File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/79/j79-1062_metho.xml
Size: 127,993 bytes
Last Modified: 2025-10-06 14:11:16
<?xml version="1.0" standalone="yes"?> <Paper uid="J79-1062"> <Title>WAS1 IINGTON DEVELOPMENTS FEDERAL DATA ENCRYPTION STANDARD APPROVED BY COMMERCE DEPARTMENT</Title> <Section position="1" start_page="0" end_page="0" type="metho"> <SectionTitle> EXPLAlATION CAPABILITIES OF PRODUCTION -BASED CONSULTATION SYSTEMS A. CARLISLE SCOTT, WILLIAM J. CLANCEY, RANDALL DAVIS, AND EDWARD H. SHORTLIFFE </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> Association for Computational Ginguistics </SectionTitle> <Paragraph position="0"> A computer program that models an expert in a given domain fa more likely to be accepted by experts in that domain, and by non-experts seeking it9 advice, if the system can explain its actions. An explanation capability not only adds to the system s credibility, but also enables the non-expert user to learn from it. Furthermore, clear explanations allow an expert to cbeck the system's &quot;reasoninqfl, possibly discovering the need for refinesents and additions to the system s knowledge baseu In a developinq system, an explanation capabilitv can be used as a debuq~lnq aid to verify that additions to the system are worckins as thev should, Thls paper dixusses the general characterlstlcs of explanation systems. what types of explanablons thev should be able to give, what types of nowl ledge will be needed in order to give these explanabiohs, and how thls knowledge mlqht be oraanlzed. The ex~lanation fae ~lltv In MYCIN [5,6,7] is discussed as an 11lustrakion of how the various pnoblems miqht be approached.</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.2 Organization of Knowledge in MYCIN . . - . 9 . 19 2.3 Scope of MYCIN'S Explanation Capability . . rn . . 25 2.4 Understanding The Questlon . . . . . . . . . - 30 2.5 Answering the Question . . . . . . . . 36 </SectionTitle> <Paragraph position="0"> Conclusions . . . . . . . . . . I I . 48</Paragraph> </Section> </Section> <Section position="2" start_page="0" end_page="0" type="metho"> <SectionTitle> 1 General Discussion </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 1.1 Consultative Production Systems </SectionTitle> <Paragraph position="0"> A consultation proqram plays the mle of an expert consultant in some domain, ~ivin~ advice or answers to non-experts with ~roblems in the domain. Users will often want to know how the svstea arrived at its results durinq a particular consultation. This Daper explalns how the implementation of such a proqram as a ~rod~ction system can fac~lltate program-generated ex~lanations.</Paragraph> <Paragraph position="1"> A production system [2] consists of three basic components. a set of production rules, a data base which is both used and updated bv these rules, and a rule interpreter. A production rule often is in the form of a situation-action rule. it describes a situation and a set of actions to be taken if tbis situation is found to exist. The rule intermeter deternines the order In which rules will be tried, checks to see if the situations exist, and undertakes the required actions. It also determines how manv of the potentially useful rules wjll be used. only the flrst (where orderinq may be predetermined or comouted dynamically), all possible rules, or enough rules to satisfy some criterion that the interpreter uses.</Paragraph> <Paragraph position="2"> In some production systems, rules are always trled in a predetermined order. In others, the order in which rules are tried varies with different consultations, since a rule will be tried as soon as the rule interpreter determines that it may be useful. In such svsteas, the common alternatives are data-directed rule invocation, in which a rule is considered &quot;usefulff if its situation part matches the data base, and qoaldirected rule invocation, in which a rule is if its action part will help the system reach its current goal. Many systems use a combination of soal- and data-directed rule invocation.</Paragraph> <Paragraph position="3"> A consultative production system need not be a psycholoqical model, imitating a human s reasoning process. The imporbant point is that the System and a human expert use the same (or similar) knowledqe about the domain to arrive at the same answer to a given problem. The svstem's rules and data base can be viewed as a knowledge base containinq the domainspeclfic knowledge of an expert as well as facts about a particular broblem. When a rule is used, its actlons make changes to the data base which are the sy5tem's decisions or deductions. Thus, a rule can be thouqht of as a plece of judgmental knowledqe, using the judqment and knowledge of au expert to make deductions.</Paragraph> <Paragraph position="4"> The process of trying rules and taking actlons can be thouqht of as llreasonsngw, and explanations consist of showing how rules used information provided by the user to make various intermediate de~uctions and finally to arrive at the answer. If the information contained in these rules is sufficient to show why an action was taken (without aettinq into programming details), an explanation can conslst of printing each rule that was used (or an Engllsh equivalent of what the rule means,)</Paragraph> </Section> </Section> <Section position="3" start_page="0" end_page="0" type="metho"> <SectionTitle> EXPLANATION </SectionTitle> <Paragraph position="0"> The three coniponents of a production system (a RULE INTERPRETER, a set of PRODUCTION RULES, and a DATA BASE 1 are augmented by an EXPLANATION CAP4BILITY. The data base is made up of general facts about the system's domain of expertise, facts that the user enters about a specific problem, and deductions made about the problem by the system's rules. These deductions form the basis of the syst en's consultative advice.</Paragraph> <Paragraph position="1"> The explanation capability makes use of the system's knowled_qe base to qive the user explanations.</Paragraph> <Paragraph position="2"> This knowledae base is made UD of static domain-specif ic knowledge ( both factual and jud mental) and dynamf,~ knowledge specific to a particular problem' Performance Characteristics of an Explanation Capability The purpose of an explanation ca~abilitv (KC) is to qive the user access to as much of the system s knowledge as posbible. Ideally, it should be easy for a user to get a complete, understandable answer to anv sort of question about the system's knowledge and pera at ion -- both in general, and with reference to a particular consultation.</Paragraph> <Paragraph position="3"> This llplies three major goals in the development of an explanation capability.</Paragraph> <Paragraph position="4"> To ensure that the EC can handle questions about all relevant aspects of the system'$ knowledge and actions. It should be capable of giving a few baslo types of explanations, for example.</Paragraph> <Paragraph position="5"> How it made a certain decision How it used a piece of information What decision ~t made about aome subproblem Why it didn t use a certain plece of information Why it failed to ma!ce a certain decision Why it required a certain piece of infornatlon Why it didn t require a certain ~iece of information How it will find out a certain plece of Information [while tRe consultatinn is in oroqress] What the system is currently doing? [while the consultation is in ~roaress] The mecific set of explanation types which are chosen as basics, however, will depend on the particular system. 2) To enable the user to get an explanation which answers the question completely and comprehensively. To make the EC easy to use. A novice should be able to use the EC without first spending a large amount of time lehrninq how to request explanations.</Paragraph> <Paragraph position="6"> We will distinguish two slighfly different functions for an EC and divide it into two components. the reasoning-status checker (RSC) to be used during the consultation, and the general question answerqr (GQA) to be used during the consultation or after the system has ~rinted its results. A reasoning-status checker will answer questions asked durinq-a consultation about the status of the systsb s reasoning process, A few simple commands are often sufficient to handle the questions that the RSC is expected to answen A genera question-answer will answer questions about the current state of the syqtem s knowledge base, including both static domain khdwledge and facts accumulated duainq the consultation. A GQA will often need the ability to recognize a wide range of question types about many aspects of the system's knowledge. For this reason, it might be difficult to define a few simple commands which would be easy to learn and still cover all the possible questions that might be asked. Consequently, natural-language processlnq in this component may be im~ortant to an explanation system s acce~tability.</Paragraph> <Paragraph position="7"> In an interactwe consultatlor!, the svstem periodically requests information about the problem. This offers the user an opportunity to request explanations while the consultation is in proaress. In non-interactive oonsultatlons, the user has no opportundtv to interact hith the system untll after lt has printed ~ts conclusions. Unless there 1s some mechanism allowing a user to interrupt the reasoning process and ask questions, the explanation capability for such a system will be limited to questions about the system's final knowledge state. It will have no reasoning-status checker, and its general. question-answerer will only be accessible at the termination of the consultatian .</Paragraph> <Paragraph position="8"> Knowledge Requirements of an Explanation Capability An EC must know what is in the system's knowledge base, and how it is organized. In order to give explanations of the svstem's current (or previous) actions, an EC also needs to understand how the system s rule interpreter works. when ruces w~ll be tried how they can fall, what causes the interpreter to try one rule but not another etc. This qeneral &quot;schemaw for how or why certain rules are used, toqether with a comprehensive record of the specific actions taken during a particular consultation, can be used a$ a basis fbr explaining the results of that cansultation. A reasoning-status checker will need d record of what the system has done so far in order to explain how it arr4ved at the current ste~. General knowledge bf how the rule interpreter works is necessary in order to explain where the current step wlll lead. The abllity to understand individual rules Also may be necessary to the extent that the content of a rule may explain why it was necessary to use this rule, or may affect which futupe rules will be trled.</Paragraph> <Paragraph position="9"> A general question-answerer will need more ~nfomatlon about the system since the scope of its explanations is much broader. its task is to answer general questions about Lhe system's knowledge base. To do this, it must know how the system stores knowledge about its area of expertise (the static kdowledge with which it starts each consultation) and how it stores facts gathered during a particular consultation (its dynamlc knowledge). These two types of information will allow a GQA to answer guestions about the substance and extent of the ~roduction system's current knowledge. If an explanation capability also is ta provide information about how the system arrived at the facts that are currently in its dynamic knowledge base, the GQA will need all the information that a reasoninpstatus checker uses. a detailed record of the consultation, an understand in^ of the rule snterpreter, and the ability to understand rules.</Paragraph> <Paragraph position="10"> These three types of knowledge could be supplemented with a limited amount of qeneral information about such thinqs as elementary losic, set theory, and arithmetic comparisons. This would allow the GQA to answer more complicated questions about why the system's knowledge base IS inb its current state, and to answer questions involving pelationshi~s between different facts in the knowledge base.</Paragraph> <Paragraph position="11"> The nature of the consult'ation domain as well as what primarv purpose the explanation capability is to serve wlll influence the range of questions that an EC should handle. In some systems, a slm~le retrieval of facts may suffice, while others may need to give detailed description 01 the product ion system's ltdeclsionfl Drocess and to make a number of deductions from facts that it has.</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> Explanation Capability </SectionTitle> <Paragraph position="0"> Access to the consultation system's knowledge base is a prerequisite for performance of the explanation capabilltv. Other types of knowledqe may be added to the system to enable the EC to answer a wider range of questlona,</Paragraph> </Section> </Section> <Section position="4" start_page="0" end_page="0" type="metho"> <SectionTitle> L </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 1.4 Program Deaign Considerations </SectionTitle> <Paragraph position="0"> The last two sections described what an explanation capability is, butlininq what tasks it should perfom, and what it requires in order to per1 om these tasks, In thisl section, we discuss design considerations for the parent production system that will enable its EC to meet the requirements that were outlined in the previous section. This discussion is not meant to define the vcorrectw way of representing or organizing knowledge, but rather to mention certain factors which should be taken into account when declding what representation or orqanization will be best for a given product Ion system.</Paragraph> <Paragraph position="1"> The first step ~s to decide what basic types of questions the system should be able to answer. This will have a direct Influence on how the EC is im~lemented. It is important, however to make the rnitial desl~n flexible enough to accomodate possible future additions to the set of basics.</Paragraph> <Paragraph position="2"> If the basic forms are diverse enough, some level of natural-languaqe understanding may be necessary. The degree of sophlstlcation of the natural-languacze processor will depend won what kind of ~erformance is. expected of the EC The format and organigation of various components of the production system's knowledge base will affect the deslgn of an EC.</Paragraph> <Paragraph position="3"> Individual ~ieces of static and dynamic knowledge presumably will be organized m some fashion which makes them accessible during the consultation. A GOA facility could make use of such organization to help In finding the information needed bo answer a question. The less organized the knowledge base the more difficult wxlf, be the task of the EC, as more complicated routines must be used in order to find the desired ~nformation.</Paragraph> <Paragraph position="4"> During the cQurse of the consultation, the system should keep a record of its actlons for use by both Co~nponents of the explanation capability.</Paragraph> <Paragraph position="5"> Where the ordering of events is important (e.~;. when the action of one rule establishes the situation necessary for a subsequent rule to yuuceed), the record should be structured in a madner which reflects the ordering of events as well as the reasons whv each event occurred. 