Style Guide for SEAS Argument Templates

SEAS Help System

seas@ai.sri.com
Back to previous page
Forward to next page
Print this page
Exit and Close Window
Introduction: About Getting Started How To
Topics: Concepts Template Style Guide Frequently Asked Questions
Viewers: Manager Publication Information Situation Information
Multi. Arg. Uni. Arg. Derivative Question Uni. Arg. Primitive Question
Multi. Tmp. Uni. Tmp. Derivative Question Uni. Tmp. Primitive Question
Table Viewer Summary Viewer Memos
Collection Manager Collection Viewers Merging Tool
Glossary: Buttons Symbols Terms

Style Guide for SEAS Argument Templates

This guide is meant to serve as an aid to authors of Seas Argument Templates. A would be author should read over this guide before beginning to author templates and might want to review this guide in conjunction with previously authored templates in hopes of improving them. Writing good SEAS Argument Templates is not easy, but it does get easier with practice.

Example SEAS Templates

A miscellaneous collection called the "SEAS Library" is home to two other miscellaneous collections called the "Q&A Library" and "Template Library". The Q&A Library houses example template questions and multiple-choice answers. The Template Library includes templates for a range of business applications and one for assessing the quality of SEAS template. While these might be directly applicable to problems of interest, they primarily are provided to illustrate how SEAS can be applied to a wide range of problems and to serve as examples of good template design.

Authoring Philosophy and Methodology for SEAS Argument Templates

SEAS emphasizes the use of simple and regular inference structures. These structures are captured by Argument Skeletons and associated Fusion Methods. The same Argument Skeleton and Fusion Method are typically used to support multiple Argument Templates over widely differing topics. The idea is that if the Argument Template author fully understands the structure of the interrelated questions that constitute the Argument Skeleton and the light propagation scheme implemented by the Fusion Methods, then the author can write the Argument Template questions and answers to fit. The simpler the Argument Skeletons and Fusion Methods, the easily it is for the author to understand.

The challenge in authoring a SEAS Argument Template is to break the problem down into a hierarchically structured set of questions that matches the selected Argument Skeleton and whose interrelationships among the answers follow the Fusion Methods. Of course, the Argument Skeleton and Fusion Methods can be adjusted as the questions and answers emerge.

The use of Regular Argument Skeletons is encouraged i.e., skeletal trees where all branches are identically structured. Regular structures help to encourage that equal time and emphasis are placed on all aspect of an analysis. If the template author does not strive to reduce clutter and eliminate trivialities, then all of its eventual users, including those authoring and browsing arguments based upon it, will pay an additional price with every use i.e., it is a situation where the template author either pays up front or its users will pay repeatedly thereafter.

Likewise, the use of Uniform or Regular Inference Methods is encouraged. A Uniform Inference Method, where every derivative question's answer is derived through application of the same fusion method, makes for the easiest arguments to understand and lines of reasoning to follow. Regular Inference Methods, ones that employs the same fusion method across all questions at the same depth in the skeletal tree, are the next easiest to understand and follow.

This philosophy is directly opposed to that of most uncertain reasoning systems. In most systems, the author begins by determining what questions might be asked and then interrelates them through a complex set of interconnections, typically annotated with conditional probabilities. As a result, the updating scheme is often complex and difficult to follow for those not versed in probability theory. While this "strong model" approach can be very effective when properly applied, we believe that the SEAS "weak model" approach is easier to understand and use. Its effectiveness is directly related to the author's ability to adapt to these simple and regular fusion structures, writing questions and answers that properly function within these constraints.

There are two distinct ways of approaching the structuring of a SEAS Argument Template: top-down and bottom-up. Using the top-down approach, one starts with the central question and attempts to break it down into a small set of supporting questions, each of approximately the same significance; then one breaks down each of those questions, attempting to break each into the same number of equally significant questions; this continues until questions are produced that can be directly answered or until the number of overall questions has become too numerous to include in a single template. In this latter case, the author might elect to limit the depth of the original template and then capture those elements that fell below that depth limit in their own templates; each of these cascaded templates would share its root question with one of the primitive questions in the original template. The relationship of these cascaded templates to the original template can be captured by adding these to the original template as discovery tools. As such, when an analyst is developing an argument based upon the original template and is confronted with one of its primitive questions, he/she can either elect to directly answer the stated question or invoke one of these discovery tools to further breakdown the question. The advantage of this approach is that the analyst determines which of these discovery tools to employ, thus choosing where and where not to spend their time.

