. Both these capabilities are easy to build into a computer system; in theory there is no difficulty at all in setting up a system of considerable range and depth. Experience tells us, though, that real life consistently falls short of expectations. It is instructive to see how this universal principle operates on Correctionetics and to extrapolate that knowledge to the real world.
Offender Data File
The underlying operating unit of Correctionetics is the offender data file (ODF), a record of 369 different facts and opinions about the offender. When the CDIP study fast began, in July 1969, the ODF included only 200 data elements. Let us note for later reference that the number of data elements found to be needed nearly doubled in about two years of planning and experimental operation of the system. The ODF begins with the name (and aliases) of the offender, three identification numbers, his date and place of birth, his first year of State residence, his ethnic origin, and his religious preference. This much information, the offender identification block, requires 139 characters of file space as a minimum for each offender. (The average offender was found to have 2.6 aliases at 33 characters per alias, and some had as many as 12.)
The ODF goes on to record the legal status, the offense history, the medical, dental, psychological, psychiatric, academic, vocational, and adjustment histories of the offender, details of his childhood, his family and its economic status, his work history in the institution, and his prospects for release and parole. The data are grouped into 17 blocks and occupy a minimum file space of 1134 characters, but the complete ODF would easily fit on a single page of typescript, since most of the information is entered in coded form.
Coding data in order to conserve storage space and to make possible a logical search for a known data entity is characteristic of computer data processing. The extent to which the coding con ventions match the underlying structure of the data determines to a very great extent the ultimate power of the computer program to handle any but the simplest sorting tasks. The coding manual for building the ODF provides explicit codes for every coded data element. Since the computer's perception of the real world takes place only through the medium of the codes (outside of literal data such as names and the like), the structure of the codes and selection of the code elements must be made with the greatest care and foresight.
In the experience of practically every organization that has developed a sizable computer data base, one of the greatest expenses in the operation comes in converting the data from conventional manual to encoded machineaccessible form. Conscientious, accurate coding demands well-trained, highly motivated clerks who can keep an extensive body of coding rules in mind and apply them quickly to an amorphous mass of real-world facts.
In the CDIP work, the coding manual is a 200-page volume explaining every possible entry in the ODF. The scope and structure of the codes themselves have apparently never been tested by processing a large number of actual correctional records, although we shall later discuss a greatly restricted pilot program and its results. The coding manual shows the extent to which standardized codes for occupations, school subjects, diseases, and similar common data entities have already been adopted among independent but parallel data-processing organizations. Academic course codes are those of the California Department of Education, and include not only introduction to data processing (MXA) and computer techniques (MXB), but also a very full range of elementary and secondary school subjects. The vocational training codes are those used in Federal government job classifications. The Federal code is considerably edited to provide a fuller breakdown of skills important to prison operation: laundry workers (36x.xxx), farm workers (2xx.xxx), food workers (52x.xxx), mattress inspectors (780.687), and the like. (There is no code provision for locksmith.) Medical diagnosis and treatment is coded according to the American Medical Association's Standard Nomenclature of Diseases and Operations. The codes for voluntary and leisure time activities presumably reflect the choices available to actual inmates of the California correctional system. They include all the familiar sports plus some surprises, such as bicycle racing (103) and golf (111). Special interest groups include aviation (706) and transactional analysis (726). That an activity code for the classification of prisoners can hold surprises is a good indication that a similar code for the public at large would run to many times 200 pages.
Data for Decision Making
During the course of the CDIP study, two pilot programs were carried out to test the preliminary design of the system and to demonstrate the operation of the system before experienced correctional managers. The results of those programs are interesting as an indicator of the potentials and pitfalls we could expect to meet in a large-scale general system.
In testing some of the preliminary design concepts of the system, CDIP planners identified the following factors in decision-making processes that use data in the way they can be provided by a largescale computerized system:
These and other more peripheral problems were tested in an experimental setting with data records of actual prisoners presented to experienced correctional officers in a simulation of computer operation. The officers decided on the disposition of three hypothetical cases: granting a minimum-security custody rating; granting a parole after a minimum sentence had been served; and revoking a parole after a borderline violation. The type of data, its order of selection, and its weight in the ultimate decision were all recorded.