1.4.3 Knowledge of What Rules Mean The explanation capability will need to understand some of the semantics of indlvldual production rules. Thls requirement could be met by having the svstem's knowledge base include a description of what each Pule means, encoded in some form which would be of use to the EC. If the format of the system's rules is highly stylized and well-defined, however, it might be possible instead to implement a mechanism for 'Iread~ng&quot; the rules. the language in which the rules themselves are written could be defined. A high-level description of the individual components of this language, telling what each component means, could be used to enable the EC to read and understand rules. If the rule set consists of a large nuaber of rules, and these rules are composed entirely of a relatively small number of primitive elements, this second approach has the advantage that less information needs ts be stored -- a description of each of the orimit~ve components, as opposed to a description of each rule. When new rules are added to the system, the first approach requires that descriptions of these rules must be added. With the second approach, provided that the new rules are made up of the standard rule components, no additional descriptite information would be needed by the explanation capabilitv.</Paragraph> <Paragraph position="6"> 1.4.4 Knowledge of the Rule Interpreter Enabling an EC to understand how t,he rule interpreter works is analoqous to enabling lt to understand rules It must be able to &quot;read&quot; the interpreter or else it must have access to some stored description of how the interpreter works. ?here is a thirkl approach for understandlnq the rule interpreter, one whlch would not be feasible for understand in,^ a larse number of rules. Knowledge of how the interpreter works could be built into the EC -- the information would not be stated explicitly, but would be used implicitly by the programmer In writing the actual code for the explsnation capability. The EC can be thousht of as a number of ms~ec1a1ist3ff, each capable of glving a single type of explanation. There could he one specialist for each of the baslc questlon tv~es that the systen can answer. Each of the specialists needs only a small amount of ~nfomation about the rule interpreter whlch could be built Into ~ts ltexplalnln~n procrea. The final type of knowledge that some general quest~on-answering facilities wlll need is lnformatlon allowinq deductions to be vade froa facts In the knowledge base. The re~resentation and extent oC thls knowledge will depend upon the types of questions that the system is to answer If logic is needetl onlv to dete~mine the answers to questions of a certain tvpe for example, the necesaarv deductions could be built into the specialist for answerinq that type of question. On the other hand, in aowe explanation capibilities, the GQA will be expanded to do more than simply qive explanations of the system s actions or to query its data base -- it will be expected to answer a wide range of questions involvlnq various kinds sf inferences about the knowled~e base. Such a CQA will need to check for equality or set membership, make arithqetic compariSons, or make logical deductions. In general most information of this type can be embodied In a new kind of specialist which is an expert at some sort of logical deduction or comparison. Representation of this sort of qeneral knowled~e will become important as the GO4 becomes not sinlply an explanation tool, but also a deduct ive one.</Paragraph> </Section> </Section> <Section position="5" start_page="0" end_page="0" type="metho"> <SectionTitle> 2 An Example -- MYCIN </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.1 Overview </SectionTitle> <Paragraph position="0"> MYCIN [5,6,7] is an example of a production-based consultation system with a well-developed explanation capability. A production run is an infectious disease therapy consultation in which MYCIN is the infectious dlseaae expert, and the user is a doctor who wants advice about the treatment of a patient.</Paragraph> <Paragraph position="1"> Knowledge that is gathered during the consultation is organized into attribute-object-value triples. In response to questions during the consultation, the user enters information about the existence of several objects, called contexts. the patient, infections that the patient has, organisms which may be causing these infections, cultures that were taken, and drugs that were qiven, The task of the consultation system is to determine the values of various attributes (called clinical ~arameters) of these contexts. For example, AGE is a clinical parameter of the patient; IDENTITY is a clinical parameter of an oreanism, with STRFPTOCOCCUS as a possible value; SITE is a parameter of a culture, with BLOOD as a possible value.</Paragraph> <Paragraph position="2"> A clinical parameter's value may be determined by asking the user, or by using decision rules. The parameter is said to be rvtracedw when the syBtem has done all it can to find out the parameter's value. Tracinq a parameter involves asking the user for a value (where applicable) and trying rules for determining the value of that parameter. Rules are tried until the value is known with certainty or there are no rules left to use.</Paragraph> <Paragraph position="3"> Each decision rule has a situatiorl part called its PREMISE. This con9ists of predicates, conditions that are tested to determine whether the indicated situation exists. If .the conditions in a rule's PREMISE are true, its ACTION will be evaluated, giving new (or updated) values to some parameter(s) . Before a aondrtion in a rule's PREMISE can be tested, the parameters that it mentions must be traced. For example, before rule 209 (below) can succeed, the system must know the site of the culture, the portal of entry of the organism, and whether the patient is a compromised host. If any of the clauses in the PREMISE is false, or if the system is unable to find out the value of one of these parameters, the rule will fail. (PREMISE) If: 1) T@e site of the culture is blood, and 2) The portal of entry of the organism is GI, and 3) The patient is a compromised host (ACTION) Then: It 1s definlte (1 -0) that bacteroides is an organism for which therapy should cover Associated with each attribute-object-value triple is a certainty factor -- a number between -1 and 1 inclusive which indicates how stronqly the system belleves that the attribute of the object has the indicated value. The user nay modify the answer to any question with a certainty factor, and all rules make conclusions which specify a degree of oertainty as well as attribute, object, and value.</Paragraph> <Paragraph position="4"> Each context is named uniquely, allowing the system to refer to CULTURE-2, meaning the second culture, or ORGANISM-3, meaning the third organism. Moreover, the contexts are orqanized into a tree known as the context tree, which defines relationships among them. For example, an organism is the direct descendent of' the culture from whicl~ it was isolated. In the portion of a tree shown in FQure 3 ORGANISM-3 hangs under CULTUR'E-2 indicatinq that STREPTOCOCCUS was isolated from the BLOOD culture.</Paragraph> </Section> </Section> <Section position="6" start_page="0" end_page="0" type="metho"> <SectionTitle> Contexts, Clinical Parameters, </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> and Values </SectionTitle> <Paragraph position="0"> The rule interpreter (MYCIV'S control structure, described in detail in ['TI) choases the pules which should be used in the particular consultation, interprets these rules, and creates a record of ~ts actions for use by the explanation system. Rules are invoked to find out values of parameters in a given context. A rule is applied to the lowest context in the context tree whose parameters are mentioned by the rule. The rule can use (or conclude about) parameters of this context, or of anv context which 1s its ancestor In the tree. For example, if RULE209 were apolied to ORGANISM-3 (see Figure 3) it would need the SITE of the culture from which the STREPTOCOCCUS was isolated. The tree indicates that this is CULTURE-2, Rather than being a sequential cycle throuqh the rule set, wher-e each rule is tried in some predetepmined order, the flow of control is qoaldirected. This means that only rules which conclllde about the current goal (to find out the value of a given parameter) are examined. The PREMISE of one of these rules may need to use some ~arameter whose value is unknown, This sets up a subgoal, namely to deterqine the value of this parameter so that the rule can be used. MYCIN's goal-direcrted approach means that the system (and not the user) takes the initiative during a consultation. The user will be asked about onLy those parameters Qhich way be relevant to the particular patient's case.</Paragraph> <Paragraph position="1"> OrganizatYon of Knowledge in MYCXN In order to give explanations of a consultation system's decisions, an explanation capability must have access to the system's knowledge base. More informative emlanations can be given if the EC also has knowledae of how the syfitem works, a record of the consultation, and possibly some domain-independent knowledae. This section discusses how MYCIN meets these requirements.</Paragraph> <Paragraph position="2"> The system's knowledee base consf sts of static medical knowledge plus dynamic knowled3e about a specific consultation. Static knowledqe is further classified as factual and judgmental. Factual knowled~e consists of facts which are medically valid inde~endent of the ~ayticular case.</Paragraph> <Paragraph position="3"> Judgemental knowledge consists of product ion rules representing deduct ions which might be made, conditional an what is already known about the case. The format of production rules and of dynamic knowledge has already been described.</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.2.3 Organization of Factual Knowledge </SectionTitle> <Paragraph position="0"> As discussed in Section 2.1, all knowledge which is gathered durinq the consultation is organized into attribute-object-value triples. For consistency, many facts in the static knowledqe base also have this format.</Paragraph> <Paragraph position="1"> This includes objects such as bacteria and antibiotics, and attributes such as the staininrg characteristics of a bacterium or the recommended dosage of an antibiotic :</Paragraph> </Section> </Section> <Section position="7" start_page="0" end_page="0" type="metho"> <SectionTitle> ATTRIBUTE OBJECT VALUE ------.I-- ------ ----- GRAM E .COLI GRAMNEG </SectionTitle> <Paragraph position="0"> DOSE GENTAMICIN 1.7 mq kg q8h IV (or IM) The remainder of the factual knowledge consists of lists and tables: pieces of aedical knowledge, organized m such a way that they can be used to augment the producbion rules. For example, one such piece of knowledqe is the list of the possible culture sites which are normallv nonsterile,</Paragraph> </Section> <Section position="8" start_page="0" end_page="0" type="metho"> <SectionTitle> NONSTERILESITES : ( CERVIX CUTANEOUS-ULCER LOCHIA NOSE SKIN STOOL THROAT URETHRA VAGINA) </SectionTitle> <Paragraph position="0"> The likely pathogens associated with the different culture sites are organized in a table, with different entries for the different sites, If: 1) The site of the culture is one of: those sites that are normally nonsterile, and 2) This organism and at least one of the likely pathogens associated with the site of the culture agree wi'th respect to the following properties: gram morph air Then: There is strongly sugaestive evidence (.9) that each of these pathogens is the identitv of the organism Note that the information in the table could have been 0r~anized as attribute-object-value triples (where the object would be a culture site). If this had been done, however, the above rule could not have been written. To accomplish the same purpose (without a change in the control structure), the svstem would have needed several rules -- a separate one for each entry in the table. Structurinq certain facts into lists an.d tables er~ables individual production rules to exDress general theories which allow a number of specific deduct ions to be made.</Paragraph> <Paragraph position="1"> Each of MYCIN'S approximately 400 rules is composed of a small number of conceptual primitives.</Paragraph> <Paragraph position="2"> A total of 60 such primitives make up the language in which rules &re written, This design facilitated the implementation of a mechanisp for translating rules into Enqliah (described in detail in 171). Each primitivs functions has a translation template with blanks to be filled in with translations of the function's arquments. A larrr;e part of MYCIN'S exolanation capabilitv depends on this ability to translate rules into a form that the user can understand.</Paragraph> <Paragraph position="3"> Having a small number of rule components also facilitates the examination of rules to see which might be apnlicable to the explanation at hand. MYCIN'S knowledge of production rules, therefore, takes the form of a general mechanlsrn for &quot;readinq&quot; rules, On the other hand, no attempt has been made to read the code of the rule interpreter. Procedural knowledge about the interpreter is embodied in nspecialistsu, epch capable of answerinq a single type of question. Each specialist knows how the relevant part of the control structure works and what pieces of knowledge it uses. In order to understand rules, the system's various specialists use a small amount of knowledge about rules in general, tosether with descriptions or templates of each of the rule components. As an example, the following rule is composed of the units SAND, SAME, and CONCLUDE.</Paragraph> <Paragraph position="4"> The morphology of the organism is coccus Then: There is strongly suggestive (-8) that the identity of the orqanism is Neisseria] [When the rule is used, the LISP atom CNTXT is bound to some object, the context to which the rule is ap~lied (see Section 2.111 The template for CONCLUDE is shown below. This describes each of the arguments to the function: first, an object (context); second, an attribute (clinical parameter); third, a value for this parameter; fourth, the tally or degree of certainty of the P,REMISE; and last, the certainty factor -- a measure of how strong our belief in this conclusion would be, assuming bhat the PREMISE of the rule is definitely true.</Paragraph> </Section> <Section position="9" start_page="0" end_page="0" type="metho"> <SectionTitle> CONCLUDE I------- TEMPLATE: (CNTXT FARM VALU TALLY CF) </SectionTitle> <Paragraph position="0"> To illustrate how this is used, consider an explanation that involves finding all rules which could cbnclude that the identity of an organism is Neisseria. The appropriate swe.cialist would start with those rules which the avstem uses to conclude values for the parameter IDENTITY.</Paragraph> <Paragraph position="1"> Using templates of the varlous ACTION functions which appear in each of these rules, the specialist picks out onLy those (like RULE009) which have NEISSERIA in their VALJ slot.</Paragraph> <Paragraph position="2"> ThQ also illustrates the sort of knowledqe that can be built into a specialist . The specialist knew that the control structure uses stored lists telling which rules can be used to determine the value of each parameter. Furthermore, ~t knew that lt was necessary to look only at the rules' ACTIONS because it 1s the ACTION that concludes facts, while the PREMISE uses facts.