Using the bottom-up approach, one starts by enumerating the detailed conditions that should lead to warning. Once these are enumerated, one begins to cluster these into coherent collections of roughly equal size and significance. One then clusters the clusters, again striving for clusters of equal size and significance, and continues this process until a single cluster remains. Each cluster should give rise to a question in the resulting template, with the nesting of the clusters captured as supporting questions. Another tool developed by SRI, called Angler (see www.ai.sri.com/~angler), can support a group in executing this bottom-up approach to template development. The result can be exported from Angler and then imported into SEAS, resulting in a new SEAS template. The new template can then be refined within SEAS.

In practice, neither the top-down nor bottom-up approach is employed in its pure form. Instead, both are typically employed at different times, one after the other, until a satisfactory result is achieved. Once the overall skeletal structure has been established, then the author's attention should turn to writing the detailed questions and answers for the template. Finally, the author should establish the situation descriptor for the new template that describes the type of situations for which the template is intended to be used.

Elements of Style for SEAS Argument Templates

The following are characteristics of good argument templates. Although it will not always be possible to adhere to these elements of good style, an argument template author should strive to do so.

Overall

Limit the overall size of a template to reduce the minimum time required to record an argument. Break large templates into multiple cascaded templates where the primitive questions in some higher level templates are the root question for some lower level templates. This allows the user to invoke the lower level templates (through discovery tools) when desirable and to ignore them when not, thus allowing the user to choose where to spend their time and how much time to spend in total.
Use a regular skeletal structure. In so doing, you help to guarantee that equal time and emphasis will be placed on each aspect of the analysis during argument creation and argument comprehension.
Since SEAS is geared to performing early warning, the preferred fusion methods are geared to propagating warning conditions up the hierarchy of questions, no matter the answer to other questions. That is, the fusion methods most commonly used resemble a mathematical maximization or minimization, rather than averaging. The argument template author needs to keep the fusion methods well in mind when developing a template's questions and answers. For early warning, each question should be associated with a warning condition. The answers must correlate with a level of warning represented by the red-to-green scale.
Use simple words as much as possible; avoid jargon. This helps to guarantee that questions and answers are properly understood.
Questions and answers should not depend too heavily on the context established by other questions, answers, or the situation description. Each should be as self contained as possible to avoid misunderstandings.
Questions and answers should be phrased in such a way that any two analyst's sharing a common understanding of a situation would chose the same answers in response to the same questions. Otherwise perceived differences of opinion will appear to exist where none do.

Questions

Questions should cover a single subject.
Questions should be in a form where the lights alone summarize the answers, e.g., Yes/No or True/False.
Don't pose categorical questions. Turn them into True/False questions by asking if the categorical assessment implies a threat. E.g., not "What are the intentions of this leader?" but "Do this leader's intentions pose a threat?". The categories can still be directly included within the multiple choice answers to help the user objectively assess the degree of threat imposed by each category.
Don't ask questions about how much the user knows regarding the answer to a question. Instead, ask the question directly and if the user's knowledge is limited, then that user should select a range of answers indicating which remain possible given their limited knowledge. Otherwise, the user's lack of knowledge is not properly captured; the inability of a user to eliminate possible answers is the way that SEAS encodes and understands ignorance.
Questions should pertain directly to the subject of the analysis rather than the availability of evidence or state of knowledge of the analyst. "Is this group supported by organized crime?" is better than "Is there evidence that this group is supported by organized crime?". The presence of evidence supporting does not imply the absence of evidence refuting (i.e., there may still be no clear answer to the question); the absence of evidence supporting does not imply that the answer is false, particularly when no evidence was sought.
Whenever possible, write questions that can be answered with objective or numerical criteria rather than with subjective criteria, which tend to be vague. "Are import levels greater than 30% of export levels?" is better than "Are import levels excessive?". When subjective answers are used, it is entirely possible for two analysts to choose different answers, although they are both in complete agreement about the situation. Thus differences in subjectively defined terms potentially end up masquerading as a material differences where none exist.
If the question asks about a possible future occurrence, state an unambiguous time frame. "In the next 12 months" is better than "this year".
Define your terms; be specific. "Will the stock market decline in value by over 50% ?" is better than "Will the stock market crash?"