The detailed analysis of the experiment appears in Appendix D of the CDIP report; it is enough here to summarize the findings which would have broader applicability to a similar task in a citizen data bank.
Data for Reports
In the second pilot test, the capabilities of a computer program package much more restricted than the full, planned correctionetics system were demonstrated to meetings of senior correctional officers at their national conventions. A special 74-item ODF was prepared from the conventional records of 5756 offenders in a cross section of the institutions of the California Department of Corrections. It is worth noting that the project found it necessary to "embellish" (CDIP's word) the original data to make them conform to the requirements of the demonstration.
In the first demonstration, at Palm Springs, a computer at Santa Monica was loaded with the data base and the demonstration programs. The terminal at Palm Springs was connected to the computer by telephone. The demonstration programs were relatively simple sorting routines which demonstrated how to generate a list of offenders to be released in the next month, and then searching the ODF for a qualified inmate to take over a clerk's job vacated by a releasee. After the prepared program application was demonstrated, the spectators were allowed to make up their own queries for the, data, although it is not clear from the report what these queries were or how well that part of the demonstration worked.
The second demonstration of the same program package was held in Cincinnati. It is a keen comment on the computer specialist's faith in his charges that the CDIP staff took the precaution of punching all the query input on paper tape beforehand, so that a keyboard mistake-alas! all too common-would not upset the demonstration. The staff also took the precaution of punching the computer's output on paper tape beforehand and taking that tape with them to Cincinnati. There, the output could be fed into the teletype printer under the control of a foot-switch, thus simulating the action of a computer at the other end of the line without exposing the demonstration to the dangers of real-life computer operation. (It is also a tribute to the candor of the CDIP staff that they fully describe this ploy in their report.) The demonstration ended with a period of genuine computer operation over the link, during which the audience had an opportunity to try the system. Typical queries from the experienced correctional officers dealt with average time served by offenders in various classifications of confinement; profiles of offenders involved in escape attempts, juvenile commitment history of selected sets of adult offenders, and other similar sorting and listing tasks.
Correctionetics as a Data Bank
What does this report about Correctionetics, an automated personal data system designed for a prison society with few of the traditional concerns for privacy, have to tell us about computers and privacy in our own wider society? Are we looking at a worst-case microcosm, one from which we can no more extrapolate to our present civil society than we can from an anthill? Even as an antihill can teach us something about living beings in general, so can Correctionetics teach us something about the intrinsic limitations computerized personal data systems have, even in the absence of manifest safeguards for privacy.
Let us look at some of the features of Correctionetics and compare them with roughly corresponding features of other personal data systems.
Scope. First, and of fundamental importance, Correctionetics stores no more data on an individual prisoner than the manual system did. In point of fact, it stores less. When the records of the sample population were being prepared for the demonstrations, it was necessary to omit all but a tiny fraction of the material in the prisoners' record jackets, many of which were half afoot thick. The material omitted was that least suited to computer treatment; that is, anecdotal and narrative records, interview reports by psychologists, extracts from correspondence, and the like. It is this sort of intelligence record that is fundamentally unsuited for computer treatment, and which would have the greatest potential for harm to privacy if it were to enter the lightly protected files of a computer data bank.
Costs. Second, Correctionetics seems to be so grossly uneconomical that there would be little incentive to adopt it in a full-scale way. As every business comptroller knows, it is almost impossible to price out a computer system before it goes into operation, and difficult enough even to measure the running operating costs. The CDIP report is reticent on costs, but we would estimate the storage and processor requirement for an offender population of 50,000 to be over 250,000,000 bytes (CDIP Table 5.4.2). Roughly corresponding commercial credit experience suggests a cost of about $80,000 per month to which staff and overhead costs would add about 50 percent to bring the total cost to about $120,000 per month. It is hard to see that the advantages of automated prison management on the scale suggested by CDIP would be defensible unless it could be carried as a partial load on some larger generalpurpose system.