</Paragraph> <Paragraph position="3"> Many of the explanation capability's specialists need a record of the consultation. This record 1s built during the consultation, and is organized into a tree structure called the history tree which reflects MYCIN'S goal-directed approach. Each node in the tree represents a goal and contains Information about how the system tried to accomplish this goal: bv asking the user or by trying rules. Associated with each rule is a record of whether the rule succeeded, and if not, why it failed. If trying some rule causes the system to trace a new parameter, thereby setting up a subgoal, the node for this subgoal is the offspring of the node containins the rule which caused the tracing. Figure 4 illustrates how part of ii history tree miwht look, In this example, R1JLE003 caused tracing of the parlameter CATEGORY which is used in the PREYISE of this rule, [RULEOO~ is shown above, see Figure 5 for RULE003 and RULE0373 goal: IRENTITY of ORGMISM-1 ask: question 7 rules: RULE009 (failed, clause 1) ... RULE003 (succeeded) ...</Paragraph> <Paragraph position="5"> system's actions, and explaining what facts the system knows. Some of the specialists for answering questions about the consultation make use of loqic J goal: GRAM of ORGANISM-1 ask: question 11 in arriving at their answers. In particular, to explain why a decision goal: CATEGORY of ORGANISM-1 rules: RULE037 (succeeded) ...</Paragraph> <Paragraph position="6"> wasn 't ade, the appropriate specialist uses the logical conclusion that the answer consists of ex~laining what prevented the system from using each of [no rules] the rules that would have made that decision.</Paragraph> <Paragraph position="8"> If deductions or comparisons are needed to answer questions of a specific type, then the necessary logic is built into the appropriate specialist. There is no general representation of knowledge about logic, arithmetic, or set theory that the explanation capability can use to make inferences from different facts in its knowledge base. To find out whether ORGANISM- I and ORGANISM-2 have the same identity, for example, it is necessary for the user to ask separately for the identity of each organism, then bo compare the answers to these questions.</Paragraph> <Paragraph position="9"> Scope of MYC~N'S Explanation Capability The ourpose of the explanation system is to enable a user to see how MCIN makes decisions, both in general and with reference to a particular consultation. To make this facility as useful as possible, we have trled to antidpate a11 types of questions whlch a user rnlpht ask, and to make every part of the system's khowledge base and reaqoning process accessible through clear explanations, The entire explanation facility consists of a number of components or wspecialistsm each capable of giving a single type of explanation. These components are grouped into three sets: one for explaining what the system is doing at a given time, one for answering questions about the system's static knowledge base, and one for answering quest ions about the dynamic knowledge base. The fir& set forms MYCIN 's reasoning-status checker ; the second and third together make up the system s general question-answer.</Paragraph> <Paragraph position="10"> Whenever MYCIN asks a question, the user is allowed to interrogate the status of MYCIN'S reanoninq chain bv asking WHY this piece of information is important. As explained in sections 2. 1 and 2.2, the system asks a question in order to find out about its curpent goal. Conslder the portion of a history tree shown in Figure 4. HOSPITAL-ACQUTRED is one subgoal, CATEGORY 13 another at the next level up, atid RULE037 links them, The ffreasonff for adkinq whether the infection was hospital-acquired, then, is based on an attempt to use this rule to determxne the hi~her subgoal. The answer to WHY at this pant is simply a display of the goals and the rules linkin8 them. We have given considerable attention to rnakin~ this as lucid a display as possible. First, since mwhvrf has several other interpretations, and no more extensive natural languase processinq is used in this component of MYCIN'S EC, the question is expanded to indicate the sense in which it will be answered. The higher goal is then presented, followed bv the rule. The clauses in the PREMISE of the rule are divided into those already esf.ablished and those yet to be determined. Finally, since rules may have multiple conelusions about different clinical parameters, the relevant conclusion is presented first and all others fbllow.</Paragraph> <Paragraph position="11"> As Fiaure 5 illustrates, further sections of the reasoning chain can be examined by repeating the WHY command. For any of the subgoals mentioned in answer to a WHY, the user may ask HOW this qoal was (or will be) achieved.</Paragraph> <Paragraph position="12"> MYCIN 's reasoning-status checker is described in more detail in b . [preceded by the first 14 questions in the consultation] .</Paragraph> <Paragraph position="13"> 15) Is the patient's illness with ORGANISM-1 a hospital-acquired infection? ** WHY [i.e. WHY is it important to determine whether or not the infection with ORGANISM-1 was acquired while the patient was hospitalized? ] f l .Ol This will aid in determining the category of ORGANISM-!.</Paragraph> <Paragraph position="14"> It ha3 already been established that [1.1] the gram stain of ORGANISM-1 is gramneq, and [I .2] the morphology of ORGANISM-1 is rod, and [1.3] the aerobicity of ORGANISM-1 is facultative Therefore, if [ 1.41 the ineection with ORGANISM-1 was not acquired while the patient was ho,s~ital ized then there is strongly suggestive evidence ( 8) that the cateaory of ORGANISM-1 is enterobacter iaceae ALSO: there is weakly dugqeetive evidence (. 1) that the identity of</Paragraph> <Paragraph position="16"> [i.e. WHY is it important to determine the category of ORGANISM-I?] [2.0]. . . in brder to determine the identity of ORGANISM-1 It has already been established that C2.11 this blwd culture was taken from a sterile source Therefore, if [2.2] this current organism and at least one of the list of members associated with the category of the orqanism agree with respect to the following properties: air conformation then There is strongly suggestive evidence (.9) that each of them is the identity of ORGANISM-1 [RULE003 1 ** HOW 1.3 [i.e. HOW was it established that tke aerobicitv of ORGANISM-1 is facultative? 1 The following were used: [ 3.1 1 RULE027 ndicat ed the~e is stronqly sugqest ive evidence ( .8 ) that the aerobicLty of ORGANISM-1 is facultative [ 3.23 RULE027 indicated there is weakly suggestive evidence ( .2) that the aerobicity of ORGANISM-1 is anaerobic Since this gave a cumulative CF of ('8) for facultative, and (.2) for anaerobic, it has been established that the aerobicity of ORGANISM-1 [user entries follow the double asterisks] The question-answering part of the systerll has natural-lanqusqe routines for analyzinq the user's input. The syatem recoanizes questions phrased in a number of ways, thereby makinq the q~est~ion-answering facil-ltv easier to use. Questions about the static knowledge base may deal with judgmental knowledge (e.g., which rules use or conclude a certain piece of information) or they may ask about factual knowledge -- entries in tables and lists. Some questions about static knowledge are shown in Figure 6.</Paragraph> </Section> <Section position="10" start_page="0" end_page="0" type="metho"> <SectionTitle> IS BLOOD A STERILE SITE? WHAT ARE THE NONSTERILE &SITES? WHAT ORGANISMS ARE I IKEL'Y TO BE FOUND IN THE THROAT? IS BACTEROIDES AEROBIC? WHAT METHODS OF COLLECTING SPUTUM CULTURES DO YOU CONSIDER? WHAT DOSAGE OF STREPTOMYCIN DO YOU GENERALLY RECOMMEND? HOW DO YOU DECIDE THAT AN ORGANISM MIGHT BE STREPTOCOCCUS? WHY DO YOU ASK WHETHER 'THE PATIENT HAS A FEVER OF UNKNOWN ORIGIN? WHAT DRUGS WOULD YOU CONSIDER To TREAT E.COLI? HOW DO YOU USE THE SITE OF THE CULTURE TO DECIDE AN ORGAN ISM'S IDENTSTY? </SectionTitle> <Paragraph position="0"> FQure 6. Sample Questions about MYCIN'S Static Knowledge Perhaps the more important art of the question-answering svstem is its abllity to answer questions about a articular consultation. While some users may be interested In checklng the extent of MYCIN'S static knowledge, most questions will ask for a justification of, or for the rationale behind, particular decisf on3 which were made during the consultation. Outlined in Flgure 7 are the types of questions about dynamlc knowledqe which can b* handled at present. A few examples of each type are given. <Cntxt> indicates some context which was discussed in the ~nnsultatlon; <parm> is some clinical parameter of this context; <rule> is one of the system's decision rules.</Paragraph> <Paragraph position="1"> Before a question can be answered, it must be classif$ed a3 belonginq to one ,of these groups. As Figure 7 illustrates, each question type includes a variety of ways in which the question can be worded, some specifying the parameter's value, some phrased in the negative, and so forth. MYCIN'S natural-language Drocessor must classify the questions, then determine what cllnLcal parameters, etc. the question references.</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2,4 Understanding The Question </SectionTitle> <Paragraph position="0"> The main emphasis in the development of the MYCIV system has been the creation of a production system which can provide sound diaqnoatic and therapeutic advice in the field of infectious disease. The explanation sfatem was included in the system's original desiqn in order to make the consultation program's decisions acceptable, justifiable, and instructive.</Paragraph> <Paragraph position="1"> Since the question-answerina facility was not the primary focus of the research, it is not designed to be a sophisticated natural-lanquage understander. Rather, it uses crude techniqyes, relying strongly on the very specific vocabulary of the domain, to ll~nderstand~~ whst information is being requested.</Paragraph> <Paragraph position="2"> The analysis of a question is broken into three phases: the first creates a list of terminal or root words; the second determines what type of question is beinq asked (see the classification of questions in Section 2.3); and the last determines what particular parameters, lists, etc. are relevant to the question.</Paragraph> <Paragraph position="3"> In the first and last steps, the system dictionary is important, The dictionary Contains approximately 1400 words that are commonly used in the domain of infectious disease. It ificludes all words that are acceptable values for a parameter, common synonyms of these words, and words used elsewhere by the system in describing the parameter (e.q., when translating a rule into English or requesting the value of the parameter).</Paragraph> <Paragraph position="4"> Reducing the Question to Terminal Words Each word in the dictionary has a synonym pointer to its terminal word (terminal words point to themselbes). For the purpose of analyzing the question, a non-terminal word is considered to be equivalent to its ( terminal) aynonym.</Paragraph> <Paragraph position="5"> Terminal words may have properties indicating: 1) that this word is an acceptable value for some clinical parameter(s) 2) that this word always implicates a certain clinical parameter, system list, or table (e.g. the word &quot;identitvll always implicates the parameter IDENTITY, which means the identity of an organism) 3) that this word might Xmplicate a certain parameter, system list, or table (e.3. the word tlpositiven might implicate the parameter NUMPOS, which means the number of positive cultures in a series) 4) that this word is part of a phrase which can be thought of as a sinqle word (examples of such phrases are &quot;transtracheal aspirationw, Ithow longw, and llnot sterilev1. The first three properties are actually inverse pointers which are generated automatically from properties of -the clinical parameters, Specifically , a word receives the &quot;acceptable valuew pointer to a paramster (property (1) above) ilp it appears in the parameter's list of acceptable values -- a list which is used during the consultation to check the user's response to 3 request for the parameter's value. Also, each clinical oarameter, list, and table has an associated list of key words that are commonly used when talking about this parameter, list, or table. These words are divided according to hoM sure we can be that a doctor is referring to this parameter list, or table when the articular word is used in a question. It is from this list that terminal words nimplicationn pointers (~Papertiea 2 and 3 in Table 1 ) are generated.</Paragraph> <Paragraph position="6"> During the first phase of parsing, each word in the original text is replaced by its terminal word. For words not found in the dictionary, the system uses Winograd's root-extraction aleorithm 181 to see if the word's lexical root is in the dictionary (ems., the root of tfdecisionw is &quot;deciden). If so, the word is replaced by the terminal word for its root. Words st111 unrewgnized aFter root extraction are left unchanqed. The resulting list of terminal and unrecognized words is then passed to a function which recognizes phrases. Usinq propertv 4 (see Table 1) of the terminal words in this list, the function identifies a phrase and replaces it with a single synonymous terminal word (whose dictionary properties may be important in determining the meaning of the question). The next step is to classify the question so that the proqram can tell which specialist should answer it. Since all auestions about the consultation must be about some specific context, the system requires that the name of the context (e.g., ORGANISM-1) be stated explicitly. This ~ives an easy way to separate qeneral questions about the knowledqe base from questions about a particular consultation. Further classification is done through a pattern matchinp; approach siailar to that used by Colby [I 1. The list of words created by the first phase is tested against a number of patterns (about 50 at present). Each pattern has a list of actions to be taken if the pattern is matched. These actions set flags which indicate what type of question was aaked. In the case of questions about judqental knowledge (called rule-retrieval questions), pattern matching also divides the question into the part referring to the rule's PREMISE and the part referring to its ACTION. For example, in &quot;How do you decide that an organism is streptococcus?~, there is no PREMISE part, and the ACTION part is &quot;an organism is streptococcus~; in &quot;Do you ever use the site of the culture to determine an organism's identity?&quot;, the PREMISE part is &quot;the site of the culturew and the ACTION part is &quot;an organism's identityn.</Paragraph> <Paragraph position="7"> The classification of a question guides its further analysis, Each question type has an associated template with blanks to be filled in from the question, The different blanks and the techniques for filling them in are listed in Table 2. With the question correctly classified, the general question-ans*erer can tell which soecialist should answer it. Fillin5 in all blanks in the template gives the s~ecialist all the information needed to find the answer, 1) <cntxt> - The context must be mentioned by name.</Paragraph> <Paragraph position="8"> 2) <rule> - Either a rule's name (RULE047) will be mentioned, or the word &quot;rulen will appear, toqether wiLh the rule 'a number ( 47).</Paragraph> <Paragraph position="9"> 3) <value> - One of the terminal words in the question has a diationary property indicating that it is a legal value for the parameter (property 1, Table 1 -- e ,q. THRDAT is a legal value for the parameter SITE).