Answers

Answers should be as concrete as possible, making it easier for the user to recognize the correct answers. The best answers correspond to things that can be directly and unambiguously observed.
Answers should be mutually exclusive. Otherwise, multiple answers might be selected despite the fact that the analyst knows the answer precisely; the selection of multiple answers is meant to convey a lack of knowledge.
Answers corresponding to a numerical range should cover the allowable range and not overlap. This guarantees that exactly one answer applies when there is sufficient knowledge.
Answers should be parallel to one another, have the same sentence structure, and as many of the same words as possible. This makes the differences in the answers easy to spot.
Answers should include only one dimension. For example, don't vary both the amount of change and the likelihood of change in one set of response categories.
Answers should all deal with the same tense. For example, don't have some of the answers deal with the present and some deal with the future. If the objective is early warning, the questions are probably best posed using future tense.
When a subjective certainty scale must be used, the following has been found to be less ambiguous than others.
Yes, almost certainly
Likely
Even, about as likely as not
Unlikely

No, almost certainly not

Example Applications of SEAS Elements of Style

Following are a number of examples highlighting the application of some of the above elements of style. Each consists of a poor example followed by an improved good example. Be sure to also see the examples in the Q&A Library, a miscellaneous collection of exemplar questions/answers.

POOR
GOOD
What are the tactics of this group? Have the actions of this group posed a threat to the security of the US within the last 12 months?
Is there evidence that the stock market will crash? Will the stock market's value drop by more than 50% within the next 12 months?
Is unemployment expected to become a serious destabilizing factor?
Highly likely
Likely
Possible
Unlikely

Highly unlikely

Is the rate of unemployment expected to substantially exceed the established norm for this country within the next 12 months?
The unemployment rate will exceed 50% of the established norm
The unemployment rate will exceed 35% but remain at or below 50% of the established norm
The unemployment rate will exceed 20% but remain at or below 35% of the established norm
The unemployment rate will exceed 10% but remain at or below 20% of the established norm

The unemployment rate will remain at or below 10% of the established norm

Is the group making an effort to send a message or publicize its cause in a way that is cause for concern? Consider the following:
  • exploits media
  • publications
  • website/internet
  • remains anonymous
very significant evidence
strong evidence
moderate evidence
minimal evidence
no evidence
Is the group disseminating or seeking to disseminate threatening messages? Consider the following:
  • Have they used mass media to distribute threatening messages?
  • Have they published threatening messages in print?
  • Have they used mailings (electronic or conventional) to distribute threatening messages?
  • Have they established a website that contains threatening messages?
The group is seeking to disseminate threatening messages to the masses
The group is seeking to widely disseminate threatening messages outside of their organization
The group is seeking moderate dissemination of threatening messages beyond their organization
The group is seeking to disseminate threatening messages among few outside of their organization

The group is not seeking to disseminate threatening messages

Is the group trying to raise funds to support its operations in a way that is cause for concern? Consider the following:
  • sponsors/benefactors
  • criminal activities
  • business ventures/front organizations
  • investments]
very significant evidence
strong evidence
moderate evidence
minimal evidence
no evidence
Is the group getting support from threatening organizations or through illegal/shady activities? Consider the following:
  • Are they engaged in criminal activities?
  • Are their sponsors/benefactors engaged in criminal activities?
  • Are they engaged in shady business ventures or investments?
  • Do they have front organizations?
The group is receiving more than 65% of its support from threatening organizations or through illegal/shady activities
The group is receiving 41% to 65% of its support from threatening organizations or through illegal/shady activities
The group is receiving 15% to 40% of its support from threatening organizations or through illegal/shady activities
The group is receiving less than 15% of its support from threatening organizations or through illegal/shady activities

The group is receiving no support from threatening organizations or through illegal/shady activities


SEAS and High SEAS 7.1 - Patent Pending and Unpublished Copyright © 1998-2007, SRI International