Impact on Decision Making. Third, the impact of Correctionetics on the actual process of prison management decision making does not seem to be all that striking. It is obvious that the computer has no difficulty in finding, for example, the average age of narcotics offenders in a particular institution, but one suspects that the warden could guess the figure closely enough for practical purposes with no aid at all. For particular tasks, such as matching parolees with job openings, the services of a computer are well defensible, but more economically carded out in a special-purpose system that only handles employment data and need not process the excess baggage of the rest of the offender data file merely to arrive at a job match. This illustrates a point that deserves emphasis again and again in designing data-processing systems: a system should be no larger than needed to do a particular task. Money spent to provide capacity for the possibility of data processing in the abstract, or merely to provide "management information" is like wagering at unknown odds. A management information program run once or twice a month on a computer system that otherwise earns its keep on accounting, payroll, and inventory yields impressive decorations for the board room and likely does no harm. But neither does it do enough good to deserve a dedicated computer system all to itself.
Safeguards for Correctionetics
Finally, we may look at Correctionetics as a test case for the application of safeguards. What effects would there be if Correctionetics gave offenders more control over information about themselves?
In the Correctionetics system there is no provision for feedback from the data subjects. The prison management's goals are defined in terms of data measurements made through the system, and the system is then used as the means of bringing operations of the prison into conformity with those goals. If a data error creeps in from any source, the system can produce a false measurement or a false operation or both; without suitable feedback, the false measurement may well reinforce the false operation instead of correcting it.
Let us look at an example as it might actually run through the Correctionetics system. Through a coding error, a prisoner's file is changed to show that he is an active homosexual. A status change report is automatically generated which removes him from a television repair course (forbidden to sexual offenders) and transfers him to a cell in a more secure block (because a profile of such offenders shows them to be, on the average, more aggressive than others). These two actions confirm the prisoner's suspicions about the prison administration and he fulfills their expectations by actually becoming sullen and aggressive, which behavior, in turn, generates another automatic transfer order to an "adjustment center." In this scenario, and in a hundred others we could imagine, an originally minor error in a record has snowballed into serious injustice.
Giving the prisoner a right to know what information his file contains would have had the immediate effect of discovering the error, provided he realized that some change in that information had taken place. In this case, the change in training status would have been an obvious clue to him. A right to secure correction of the data would have stopped its propagating in the program and would have prevented or undone the subsequent actions the system made on the basis of the error.
Thus, the possibility of feedback from the data subject to the data bank can act as a powerful brake on the freedom of an authority to take arbitrary action. It is obvious that this would have clear benefit for a person at the bottom of the heap, but we wish to point out that it also protects the authority taking action. If we make the assumption that administrative injustice will eventually come to light and be dealt with through the law, it is very much to the benefit of the warden, in our example, to insure that his decisions are based on the best data he can command. Rules to ensure that errors in personal data banks are discovered and corrected promptly will go far toward preventing abuse of even so stern a system as Correctionetics.
Computerized Decision Making
The deeper question of the actions that an automated system such as Correctionetics can take on the basis of even perfect data also deserves careful consideration. In our example from the actual program, a record as a sexual offender was automatically treated as sufficient cause to disqualify an innate from training as a television repairman. This is a simple decision to program, and one presumably based on an actual rule of the California Department of Corrections. In pre-automation practice, the application of such a rule would usually take place in a context such that knowledge of other factors in the offender's record would come to the attention of the training officer. He might give the rule only as much weight as he thought appropriate in the light of all the factors in an individual case, and could certainly at least take initiative to seek occasional exceptions from the rule.
It is precisely that sort of personal initiative which seems to be the most strongly appreciated advantage of human over computerized administration. Although we have all experienced occasions in which a bureaucrat acted like a computer, we also recognize those occasions as the exceptions to our usual experience with human decision making.
To be fair, it is possible in theory to program a computer to simulate human decision making. In practice, though, it is obvious from the Correctionetics experiment that we are far from attaining that end.
1Correctional Decisions Information Project, Correctionetics: Modular Approach to an Advanced Correctional Information System (Sacramento, Calif.: American Justice Institute), 1972.
2 Not italicized in original text.
3 Not italicized in original text.