</Paragraph> <Paragraph position="10"> 4) <parm> - All of the words in the list are examined to see if they implicate any clinical parameters, Strong implicatians come from words with properties showinrf that the word is an acceptable value of the parameter, or that the word always implicates that parameter (properties 1 and 2, Table 1). Weak implications come from words with nroperties showing that: they might implicate the parameter (property 3, Table 1). The system uses an empirical scoring mechanism for picking out only the most likely parameters.</Paragraph> <Paragraph position="11"> Associated with certain parameters are words or patterns which must appear in the question in order for the parameter to be implicated. This scheme allows the system to distinguish related parameters which may be implicated by the same key words in the first pass. For example, the word nPMNvf implicates parameters CSFPOLY (the percent of PMNs in the CSF) and PMN (the percent of PMNs in the complete blood count). These are distinguished by reauirinq that the word &quot;CSFn be present in a question in order for CSFPOLY to be implic.at ed .</Paragraph> <Paragraph position="12"> 5 <list> - System lists are indicated in a manner slmilar to parameters, except that scoring is not done.</Paragraph> <Paragraph position="13"> Lists, like parameters, ,may have associated patterns which must be present in the question. Furthermore, lists h~ve properties telling which other system lists are their subsets. If a question implicates both a list and a subset of that list, the more general (larger) list is discarded.</Paragraph> <Paragraph position="14"> As an example, the question &quot;Which drugs are aminoglyco~ides?~~ implicates two lists: The list of all drugs and the list of drugs which are aminoglycosides. The system only considers the more specific list of aminoglycosides when answering the question.</Paragraph> <Paragraph position="15"> 6) <table> - Tables are indicated in a manner similar to lists except that an entry in the table must also be resent in the question. For exam~le, the word norganismw may indicate two tables: one containing a classification of organisms, and the other containing normal flora of various portals. The question &quot;What organisms are considered to b'e subtypes of Pseudornona~?~' will correctlv implicate the former table, and &quot;What are the organisms likely to be found in the throat?ft will implicate the latter, because FSEUDOMONAS is in the first table and THROAT is in the second.</Paragraph> <Paragraph position="16"> whether the organism is a contaminant 6, 31, 351, 39, 41, 42, 44, 347, 49, 106 kich do you wish to see? ** 6 --ow---If: 1) The culture was taken from a sterile source, and 2) It is definite that the identity of the organism is one of: staphy7~coccus-coag-neg bacillus null subt ilis corynebacter ium-non-diphtheriae Then: There is stronqly suqqestive evidence ( .8) that the orqaniqm is a contaminant [User input follows the double asterisks. 1 [ 1 1 The quest ion is reduced to a list of terminal words.</Paragraph> <Paragraph position="17"> C21 Pattern matchinq classifies the question as a rule-retrieval question, and divides it into a premise part and an aetfon part [3] Dictionary oroperties of the temnfna1 words are used to determine which parameters (and their values) are relevant to each part of the question. These vocabulary clues are listed in the form (<parm> (<values>) weight) where weight is used by the scoring mechanism to determin which parameters should be eliminated from consideration, [ 41 After selecting only the most strongly indicated parameters, the final translation tells what rules can answer the question: there are no restrictions on the PREMISE, and the ACTION must contain the parameter CONTAMINANT (with any value).</Paragraph> <Paragraph position="18"> [5] The answer con8fsts of findinp; all rules which meet these restrictions, and printing those that the user wants to see.</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.5 Answering the Question </SectionTitle> <Paragraph position="0"> Corresponding to each question type, there are a number of' possible answer templates. For example, for questions of the fom ttHow do you know the value of <parm> of <cntxt>?&quot;, two of the answer templates are: I used <rule> to conclude that <parm> of <cntxt> is <value>, This gave a cumulative c.f. of (certainty factor), The last question asked before the conclusion was made was (question number>.</Paragraph> <Paragraph position="1"> In answer to question (question number) you said that <parm> of <cntxt> is <value> The specialist for answering ,questions of a given tvpe will need to check the history tree or the system s knowledge base in order to determine which of the answer templates is appropriate for a particular question, Some blanks in the answer template will be filled in by the same items that filled blanks in the question tern~late. The remainder are filled by the specialist with information which will answer the question. In the above example, the slots <parrn>, <cntxt>, and possiblv <value> would be filled in from the question, and the other slots would be fllled from the historv tree.</Paragraph> <Paragraph position="2"> General questions about MYCIN's factual knowledqe are the easiest to answer. The specialist that provides answers to such questions must look up the desired information in the system's static knowledge base. Generally, an answer consists of simply displaying this information. As Figure 9 illustrates, the most processing that is ever done is to check the retrieved information against some item in the question (for eauality of set membership).</Paragraph> <Paragraph position="3"> ** IS BLOOD A STERILE SITE? Yes. Blood is one of those sites that are normally sterile.</Paragraph> <Paragraph position="4"> The specialist thzt answers questions about j~dqUIenta1 knowledge is slightly !nore complicated. Answering these questions (Flqure 10) involves built-in knowledge about the rule set, plus the ability to tlreadll the rules. By the time the question has been analyzed, the mecialist knows exactly which parameters must appear in the PREMISE, and which must apDear in the ACTION of any rule which answers the question.</Paragraph> <Paragraph position="5"> Values may be specified for any of the parameters. To answer the question, the rule-retrieval specialist must first find every rule whose PREMISE and ACTION satisfy these constraints. To do this, it needs to know that there are two special lists associated with each parameter: one contain in^ every rule that uses the parameter in its PREMISE, and the other containinq every rule that concludes about the parameter in its ACTION. Using these lists for the various parameters mentioned in the quest ion, the specialist can find those rules that might answer the question. If no values were soecified, the job is done and the relevant rules can be displayed without further anslvsis; otherwise, it is necessary to read each of the rules in the list and to eliminate those which do not mention the correct values for the ~arameter, The rule-retrieval specialist also makes use of a piece of MYCIN'S knowledge which was not discussed earlier. The system contains models of its own knowledge (called rule models) which are used primarily during acquisition of new medical knowledge from an expert [4]. These models, however, can be put to many uses -- one is to explain aeneral patterns in decision making. The rule models are abstract descriptions of a subsets of rules and are generated automatically by readinq the rules. For examole, the model for IDENT-IS-PSEUDOMONAS tells what features are common to the majority of rules which conclude that the identity of an oraanism is pseudomonas .</Paragraph> <Paragraph position="6"> If a model exists describing the rules about which the question is asking, the rule-retrieval soecialist incorporates this model's information into its answer (Figure 11). Thus the question-answering facility is able to give some information about strategies for achieving some of the svstem's goals, as well as the individual rules which use the strateqies.</Paragraph> <Paragraph position="7"> ** WHY DO YOU ASK WHETHER THE PATIENT HAS A FEVFR OF UNKNOWN ORIGIN? The rules listed below use: whether the patient has a true fever of unknown ori~in The rules listed below use: whether the patient has had a senito-urinary manipulative procedure to conclude about: the identity of the organism 156, 163, 190 Wfiich do you wish to see?</Paragraph> <Paragraph position="9"> If: 1) The site of the culture is blood, and 2) The gram stain of the organism is gramneq, and 3) The morphology of the orqanism is rod, and 4) The portal of entry of the organism is urine, and 5) The patient has not had a qenito-urinary manipulative procedure, and 6) Cystitis is not a problem for which the patient has been treatel? Then: There is suggestive evidence (.6) that the identity of the orqanism 1s e. coLi Figure 10. -Rule-Retrieval Questions</Paragraph> </Section> </Section> <Section position="11" start_page="0" end_page="0" type="metho"> <SectionTitle> ** HOW DO YOU DECIDE THAT AN ORGANISM MIGHT BE PSEUDOMONAS AERUGINOSAQ </SectionTitle> <Paragraph position="0"> Rules which conclude that the identity of the organism is pseudomonas-aeruginosa generally use one or more of the following pieaes of information: the sfte of the culture the Gram stain of the organism the morphology of the oreanism Furthermore, the following relationships hold: The gram stain of the orsanism, and the morphology of the organism tend to appedr together in these rules. RULE184, RULE1 16, RULE047, RULE085, RULE040 conclude that the identity of the organism is pseudomonas-aeruqinosa. Which of these do you wish to see?</Paragraph> <Paragraph position="2"> If: 1 ) The category of the organism is not known, and 2) The aram stain of the orsanism is qrarnneg, and 3) The morphology of the organism is rod, and 4) The aerobicity of the organism is facul Then: There is weakly sugqestlve evidence (.I) that the identity of the orqanism is pseudomonas-aeruqinosa Figure 1 1. Question Which Uses Rule Models One of the simplest quest ions about a spec if ic consultation inquires about the value of a particular parameter (Figure 12). The specialist which answers these questions must know how to retrieve this information, In soae cases it will also be necessary to make comparrsons to see whether a value specified in the auestion is one of the values which was deduced for that parameter.</Paragraph> <Paragraph position="3"> Answerinq other tllrpes of consultation-specific questions involves the use of knowledge about the control strbcture, including: 1) how the system acquires information 2) why the system tries to find out about parameters 3) how information is used once it is acquired 4) what causes a rule to be tried 5) what causes a rule to fail, The specialist for answering questions like !!How do vou know the value of <parm> of <cntxt>?w, knows that the value of a narameter can come from two sources: it can be deduced by rules, or the user can tell us about it in response to a question. The history tree (see Section 2.2) will show whlch (possibly both) of these sources provided the particular informatton mentioned in the question (Figure 13).</Paragraph> <Paragraph position="4"> ** HOW DO YOU KNOW THAT CULTURE-1 WAS FROM A STERILE SOURCE? I used rule189 to conclude that this blood culture was taken from a sterile source. This gave a cumulative c.f. of (1.0). The last question asked before the conclusion was made was 7, ** DID YOU CONSIDER BACTEROIDES AS A POSSIBILITY FOR ORGANISM-I? Yes. I used rule095 to conclude that the identity of ORGANISM-1 is bacteroides. This gave a curdulative c.f. of (,7). The last question asked before the conclusion was made was 20. ** HOW DO YOU KNOW THAT ORGANISM-1 IS E.COLI? The following rules made conclusions about whether the identity of ORGANISM-1 is e.coli cumulative certaimy last quest ion asked If the question is phrased in the negative, it is neeessarv first to find all the ways the conclusion could have been made (this is a simple task of rule-retrieval), then to explain why it wasn't made in this consuLtation</Paragraph> <Paragraph position="6"> The following rules could have b?en used to determine that the identity of ORGANISM4 is streptococcus : RULEO33. However, none of these succeeded in the context of ORGANISM-1.</Paragraph> <Paragraph position="7"> If you would like an explanation for why any of these rules failed, please enter their numbers: *# 33 Clause 2 of rule033 [&quot;the morpholoqy of the orqanasm is coccusw] was already known to be false for ORGANISM-1, so the rule was never tried.</Paragraph> <Paragraph position="8"> ** WHY DON 'T YOU THINK THAT THE MORPHOLOGY OF ORGANISM-1 IS COCCUS? It is definite that the morphology of ORGAVISY-1 is rod. Knowing this with certainty rules out all other values for the the morphology of 0RQ.ANISM-1, includinq coccus.</Paragraph> <Paragraph position="9"> Figure 14. Questions Regarding Why a Conclusion wasn't Made The specialist for answerinq questions of the fom &quot;How did vou use <parm> of <cntxt>?IV, needs to know not only how to find the svecific rules which might use a parameter, but also how a ~arameter can cause a rule to fail 2nd how one parameter can prevent another from being used. The history tree can be checked to see which of the relevant rules used the oarameter, which failed because of the parameter, and which failed for some other reason, preventin@ the parameter from being used (Figure 15). ** HOW DID YOU USE THE AEROBICITY OF ORGANISM-17 The aerobicity of ORGANISM-1 was used in the following rules: RULE003, RULE005, RULE084.</Paragraph> <Paragraph position="10"> The aembicity of ORGANISM-1 caused the following rules to fail: RULEO35, RULEO51, RULE052, RULEO'i3, RULE111c Thc aerobicity of ORGANISM-1 also would have been used in: RULE037, RULEOSO. RULE058, RULE086, RULE110, RULE184, RULE203, RULE204, RULE205. However, none of these succeeded In the context of ORGANISM-1. If you would like an explanation For why any of these rules failed, please enter their numbers: ** 37 Rule037 was tried in the context of ORGANISM-1, but it failed due to clause 1 [&quot;the categorv of the organism is not</Paragraph> <Paragraph position="12"> The fact that Fred Jones is not a compromised host caused the following rules to fail: RULE085, RULE106.</Paragraph> <Paragraph position="13"> The fact that ~rea Jones is not a compromised host also would have been, used in: RULE109. Howeveri none of these succeeded in the context of PATIENT-1.. If you would like an explanation for why any of these rules failed, get as much information as is desired. The first answer given was not really complete In itself, but it led the user to ask another question to yet more information. Then another questlgn was asked to determine why clause 1 or RULEI~~ was false. The answers to the first two questions both mentioned rules which could be printed if the user wanted to examine them. The special command PR (Print ~ule) ig for the user's convenience. It requires no natural-languaqe processing and thus can be understood and answered immediately. (What is rule or *'Print rule 109 .&quot; also would have been</Paragraph> <Paragraph position="15"> During t be preceding consultation, I never had reason to try to find out whether the results of the CBC from the time period of this blood culture are available. It would have been necessary to know this information in order to use any of the following: RULE109. However, none of these succeeded in the context of this blood culture. If you would like an explanation for why any of these rules failed, please enter their numbers: ** 109 Rule109 would be relevant to this blood culture when ap~lied to: ORGANISM-1. I never had reason to trv to apply rule109 to ORGANISM-1. I would have needed to use rule109 in order to find out whether ORGANISM-1 is a contabinant. However, during the preceding consultation, 1 never had reason to trv to find out whether ORGANISM-1 is a contaminant.</Paragraph> <Paragraph position="16"> ** WHY DIDN'T YOU NEED TO KNOW WHETHER ORGANISM-1 IS A COWTAMINANT? During the preceding ~onaultation, 1 never had reason to try to find out whether ORGANISM-1 is a contaminant. It would have been necessarv to know this information in order to use any of the followinq: RULEOO6, RULE007, RULE906, RULEt08, RULE109, RULl3159. However, none of these succeed'ed in the context of ORGANISM-1. If you would like an explanation for why any of these rules failed, please enter their numbers: If: 1) The cuJture was taken from a sterile source, and 2) Xt is not known whether tb e orqanism is a contaminant, and 3) The patient is not 9 compromised host, and 4) The patient was febrila due to the infection, and 5) The results of the cbc from the time ~eriod of the culture are available, and 6) A - The white count from the CBC is qreater than 10.5, or B - The percent of pmn's in the cbc at the time the culture was obtained is qreater than 78, or C - The percent of wbc's which were bands in the CBC at the time the, culture was obtained is qreater than 10 Then: There is strongly suggestive evidence (.a) that the orqanism is not a contaminant Figure 16. Questions Regarding Why a Parameter wasn't Traced In questicns asking about the application of a rule to a context there are three poss*~bilities: the rule told us somethinq about the context; the rule failed when applier! to that context; or the rule was never tried in that context, The historv tree tells which of these ia the case, Furthermore, if a rule succeeded, there is a record of all the conclusions it made, and if it I&quot;3ileb, the reason for failure is recorded. 9s Fiqure 17 illugtrates, rules are only applicable in certain contexts. If the rule could not apply to the context mentioned in the question, but mi~ht be relevant when applied to soae other context [which is an ancestor or descendent in the context tree], then the question is answered as if it had mentioned the relevant context.</Paragraph> <Paragraph position="17"> There are interactions among the different types of questions which weren't mentioned earlier for the sake of clarity. Sometiaes a question ray indicate an incorrect assumption about the sate of MYCIN'S knowledge. For examole a question could ask &quot;Whv don't you think that ORGANISM-1 is E.coli?&quot; when, in fact, the system has concluded that the organism is E.coli. To answer this question, the explanation system would explain how it did make such a decision, despite the ~hrasinq of the question. Similarlv, when asked how the system used (or concluded about) some mrameter that was never traced, the system explains why it didn 't need to find out about the parameter, reflardless of how the question was bhrased.</Paragraph> <Paragraph position="18"> ** DID RULE 116 TELL YOU ANYTHING ABOUT INFECTION-I? ~ulell6 would be relevant to the primary-bacteremia when applied t: ORGANISM-1 Clause 5 of rulel 16 [&quot;the infection is meninqitisn] was already known to be false for ORGANISM-I, so the rule was never tried.</Paragraph> <Paragraph position="19"> ** HOW WAS RULE. 178 HELPFUL WHEN YOU WERE CONSIDERING ORGANISM-13 ~ule178 would be relevant to ORGARISM-1 when applied to: CULTURE-I I never had reason to try to apply rule178 to this blood culture. X would have needed to use rule178 in order to find out whether this blood culture was taken from a sterile source.</Paragraph> <Paragraph position="20"> However, I was already certain that this blood culture was t&ken from a qterile source, ** WHAT DID RULE 295 TELL YOU ABOUT ORGANISM-I? I used rule295 to conclude that the identity of ORGANISM-1 $3 hemophilus-influenzae. This gave a cumulative c.f. of (25).</Paragraph> <Paragraph position="21"> The last question asked before the conclusion was made was 36.</Paragraph> <Paragraph position="22"> +* WHY DIDN'T YOU USE.RULE112 TO FIND OUT ABOUT ORGANISM-I? RULE1 12 was not executed because it would have caused circular reasoning when applied to ORGANISM-?. Would you like ta see the chain of rules and parameters which makes up this circle? ** YES 1 wanted to know about the identity of ORGANISM-1 -because I try to find out the identitv of the organism for all current organisms of the patient.</Paragraph> <Paragraph position="23"> To find out about the identlty of ORGANISM-1, I tried to use ruleO21. Before I could use rule021, I needed to know about a prior orqanism with possibly the same identity as ORGANISM-1 .</Paragraph> <Paragraph position="24"> To find out about a prior organism with possibly the same identity as ORGANISM-1, I trred to use rule005. Before I could use rule005, I needed to know about the aerobicity of ORGANISM-1 .</Paragraph> <Paragraph position="25"> To firid out about the aerobicity of ORGANISM-1, I tried to use rule031. Before I could use rule031, I needed to know about the category of ORGANISM-1.</Paragraph> <Paragraph position="26"> To find out about the category of ORGANISM-1, I tried to use rulel 12. Before I could use rulel 12, I needed to know about the identity of ORGANISM-1 .</Paragraph> <Paragraph position="27"> But this is the unknown parameter I sought originally.</Paragraph> <Paragraph position="28"> Figme 17. Question Regarding the Application of a Rule</Paragraph> </Section> <Section position="12" start_page="0" end_page="0" type="metho"> <SectionTitle> 3 Conclusions </SectionTitle> <Paragraph position="0"> Consultation system8 which give expert advice in some domain form one class of artificial intelligence programs which can provide useful solutions to real-world ppoblems. The utility of such a system, however, depends on its acceptability to human users. One feature which can Increase a system's acceptability is a mechanism whereby the system can explain or justify its advice.</Paragraph> <Paragraph position="1"> The development bf an explanation mechanism for a consultation system is very much related to the problems of representing knowledge and of making use of different sources of knowledge, Since the production system formalism provides a unified way to represent modular pieces of knowledae, the task of designing an explanation capabilitv is simplified for production-based consultation systems, The example of MYCIN shows how this can be done and illustrates further that a svsterll desiqned for a single domain with a small, technical vocabulary can give comprehensive answers to a wide range of quegtions without sophisticated natural-language processing.</Paragraph> </Section> <Section position="13" start_page="0" end_page="0" type="metho"> <SectionTitle> Acknowledgments </SectionTitle> <Paragraph position="0"> The authors wish to express their qratitude for the inbrest and advice of Drs. Bruce Buohanan and Cordell Green (Computer Science Department). We also are indebted to the followinpr MYCIN Project coworkers: Jan Aikina, Stanton Axline, Stanley Cohen, Larry Fagan, Frank Rhame, Bill van Melle, Sharon Wraith, and Victor Yu. Special thanks are due to Bill van Melle and Bruce Buchanan who made numerous helpful comments on earlier drafts of the paper.</Paragraph> </Section> <Section position="14" start_page="0" end_page="0" type="metho"> <SectionTitle> NASHINGTON DEVELOPMENTS SUPREME COURT DENIES CBCT PETSTIONS; COMPTROLLER CURRENCY TO ISSUE NEW CBCT &GULATIONS </SectionTitle> <Paragraph position="0"> Consolidated petitions from two Chicago banks and the Comptroller of the Currency seeking review of Pedeyal Appeals Court rulings which equate customer-bank conlmunications terminals (CBCTs) with bank branches, and thus forbid CBCTs in nonbranching states, were denied by the Supreme Court last month ttwithout dissent or explanation .&quot; Banks ' response. The petitioning banks, Gontinental Natiofial Brink and Trust Co. of Chicago and the First National Bank of Chicago, this month deactivated their combined total of seven CBCTs . The terminals' three uses permit deposits to be made, funds to be transferred, or funds to be borrowed; each of the5e functions, various appea-1s courts have held, constitute branch banking under the National Bank IMcFadden) Act.</Paragraph> <Paragraph position="1"> New CBCT regutatiom. The Currency Comptrollerts Attorney Edward Jiran told AFIPS P~ehington Report in October that the Comptroller will issue new regulations for nat%onal bank CBCTs this month.</Paragraph> <Paragraph position="2"> Rules may be written to liberalize minimum capitalization requirements for terminals shared by more than one financial institution, according to the Comptroller s Electronic Banking Systems Division Chief Claude A. Raworth.</Paragraph> <Paragraph position="3"> Pawortb told AFIPS that the new regulations will permit banks Itto use the [EFTS] technology without heavy capitalization requirements; [otherwise, 1.</Paragraph> <Paragraph position="4"> some thinly capitalized [banks] would have problems .&quot; Thus, according to the Currency Comptroller official , the regulations would encourage sharing EFTS facilities with smaller banks.</Paragraph> <Paragraph position="5"> NCEFT'and ZegisZatim acti~ties. Prior to the Supreme Court action denying the CBCT petitions, National Commission on Electronic Funds Transfer (NCEFJ') Executive Director John B. Benton told this writer in Washington that the braaching issue should be resolved in the U.S. Congress, Benton added that &quot;it's going to be some time before 1egi.slation is passed in CongressIt on whether CBCPs are branches.</Paragraph> <Paragraph position="6"> Last month, the NCEFT convened hqar~ngs osl both brahching and corlsumer issues. The Senate Subcommittee on Financial ~hsstitutions, chaired by Sen. Thomas dr McIntyre ID-N .H .) , wi 11 hold hearings on the branching question in December. CBCT court action. Fedgral courts in various jurisdictions have bean almost unanimous in their interpretation of the McFad&n Act disallowing off-premises CBCTs. Last August, the U.S. Court of Appeals for the Tenth Circuit, ruling in a Colorado case, upheld a lower district court judgment that receipt of deposits by CBCTs vialates the McFadden Act.</Paragraph> <Paragraph position="7"> The U. S. Court of Appeals for the Eighth Circuit, in a separate case, last July upheld a district court ruling in Missouri that the First National Bank in St. Louis must remove CBCTs at two St. Louis County locations. The district court said that any one of the three major services provided by CBCTs constituted branch banking, which is limited in Missouri.</Paragraph> <Paragraph position="8"> In the cases of the two petitioning Chicago banks, decided by the Federal Court of Appeals in Illinois last May, all functions performed by CBCTs were deemed to constitute branch banking, which is forbidden in 11,Jinoi.s.</Paragraph> <Paragraph position="9"> However, a Federal District Court in Oklahoma approved all three uses of CBCTs, finding terminals &quot;only processors ,Iv and as such legal. An appeal of the case was mooted when the. state legislature authorized legislation aaactioning the use of CBCTs.</Paragraph> <Paragraph position="10"> GOVERNhIENT OPERATIONS COMMITTliE CRITICIZESNONCOMPETITIVljG PROCUREMENTS Following hekrings last June (Washinqton Re~mt, 8/76), the House Cornittee on Government Operations stated in October that &quot;the low level of fully competitive ADP promrements, coupled with the apparent lack of effective utilization of ADP resources will, if allowed to continue, ultimately result in seriously jeopardizing the effectiveness of the Brooks Act [P.L. 89-3061, at a cost of millions of dollars annually to the taxpayers.&quot; Reexam.inution of 'true -ct of computere ' suggee.ted. In a report, titled Administration of P. L. 89-306, Procwernnt of ADP Re8omes By ths FedemZ Covement (II,), the Committee said: &quot;If'the full benefits of the'Act are to be achieved, GSA, OMB, NBS, and the user agencies must join together in a commitment tb fully support and adhere to the provisions of the Act.&quot; The report suggested a reexamination sf &quot;the true impact of computers on government operations.&quot; It raised the question of whether &quot;acquisition of ADP resources is necessary to carry on an essential program, or whether such acquisition will foster non-essential activity because the computer is available.&quot; ADP ~t. . . receive attenth of top management. The Committee concluded that ADP Itcam no longer be viewed as a mere tool, but must now be treated in the same way as other major programs and, as such, receive top attention of top management.&quot; The report attributed the basic causes of noncompetitive procurements to a lack of: (1) adequate justifications for ADP acquisition; (2) long-range planning; (3) standards ; (4) high level languages ; (5) utilization ~euiews; and (6) use of functional specifications.</Paragraph> <Paragraph position="11"> OMB directs GSA to use vfbwtiomZ ADP specifiwtions*.</Paragraph> <Paragraph position="12"> Responding directly to the Committee report and other criticisms, OMB Dlrector James T. L letter last month to GSA Administrator Jack ECkerd, wrote that &quot;[t $a o preclude in a the adverse effects of unduly restrictive specifications, agsncies shall, to the maximum practical exrent, express their ADP requirements in terms of functional performance specifications rather than equipment specifications . &quot; Lynn added: &quot;GSA should strive for expeditious review of agency submission to avoid delays in the procurement process.</Paragraph> <Paragraph position="13"> Agencies should cooperate with GSA to expedite these reviews. An undue length of time for the normal ADP procurement process may be considered a disagreement by GSA with the request of the agency, and may be appealed to the Office of Management and Budget.</Paragraph> <Paragraph position="14"> fvConsistent with the poliw expressed herein, any such appeal will be handled expeditiously, but the burden will be upon the agency to demonstrate the need to specify the particular make or model of equipment or its functional equivalent when such specification is at issue,&quot; he said.</Paragraph> <Paragraph position="15"> The GSA has 60 days fi- October 6, the date of thb letter, to comply with the directive.</Paragraph> </Section> <Section position="15" start_page="0" end_page="0" type="metho"> <SectionTitle> HOUSE COMMUNICATIONS SUBCOWITTGE CHAIRMAN RECOMMENDS OVERHAUL OF 1934 COM~UNICATIONS ACT </SectionTitle> <Paragraph position="0"> A &quot;basement - to-at t icl' overhaul of the 1834 Federa2 Commmications Act, the statute which provides the basis for regulation of the nation's telecomunications systems, was proposed to the next Congress last month by Rep. Lionel Va Deerlin (D-Calif.), chairman of the House Communications Subcommittee.</Paragraph> <Paragraph position="1"> The Submcomittee has just completed three day* of hearings on the proposed Cone~wr Comnications Refon Act of 19F6 (Waehh-gton Report, lo/ 76) .</Paragraph> <Paragraph position="2"> Joined by Rep. Lou Erey.@-Fla.), Van Deerlin said the original Act is outdated. null According to the trade press, issues to be considered by the Subcommittee include : whether the FCC has established fair coppetiti:~.: ground rules; whether business telephone services really subsidize residential services; and what the implications of changing FCC policies are toward competition.</Paragraph> <Paragraph position="4"> President Ford last month signed into law the Tm Reform Act of 1076, recently passed by Congress (va~hinqtofl Repopt, 9/ 76 ) , which includes an amendment pemktting state governments to demand and retain an individual's Social Security Number [SSN) to aid in searches for parents not supporting their children. The amendment repeals part of the Acivacy Act of 1974.</Paragraph> <Paragraph position="5"> In addition, state and local governments can now use the SSN as an identifier for tax administrative programs, for drivers ' licenses, and for motor vehicle registration. However, government officials who disclose the nmber without authorization are subject to a fine o!f up to $1,000 or imprisonment of up to one year.</Paragraph> </Section> <Section position="16" start_page="0" end_page="0" type="metho"> <SectionTitle> NEWS BRIEFS </SectionTitle> <Paragraph position="0"> Rep. TiarWirth [D-Colo.) last month wrote Office of Teleconmnmications Policy ETP) Director ,momas Houser (who had reportedly c~ntacte~ Assistant Attorney General Donald Baker to discuss the &quot;economic rationale&quot; for the Governmpnt ' s antitrust action against ATBT) tliat Itany further contacts between OTP and the Departwent of Justice, concerning the [suit]. . .</Paragraph> <Paragraph position="1"> would be most iaappropriat e .</Paragraph> <Paragraph position="2"> The - FCC last month set new dates for providing comments on the second &quot;Computer InquiryH and rulemaking regarding the use of computers by common carriers jn -providing communciations or data processing services ; comments are now due by January 10, 1977; replies, by February 24, 1977.</Paragraph> <Paragraph position="3"> Responding to protests by -9 AT&T the - FCC last month revised registration specifications for modems to facilitate direct interconnection of modems to the telephone network.</Paragraph> <Paragraph position="4"> The House Government Operations Committee last month failed to report to the House floor a bill that would have authorized the Federal Assistance The Institute for Cowuter Sciences - and Technology - [ICST) of the National Bureau of Standards QBS) last month issued DisR Drive Interface CharmteAetics (# --enclose $5.50) ; the report was prepared by Auerbach Associates under NBS contract, and deals with charact-eristics of the interfaces between high performance disk drives and their controllers.</Paragraph> <Paragraph position="5"> The National Communi-cations System (NCS) last month announced proposed Federal standards for data communications interfaces; the proposed standards specify the general purpose electrical characteristics to be applied to data comrnunicatons interfaces.</Paragraph> <Paragraph position="6"> The National Science Foundation [NSF] last motlth said several of its staff members hyve been appointed to provide initial support for the Office of Science and Technology Policy (OSTP).</Paragraph> <Paragraph position="7"> The first Federal $oft?&&; Excchg.e CataZog, designed to promote the exchange and sharing of software among ~ederal AW units, is-scheduled to be released th$s January by the Agency Services Division of the General Services Administration (GFA) .</Paragraph> <Paragraph position="8"> The Bure,au of the Census last monthanounced the availability of a 584-page report, tit led Conputer Pwgrwns for Demographic Analysis (p --enclose $5. SO), first published last June.</Paragraph> <Paragraph position="9"> The Na'tionql Bureau of Standards (NBS) is holding a Bicentennial Conference on MathematkaZ Progdng November 29 through December 1 in Gaithersburg, Md.; the conference is being co-sponsored by NBS and the ACM Special -. Interest Group on Mathematical Programming (SIGMP).</Paragraph> </Section> <Section position="17" start_page="0" end_page="0" type="metho"> <SectionTitle> AFIPS IN WASHINGTON </SectionTitle> <Paragraph position="0"> A Planning Conference to provide technological background for the second FCC &quot;Computer Inquiry, tt organized by AFIPS in conjunction with the FCC (Wushinqton Report, 10/76) , was convened in Washington November 8-9. The conference focused on computer communications, especially as it relates to regulatory policy.</Paragraph> <Paragraph position="1"> Detailed coverage will appear in the December AFIPS Washington Report.</Paragraph> </Section> <Section position="18" start_page="0" end_page="0" type="metho"> <SectionTitle> WASHINGTON OFFICE ACTIVITIES DESCRIBED AT DPMA INFO/EXPO '76 IN IAS VEGAS AFIPS Washington Office Director Philip S. Nyborg last month chaired a panel </SectionTitle> <Paragraph position="0"> session on major activities of the Washington Office at the DPMA INFO/EXPO '76, in Las Vegas October 26. The session covered the Office's activities in such areas as privacy, EFTS, legal protection for softwere, computer security, data conununications , and technology transfer.</Paragraph> <Paragraph position="1"> The panel opened with a discussion of the historical perspective of the office and its outlook for the future by Mr. Keith Uncapher, chairman of the AFIPS Washington Activities Committee. Nyborg f o1,lowed with a description of the dai ly operations of the Office, outlining the manner in which comment has been organized by AFIPS on specific issues.</Paragraph> <Paragraph position="2"> Mr. William Moser, DPMA International Vice President for Industry and Government Liaison, described his role as DPMA liaison to the Washington Office. Moser emphasized his interaction with DPMA members h devs Isping comment on testimony , and he encouraged their continued participation.</Paragraph> <Paragraph position="3"> Dr. A. Michael Noll, formerly the staff member for computers in the White House Office of Science and Technology (OST), described past and present Presidential science advisory mechanisms. No11 ' s presentation also covered the new Office of Science and Technology Policy, and its potential relationship to the field of information processing.</Paragraph> <Paragraph position="4"> The AFIPS Washingtom Report is researched and written by Pender M. ' ~ccarter, Research Associate, AFIPS Washingtcn Office.</Paragraph> <Paragraph position="5"> AFIPS societies have permission to use material in the AFIPS Washington Report for their own publications, except where an article title appears with an clearance must first be obtained from the AFIPS Washington Office. Documents indicated by the symbol are available on request to the AFIPS Washington Office. Where price is noted, make checks payable to IrAFIPS.&quot;' appeal a September ruling by a Federal District Court Judge in Cincinnati that Federal Home Loan Banks can not provide computer services to savings and loan associations. Five of the 12 regional Home Loan Banks, which are or at one time have been suppliers of data processing services, are involved in the Cincinnati appeal.</Paragraph> <Paragraph position="6"> ADAPSO, United Data Procesahg obtain irr;junct;on. The original case was brought against the FHLBB and a Cincinnati Home Loan Bank by the Association of Data Processing Service Organizations (ADAPSO) and United Data Processing, Inc, In October, 1973. The remaining four Home Loan Banks joined the suit as intervenors in November, 1973. The district court ruling enjoined all five banks from offering data processing services beginning January 1, 1977. A modified stay of the injunction has been obtained by the FHLBB pending the present appeal.</Paragraph> <Paragraph position="7"> District court bZds that FehZ Rome Loan Banks aro not permitted to provide D In the original suit, ADAPSO argued that the Federal Home Loan Banks posed &quot;a form of unfair competition1' to private enterprise because they are exempted from taxat ion. William W. Fletcher, president of Midwest Advanced Computer Services, inc., Warren, Mlcigan, and ADAPSO member, has been quoted as saying that hls firm can not compete with the Federal Home Loan Banks I &quot;30 per cent lower&quot; rates. In addition, ADAPSO said that the PederaZ Rome Loan Bank Act of 2932 forbids banks from participating In nonbanking commercial activities such as those provided by computer service centers.</Paragraph> <Paragraph position="8"> Federal District Judge Carl B. Rubin, in his ruling last September, held that &quot;the providing of data processing services to member institutions is not an express power granted to the Federal Home Loan Banks or Federal Hone Loan Bank Board1' by the FedsmZ Home Loan Bunk Act. ADAPSO Executive Vice-President Jerome L. Dreyer has criticized the FHLBB for encouraging its members to market data processing services to the same banks which the FHLBB regulates.</Paragraph> <Paragraph position="9"> Further adverse court ruzings eouM proscribe future DP offerings by Home Loan Banks. ADAPSO Attorney Herbert E. Marks, who is handling the FHLBB case, told AFIPS Phshington Report that the New York Federal Home Loan Bank has ceased offering data processing services since the district court injunction was obtained. Marks noted that further adverse rulings against the Home Loan Banks could proscribe future data processing offerings by the seven other Federal Home Loan Banks. He also said that if the district court is upheld by the appeals court in Cincinnati, the Federal Home Loan Bank Board could appeal to the Supreme Court.</Paragraph> <Paragraph position="10"> A spokeswoman for the FHLBB in Washington told AFIPS that the FHLBB had no comment on the case since it is in litigation.</Paragraph> <Paragraph position="11"> She stated that the New York Federal Home Loan Bank ceased providing data processing services prior to the district court ruling, not after the injunction, as Marks contended.</Paragraph> </Section> <Section position="19" start_page="0" end_page="0" type="metho"> <SectionTitle> NEW CBCT REGULATIONS MAY SANCTION OPERATION OF ,NON-FEDERALLY REGUtATED EFTS TERMINALS </SectionTitle> <Paragraph position="0"> Acting Comptroller of the Currency Robert Bloom last month announced new customer-bank comntunicat ions t erkninal (CBCT) regulations for national banks.</Paragraph> <Paragraph position="1"> The new regulations allow national banks to apply for permission to establish CBCTs as branches in states where state-chartered banRs are permitted to establish branches. Bloom's action follows the Supreme Court's refusal to consider consolidated petitions from two Chicago banks and the Comptroller of the Currency seeking review of Federal Appeals Court decisions which equate CBCTs with bank branches and forbid CBCTs in non-branching states (Washington R6pPt, 11/76].</Paragraph> <Paragraph position="2"> The Comptroller has unsuccessfully argued in the courts that CBCTs are not branches, and are not forbidden in states which limit or disallow branching.</Paragraph> <Paragraph position="3"> The trade press suggests that the new regulations may sanction the operation of non-federally regulated EFTS terminals by retailers, shopping mall operatorsf owners of apartment complexes, or operators of transportation terminals. Thus, the effect of these regulations may be to extend the present list of EFTS ilprovidersff beyond financial institutions, i.e., to include what were formerly on13 EFTS users.</Paragraph> <Paragraph position="4"> The Comptroller's new regulations state that capitalization required for a CBCT branch may now be shared among the participants, thus lowering the cost of operating a CBCT, and presumably making CBCTs available to smaller banks with less capital (see Washington Report, 11/76). Also, in contrast to the $500 application fee for a traditional branch, the application fee for a CBCT branch is $200.</Paragraph> <Paragraph position="5"> In the section of the Acting Comptroller's statement, said to sanction operation of non-federally regulated EFTS terminal^, Bloom held that &quot;any CBCT which is not established by a national bank . . . is not a branch of a national bani, and not subject to the provisions oftf the applicable section of the federal bank regulation, the McFadden Act, which the courts have held forbids CBCTs in non-branching states .</Paragraph> </Section> <Section position="20" start_page="0" end_page="0" type="metho"> <SectionTitle> WHITE HOUSE ANNOUNCES APPOINTMENTS TO PRESIDENT'S COMMITTEE ON SCIENCE AND TECHNOLOGY - </SectionTitle> <Paragraph position="0"> The White House has announced appointments to the President's Committee on Science and-Technology [PCST), which will conduct a two-year review of science and technology as it relates to the Federal government.</Paragraph> <Paragraph position="1"> The Coanittee was created by the same legislation [WaehCngtm Report. 5/76) which established the new White House Office of Science and Technology Policy [OSTP).</Paragraph> <Paragraph position="2"> As previously announced, the chairman of the committee is Dr. Simon Raao, co-founder of TRW. Inc. Vice chairman is Dr. William 0. Baker, president of Bell Labs. Both Rmo and Baker were chairmen of cornittees organized last year by President Ford to plan for OSTP.</Paragraph> <Paragraph position="3"> Other members of the PCST are: Otis R. Bowen, governor of Indiana; W. Glenn Campbell, director of the Hoover Institute on War, Revolution and Peace, Stanford University; Edward E. David Jr., vice president of Gould, Inc., and former science adviser to President Nixon; Elizabeth H. Leduc, professor of biology, Brown University; Fritz J . Russ , president, Systems Research Laboratories, Dayton, Ohio; Charles P. Slichter, professor of physics, University of Illinois; Charles H. Tomes, professor of physics, University of California at Berkeley; and W. Bradford Wiley, chairman of John Wiley and Sons.</Paragraph> </Section> <Section position="21" start_page="0" end_page="0" type="metho"> <SectionTitle> FEDERAL ENERGY ADMINISTRATION ASSISTS WITH COMPUTERIZED CONSTRUCTION FORE,- CASTING SYSTEM </SectionTitle> <Paragraph position="0"> The Federal Engergy Administration (FEA) is assisting in developing a computer-based system to permit national and regional projections of construction labor needs in relation to energy development. The system will also provide long and short-range forecasts in local areas, FEA Administrator Frank G. Zarb said last month. It will be implemented late next year, Zarb added.</Paragraph> </Section> <Section position="22" start_page="0" end_page="0" type="metho"> <SectionTitle> NATIONAL SCIENCE FOUNDATION RELEASES GUIDE TO PROGRAMS REFLECTING APPROPRIATIONS </SectionTitle> <Paragraph position="0"> The NationaL Science Foundation INSF) last month released its Guide to Programs (#-- enclose $1.353 reflecting NSFts appropriations for FY 1977. Programs in theoretical computer science, software systems science, software engineering, intelligent systems, computer systems design, and special projects are described.</Paragraph> <Paragraph position="1"> Proposals for support are ordinarily assigned to the appropriate NSF division or office for review and evaluation.</Paragraph> <Paragraph position="2"> The theore ical computer science program encompasses the theory of computation, numerical analysis and computational ~nathematics , theory of formal languages, and analysis of algorithms. The software systems science program covers &quot;f~ndamental'~ questions of communicating with and controlling computer syszems. The software engineering program includes the methods, tools, and techniques for specifying, designing, and implementing 'lqualityl' software. The intelligent systems program covers computer-based systems which have such characteristics as pattern recognition, pattern generation and knowledge representation.</Paragraph> <Paragraph position="3"> The computer systems design program includes the principles of computer systems design such as: computer system architecture, performance, graphics, man-machine interaction and logic design. The special projects program encompasses research projects, studies, workshops, and other activities which '?miBht encourage the development of new fields of computer science research.&quot; Should future legislation result in some modification to the FY 1977 program, changes will be announced in the monthly ASP BuZZsth~.</Paragraph> <Paragraph position="4"> Program schedules, deadlines, updates, and availability of brochures describing individual programs ape also published in the BuZZetdn.</Paragraph> <Paragraph position="5"> It may be obtained free-of -charge by writing: Editor, NSF BuZZetin, Public Information Branch, NSF, Washington, D.C.</Paragraph> <Paragraph position="6"> 20550.</Paragraph> </Section> <Section position="23" start_page="0" end_page="65" type="metho"> <SectionTitle> NEWS BRIEFS </SectionTitle> <Paragraph position="0"> The at press time late last month reversed the Chief of the Common Carrier Bureau's rejection of ATqT's Dataspeed 40/4 filings (Washington Report, 4/76) ; the Commission concluded that the 40/4 service is &quot;not inconsistent1' with its existing computer rule; detailed covWaga will appear next month. FCC Chairmum Richard 8: wiiey early last mnth said the - FCC is-considering whether it has authority to ask the courts to modify the 1956 ATeT antitrust consent decree; the Justice Department has held that the consent decree bars AT4T from interstate marketing of the telephone company s Dataspeed 40h4 as &quot;incidental&quot; to regulated communications, The director of the White House Office of Telecomunications Policy @W] , Thomas J. Houser. last month criticized the Federal Reserve System for emandinn - . . - - - - . - .- -. - - - - . . . - - - - . - . . &quot; &quot;its operit ional role in interbank EFT with pilot pro j bctslt linki& automated clearing houses (ACHs) with telecolmnmications (Washington Report, 9/76) ; however, H. L . Baynes , president , National Automated Clearing House Association (NACHA], which is operating the project with the Federal Reserve, told this writer last August that the Fed has &quot;the right to develop an ACH processing capability [and to go] interregional without involving the private sector.</Paragraph> <Paragraph position="1"> The General Services Administratioq (GSA) last month amended its Mvaoy Gu--&de-Zines adding privacy ad security considerations for use in ADP or tqlecommunicat ionsT$st ems solicitat ioki and corntracts ; the amendments also require that agencies provide an inspection program in system specigications and contracts that will ?'ensure coptinuous, efficacious and efficient safeguards, and provide for the discovery and the countering of any new threats or hazards. &quot; The National science Foundation [NSJSP) last month said indtutrial spending for research and development reached $23.5 billion in 1975, up five per cent from the 1974 level of $22.4 billion.</Paragraph> <Paragraph position="2"> A study which lists fffactors important to successit in the mainframe computer industry, titled The American Capzrter Industry in ite Intsrnat$omZ Covetit4ve Grzuirommt (#--enclose $1.70) was released last month by the Domestic and International Business Administration of the Department of Commerce.</Paragraph> <Paragraph position="3"> Information which may be needed by managers considering data base implementations is featured in Data Bass L%reotions--The flext Steps (#--enclose $2.401, a 158-page special publication of the National ~ureau of Standards (NBS~, first announced last October; the document incorporates the Proceedings of a 1975 workshop co-sponsored by NBS and the ACM.</Paragraph> <Paragraph position="4"> li-The NatiomZ C<viZ Serv&?s LeagUeCareer SsrzJ%ce Award for 2976 was presented last month to Dr. Ruth M. Davis, Director of the Department of Commerce's National Bureau of Standards CNBS) Institute for Computer Sciences and Technology (ICST)</Paragraph> </Section> <Section position="24" start_page="65" end_page="65" type="metho"> <SectionTitle> AFIPS IN WASHINGTON FCC COMPUTER COMJNICATIONS PLANNING CONFERENCE: 'NO LOGICAL TECHNICAL BOUNDARY' BETWl!EN COMPUTING AND CO~ICATIONS </SectionTitle> <Paragraph position="0"> Summarizing the technical presentations at the FCC P~unning C'onfermee on Comter Colmnurica6~ons Novenber 8-9 in Washington, organized by the AFIPS Washington Office, rrr. Vinton G. Cerf of the Information Processing Techniques Office, Defense Advanced Research Projects Agency (ARPA] , noted: (1) There is no &quot;logical technical boundary'? between computing and communications; (2) Packet switching is having an ttimportant effectw on computer comnmications today, and will have a ftprofound effect&quot; on frequency allocation in the future; (3) High local access costs, international link costs, and user learning costs are (Ilimiting&quot; the growth of computer communications services; (4) The prime opportunities for network costs and tariff improvement lie in the development of new facilities for local access and for intercontinental links; and (5) A regulatory climate is needed &quot;which encourages innovation, ensures reliable interconnection of primary services, and promotes the acquisition of capital for growth.11 Witey says FCC p02ioy deoisions must be bused on under8tunding of technoZogy.</Paragraph> <Paragraph position="1"> Opening the conference, FCC Chairman Richard E. Wiley stated his view that a computer is a comunications device, not a data processing device, when it is used in a traditional communications service. However, Wiley also noted that the use of computer technology may result in an enhanced service, offering more than traditional communications.</Paragraph> <Paragraph position="2"> He predicted that Similar technological forces will develop in several areas of communications service. Wiley said: &quot;1 firmly believe that, in this grbat free enterprise society of ours, government regulations must not be permitted to stand in the way of thchnological development. If the new technology creates administrative difficulties for the government, it is the regulations that must be conformed and not the technology.&quot; According to the FCC chairman, &quot;We will see more instances in the future where, as here, the technology is developing in a manner that blurs traditional interfaces. In cases of this sort, the policymakers mu8t have direct access to technical experts and decisions must be based upon an understanding of the technology.</Paragraph> <Paragraph position="3"> Wiley also said, &quot;It is our belief that AFIPS is a particularly appropriate entity to present this conference fur the Commission, in view of the Federation's sponsorship of the National Computer Conference, and the Joint Comput el' Conferences prior to that. Moreover, it is also our view that AFIPS is a basically impartial organization in relation to our policy concerns,bringing to us the objective views of a group of highly qualified experts .'I Computer communieutions aceaunts for 20 per cent of a22 acpenditmes in computgr fietd. In the conference Is first presentation on &quot;Computer Communications ; An Introduct $on and Overview ,&quot; Lynn Hopewell (chairman, IEEE Computer Society Technical Committee on Computer Communications ; and senior member, executive staff , Coihputer Sciences Corp.] said that computer communications systems had first been used in control I i n g ~lcomplexfl operations of industry and government. Hopewell added that the most common use of computer commun2cations is in data base applicat ions.</Paragraph> <Paragraph position="4"> The speaker suggested that the computer communications industry accounts for 20 per cent of all expenditures in the computer field, Examples of industrial areas affected by computer communications technology were listed as law enforcement, securities brokerage, insurance, reservations systems and banking. Hopewell said that 75 per cent of the top 500 industry corporations are now using camputer communications.</Paragraph> <Paragraph position="5"> Distributing samples of a microcomputer-on-a-chip , he noted technological changes that have produced a 1976 microcomputer CPU costing $20 in contrast to a 1960 IBM CPU costing $30,000.</Paragraph> <Paragraph position="6"> Hopewell added that microprocessors which cost $20 today cost $100 only a year ago.</Paragraph> <Paragraph position="7"> He indicated that such differentials represent only the beginning of extremely low cost computing.</Paragraph> <Paragraph position="8"> Hopewell said that the proposed definitions employed in the new &quot;Computer Inquiryff aid in qlclarifying the permitted uses of computers by common carriers, but in no way allow the proposed abandonment of the 'hybrid1 service concept He noted that the hybrid concept should be retained &quot;because some service offerings have mixes .ob both computing and communications function^,^' not because of unclear definitions.</Paragraph> <Paragraph position="9"> Since &quot;hybrid services will become even more common in the future,&quot; Hopewell continued, elimination of the hybrid concept will &quot;ineyitably suppress innovation because any services that have communication factions inextricably bound into them will be defined as subject to regulation.&quot; The speaker concluded that he saw no &quot;technological or economic reason for the regulation of the resale industry.?l Te~htu,Z~gy $8 'mov;ng td8 tntegmted &deband 8ehes. ' In his presentation on &quot;Research Topics in Computer Conwnmicati~n ,&quot; Vinton G. Cerf summarized t ethnological factors influencing the development of computer comhunicat ions, : the availability of wideband transmission media; the use of computers to control the switching and allocation of transmission bandwidth; the trend towards all-digital telephone networks; and interconnection of computer cornmuhications networks. Cerf said the &quot;technology is leading away from separate, distinct narrowband services . . twoards integrated widenband services. The potential social impact is pervasive; the possibilities for new wealth-creating activity, almost incalculable; and the need for regulatory adjustment, inevitableeft With the juxtaposition of computing and connnunication, he suggested the FCC could: reallocate the radio spectrum using computer-controlled demand access; mandate computer communication network interconnection through the enformcement of interconnection standards; and consider proposals by =plated carriers as well as others &quot;to provide previously impossible combinat ions of information processing and traditional communication services.</Paragraph> <Paragraph position="10"> 'Boundary betueen corrmunicat~ng und computhg wit2 become mare bZurred. ' In his presentation on the MDimensions of t&e Need for Computer Communications ,&quot; Alex Cupan, president, BNR, Inc., suggested that domestic users must seek to reduce the cost of network access for smaller users, especially in the ltsmaller urban centers .'! FCC Interconnection Chief Louis Feldenr , in a question-and-answer period following Currants presentation, stressed the Itneed for computer powervf in rural areas.</Paragraph> <Paragraph position="11"> In his formal presentation, Curran continued that the cost of intercontinental transmission must also be reduced. According to the speaker, these measnres would help to insure that Ifthe common carrier networks can cope with the data communications traffic of commercial users.11 Curran stated that some of the office and home services associated with computer communications could require reintegration of voice and data capabilities.</Paragraph> <Paragraph position="12"> He provided several current examples of personal computing including: Vieuduta, the embryonic electronic newspaper in the' United Kingdom; Icasthzg, instantaneous voting and data collectio~~ in Canada; MzCZgram, a precursor of electronic mail in the U. S.; and CAJ, computer-assisted instruction.</Paragraph> <Paragraph position="13"> Curran also said that other services will require switching machines &quot;capable of recognizing information [interest] specifications as valid addresses For example, swi thcing machines could recognize selective informat ion dissemination interest profiles as valid addresses, he noted. Curran concluded : 'Thus, the boundary between communicating and computing will become more blurred. There are economic arwents for suggesting that the boundary definitions be relaxed to encourage a new cycle of innovation FCC decisGm8 my ZMt the growth of carr%ers, not necessaAZy the g~oth of wegu&%t@d supptiero in data pmoeseSng. In his presentation, titled frLimitations on the Growth of Computer-Communication service^,^^ Prof. Donald A. Dunn, Engineering Economic Systems Department, Stanford University, indicated that user learning costs will limit the rate at which new computer ~ommunication services can be introduced to markets serving non-computer professionals. Dunn suggested that, in future regulatory decisions, integrated service packages (designed to minimize user learning and operation costs) might be used in lieu of individual component services as the unit of,service that is judged cohnnunications or data processing. He added that regulatory limitations 'on the rate of return and regulatory policies on depreciation allowances can restrict technological change as well as limit the rate of introduction of new equipment by carriers.</Paragraph> <Paragraph position="14"> According to the speaker, the earliext ftComputer Inquiry&quot; rule, requiring data processing $entices to be provided by carriers through a separate affiliate, may limit the growth of carriers. But he added that the rule does not affect the growth of the data processing industry since unregulated suppliers can respond to this market. Dunn also noted that the resale and sharing decision, imposing regulation on resale carriers, would not necessarily inhibit the growth of the industry Itsince separate data processing affiliates will not be required of resale carriers that do not provide monopoly services.&quot; He stated that the resale and sharing decision &quot;removed some of the most serious limits to the growth of this industry by opeining the market for network services to essentially any firm willing to operate as a resale carrier.&quot; Dunn concluded: &quot;Pressures are likely to develop soon to regulate providers of information service packages that may offer computer message services to users that obtain network service from resale carriers. Such regulation would inhibit the growth of the industry, and is not needed to protect the interests of users. lf &quot;There is no mtmZ bozbzdary' bsttdem cormnm~cat<ons and computtnq. In their presentation on &quot;The Future of Computer Communications .&quot; Vinton G. Cerf and Alex Curran said that &quot;we cannot offer a solution to the definition of a boundary between communicating and computing--in fact, technical considerations convince us that there is no natural boundary.&quot; Cerf and Curran urged the FCC to support the development of ffcompetitive servicestt; to insure that a ''sufficient setf1 of standards is created to facilitate interconnection of 'prime servicesf1; create a climate in which both computing and carrier interests profit from the installation of &quot;reliablefv facilities; and broaden the base for the acquisition of capital so as to eliminate a possible constraint on growth.</Paragraph> <Paragraph position="15"> Conference preeenutiom oiZZ become prt of fumt record of Caputer InquCry.</Paragraph> <Paragraph position="16"> The conference, open td the public, was attended by over two hundred people, including those who watched the proceedings via closed circuit television in an adjoining room. As previously announced. the presentations will become part of the formal record in the FCC's tlComputer Inquiry.'l A Proceedings containing all the papers is available at $10 per set from AFIPS Headquarters, 210 Sumtit Avenue, hlontvale, New Jersey 07645. Headquarters telephone number is (201) 391-9810.</Paragraph> </Section> <Section position="25" start_page="65" end_page="65" type="metho"> <SectionTitle> AFIPS PAN)% mMsEM COktbfENT ON USE OF SOCIAL SECURITY NUMBER AT REQUEST OF PRIVACY COPMISSION </SectionTitle> <Paragraph position="0"> Members of the AFIPS panel on private sector usage of the Social Security Number (SSN) , organized at the request of the Privacy Protection Study Commission (washington Report, 2/76) , last October responded individua lly to a staff memo copcerning the use of the SSN, submitted to them by Privacy Commission Executive Director Caroler W. Parsons.</Paragraph> <Paragraph position="1"> Daniel D. McCracken, independent consultant, said he favors legislation ?'to prohibit unauthorized matchihg of records&quot; through use of the SSN as a universal identifier . McCracken wrote Parsons, saying that lacking such legislation, &quot;1 would argue for restrictions on the use of the SSN as a partial subsititue, and as a way to keep the more basic issue alive and visible.&quot; McCracken, who is ACM vice-president, is responsible for passage of an ACM resolution opposing the use of the SSN as a universal identifier., Herbert S. Bright, president, Computation Planning Inc., who also filed a response to the memo, said use of the SSN in licensing drivers opens SSN files to insurance companies as well as list compilers and 6ther vendors. Bright added: &quot;Continuing progress in cross-linking practice between insurance companies is increasing the justification for vigorous efforts by the Commission to examine such attacks on privacy and to place the facts before the public and the Congress .&quot; Bright is also a member of ACM.</Paragraph> <Paragraph position="2"> Willard E. Hick, auditor, Massachusetts Mutual Life Insurance Co., Springfield, Massachusetts, responding to the memo, wrote that the Commissionts statement supporting continued use of the personal identifier tlshould emphasize need and not conaentrate on counteracting sugges*ions that have been made .I1 Hick questioned why the Conunission does not &quot;address in more detail the reason universal identifiers are necessary on a positive rather than negative note.&quot; He told AFIPS Washington Report that, to the best of his knowledge, there is no I1crosslinkingw between insurance companies. Hick is a member of the Institute of Internal Auditors.</Paragraph> <Paragraph position="3"> Also responding to the memo, with letters not available at press time, were: Jeffrey V. White, president, The Credit hreau, Inc., Atlanta, Georgia; John J.</Paragraph> <Paragraph position="4"> Stiglemeier, director, Information Center on Education; Roger E. Creel, assistant vice president-Syst ems 6 Processing, Employers Insurance Co. of Wasau, Wausau, Wisconsin; Alden R. Dalzell, director of Data Processing, Ohio University, Athens, Ohio; J. M. Moore, section head, Exxon Corp., Florham Park, New Jersey; and William E. Perry, director of EDP and Research, the Institute of Internal Auditors, to use material in the MIPS Washington Report for their own *publications, except where an article title appears with an If(*),&quot; clearance must first be !obtained from the AFIPS Washington Office.</Paragraph> <Paragraph position="5"> Documents indicated by the s];sbol '&quot;(ff)&quot; are available on request to the AFIPS Washirgton Office.</Paragraph> <Paragraph position="6"> Where price i \noted, make checks payable to &quot;AFIPSal!</Paragraph> </Section> <Section position="26" start_page="65" end_page="65" type="metho"> <SectionTitle> FEDERAL DATA ENCRYPTION STANDARD APPROVED BY COMMERCE DEPARTMENT </SectionTitle> <Paragraph position="0"> A data encryption algorithm, designed to protect digital information, was approved in November ad a Federal .Information Processing Standard (FIPS) by the Department of Commerce. The new standard will be described and promulgated in FIPS PUB 46.</Paragraph> <Paragraph position="1"> The algorithm was first proposed by the Institute for Computer Science and Technology (ICST) of the National Bureau of Standards (NBS') in 1973. The encrypt ion algorithm converts informat Ion into -an encoded [C. e. , encrypted) form which can be transmitted or stored, but which cannot be read without the appropriate key. Thus, it can be used to protect inforfiation communicated between a terminal and a computer, as well as between computers; in addition, it can be used to protect information stored in off-line media.</Paragraph> <Paragraph position="2"> Mandatory stundiud for Federa2 agewies. As a FIPS, Federal agencies wlll be requ~red to use the encryption algorithm, whenever encryption of data is deemed necessary. However, data subject to the ilu:iona2 Security Act of 1967 and the Atomic Enenergy Act of 1952 are exempted from the standard. The encryption algorithm may also be used on a voluntary basis by organizations and individuals outside the Federal government.</Paragraph> <Paragraph position="3"> Primary initial appJication in EFTS. The new standard's primary application is, initially expected to be in Electronic Funds Transfer Systems [EFTS). The algorithm can be employed in EFTS to discourage l1skinuning&quot; of bank card information for the purpose of gaining access to automated teller machines (ATMs). It can also be employed to prevent interception of compute1 messages authorizing transactions at the ATM. Several non-standard encryption algorithms, including one developed by Diebold, Inc,, are already in use by some banks.</Paragraph> <Paragraph position="4"> Strength ofaZgcwith estimated. NBS said technical workshops have been held to measure the strength of the algorithm in terms of time and money that must be spent in finding the key. In November, at the Dartmouth College Symposim on Man and the Computer, ICST Director Dr. Ruth M. Davis stated that, while go code is ~theordtically unbreakable,&quot; 2,500 years of computer tine on a general-purpose computer Ilslgnlficantly faster&quot; than a CDC 7600 would be required to derive a key. With respect to future technology, Dr. Davis estimated that &quot;working from matched sets of clear and clpher data, well over $300 millionH could be spent five years from now to find a key. With technology available in 10 years, it could take $70 million to derive a key, she said.</Paragraph> <Paragraph position="5"> IBM 'e cr~~orithm choeen; compan%es to manufauture devices $morporatl'ng c~ta?t.durd dthuut paying royuztiee.</Paragraph> <Paragraph position="6"> ICST began soliciting encryption algorithms in 1973 and 1974.</Paragraph> <Paragraph position="7"> IBM's proposed algorithm was chosen in 1975.</Paragraph> <Paragraph position="8"> While IBM has patent rights in the standard, the company has .agreed to grant royalty-free licenses to other companies which wish to manufacture devices incorporating the algorithm.</Paragraph> <Paragraph position="9"> The trade press notes that Motorola an& Rockwell ~nternational are planning to manufacture integrated cirduit chips which incorporate the encryption algorithm.</Paragraph> <Paragraph position="10"> Stundcql -too be pubtished 2/I5/77; oonfepeme echeduted.</Paragraph> <Paragraph position="11"> PIPS PUB 46 is.</Paragraph> <Paragraph position="12"> expected-to be published February 15, 1977. and to become effective Augrut 15, 1977. Copies of the standard will be available after February 15 from the National Technical Informatibn Service (NTIS), DOC, 5285 Port Royal Roa& Springfield, Virginia 22161, or through the MIPS Washington Office.</Paragraph> <Paragraph position="13"> NBS will hold I conference on February 15, 1977, at its Gaithersburg, Maryland, facil~tiss to discuss !'IPS PUB 46 in relation to other security measures,</Paragraph> </Section> <Section position="27" start_page="65" end_page="65" type="metho"> <SectionTitle> TERMINAL MANUFACTURERS, INDUSTRY ASSOCIATIONS CRITICIZE FCC APPROVAL OF DATASPEED 4014 </SectionTitle> <Paragraph position="0"> Independent terminal manufacturers and industry associations Bast month criticized the Federal Communications Commission (FCC) for its approval of tariff filrings on AT&Tts Dataspeed 4014 terminal device (Washington Report, 12/76). me FCC's approval, announced in November, reverses a previous proposed decision by the Commission's Common Carrier Buresd. which rejected the Dataspeed 40/4 tariff (Washington aeport, 4/76). Under the new FCC ruszng, ATGT will be permitted to provide the Dataspeed 4014 service.</Paragraph> <Paragraph position="1"> Primmy Issue. pe debate centers on whether the terminal bffering constitutes a data communicat~ons service or a data processing service. If it is a data conununications service, as held in the present donunission ruling, ATGT can provide the service. However, if it is a data processing service, as the Chief of the C~mon Carrier Bureau stated last March, ATET cannot provide the service The FCC had held earlier in its original lvCompute]r Inquiry&quot; that common carriers may pr(wi.de data processing services, but only through unregulated subsidiaries. However, AT&T is even pre6luded. Zrom this option by its agreemept in a 1956 consent decree (with the U.S. Department of Justice) not to participate in unregulated industries.</Paragraph> <Paragraph position="2"> Specific industry criticism. The independent terminal manhfacturers and computer industry groups charged that the development of the Dataspeed 40/4 terminal is subsidized by profits from ATGT's regulated communications service offerings. Thus, they contend that the device represents &quot;unfairff competition by ATGT against unregulatedb industries.</Paragraph> <Paragraph position="3"> Possible further aeth. The FCC has stated that its latest ruling is contingent on the outcome of the Comrnissionts new 'IComputer Iuquiryvt (see Wash6zgtun Report, 8/76). (Comments in this proceeding are now due April 11, 1977; replies are due May 25, 1977.) However, the trade press speculates that it is unlikely that the Dataspeed 40/4 offering, once tariffed, will be &quot;detariffedat' The Computer and~Communications Industry Association [CCIA) has indicated its intention to file suit against the Commission in the U.S. Court of Appeals in the District of Columbia. CCIA seeks to enjoin implementation of the FCC's ruling, pending a full wurt review on the merits.</Paragraph> </Section> <Section position="28" start_page="65" end_page="65" type="metho"> <SectionTitle> CCPA APPROVES PATENT FOR SOFTWARE PROGRAM SECOND TIME </SectionTitle> <Paragraph position="0"> Some software programs are eligible for patents, according to the U.S. Court of? Customs and Patent Appeals (CCPA), which last November reversed %he U.S.</Paragraph> <Paragraph position="1"> Patent Office for the second time on the issue of software patentability.</Paragraph> <Paragraph position="2"> The CCPA decision involves the l'Regulatort' program (developed by Glen F.</Paragraph> <Paragraph position="3"> Chatfield, president of Duquesne Systems, Incl), whfch adjusts the priorities op computer programs.</Paragraph> <Paragraph position="4"> In two previous cases, the Supreme Court (prior to the latest Chatf5eZd decision)has reversed the CCPA, deciding in favor of the Patent Office on limited grounds and not addressing the question of software patentability.</Paragraph> <Paragraph position="5"> In CottschaZk v. Benson, the Court ruled four years ago 'that a program to convert binary-coded decimal was not a patentable llprocessh as defined in ttie U.S. Code, In 'Dann, ComrnissZoner of Patenes ad Truemarks v. Johnston, the Cmrt held last April (WashCngton Report, 5/76) that a computerized recordkeeping system, developed for banks by Thomas R. Johnston, is ltotnrious to one reasonably skilled in [the applicable] art, and is therefore not patentable.</Paragraph> <Paragraph position="6"> CCPA Chief Jud e Harold T. Markey ruled in the ChatfisZd case that &quot;no prior art was cited f by the Patent office] against the appealed claims, and, accordingly, [the] invention must be considered to have been new and unobvious.&quot; The judge also held that' Chatfield's invention meets the U.S. Code requirement for tlpro~e~~'l claims.</Paragraph> </Section> <Section position="29" start_page="65" end_page="65" type="metho"> <SectionTitle> TELEPROCESSING SERVICE PROGW INITIATED BY GSA, TO HELP REDUCE $10 BILLION FEDERAL DATA PROCESSING BUDGET </SectionTitle> <Paragraph position="0"> The General Services Administration (GSA) in November signed its first two teleprocessing services schedule contracts, permitting agencies to order services at government-wide volume discounts!, rather than make separate procurements at greater cost. The contracts, effective December 1, provide for interactive and remote batch proc'essing services &timated at between $50 to $80 million over the next year. Computer Network Cow. and First Data Corp.</Paragraph> <Paragraph position="1"> received the first two contracts.</Paragraph> <Paragraph position="2"> The trade press suggeststhat the teleprocessing services program is the fi~st of several steps the Government is implementing to reduee its $10 billion in yearly expenditures on data processing. For example, the Office of Management and Budget (OMB], Executive Office of the President,is encouraging greater use of outside service bureaus as well as third-party leasing firms. In addition, each Federal agency has been directed by OMB to determine five operations now being performed on its own computers that could be performed on service bureau computers.</Paragraph> <Paragraph position="3"> OMB is expected to continue ~ts moves to cut DP expenses through the transition t~ the Carter Administration.</Paragraph> </Section> <Section position="30" start_page="65" end_page="65" type="metho"> <SectionTitle> @lNSUMER BENEFITS IF EFT TERMINALS ARE NOT CONSIQERED BRANCHES: NCEFT TELLS SENATq </SectionTitle> <Paragraph position="0"> Members of the National Commission on Electronic Fund Transfers [NCEFT) last month told a Senate Subcornittee that the Arner1ea.n consumer &quot;may suffer unnecessary inconveniences and may pay higher prices for payment services&quot; if EFT terminals are governed by the saw, laws and regulations that apply to bank branches.</Paragraph> <Paragraph position="1"> Such Federal laws equate EFT terminal's wiTh branches thus forbidding EFT terminals in non-brandring states.</Paragraph> <Paragraph position="2"> [See Wash5ngton Repop** 12/76.) In testimony presented to the Subcommittee on Financial Iristitutions, chaired by Sen. Thomas d. McIntyre (D-N.H.), NCBFT Chairman William B. Widnall and Executive Director John 0. Benton agreed that &quot;[tlhe greatest current impediment to putting into place the technology which can provide the benefits of EFT to the consumer is the uncertainty which exists about the legal environment within wllich EFT systems will operate in the future.&quot; The chairman and executive director, joined by three NCEFT commissioners, said that a competitive imbalance is created when federally-chartered savings institutions have greater freedom in deploying EFT terminals than national bahks,which are governed by more restrictive legislation. The NCEFT representatives+ added that business enterprises, such as retail chains, &quot;should not be considered , , . regulated institutions or branches thereof,&quot; when they employ electronic terminals to communicate with depository institutions.</Paragraph> <Paragraph position="3"> Last month, the NCEFT convened hearings on competition and sharing. The Commi$sionfs interim report is due February 23.</Paragraph> </Section> class="xml-element"></Paper>