Personalized and prescriptive decision aiding involves three main steps:

1. It starts with careful study of users' preferred ways of representing information and solving problems. (This is not to be confused with status quo procedures for accomplishing the task.) This part of the process may draw on manual or automated methods for elicitating user mental models and processing strategies.
2. The strengths and weaknesses of these preferred structures and strategies are then analyzed critically in the real-world context in which they are to be used. (This is not to be confused with a purely formal evaluation of logical consistency associated with some of the literature on decision making biases.)
3. The decision aid is then designed to support the user's preferred approach (hence, personalized) while at the same time complementing its weaknesses and guarding against potential pitfalls (hence, prescriptive). The decision aid is tailored to users' mental models and decision making strategies, but at the same time promotes the evolution of better mental models and decisions through a process of critical thinking

An adaptive cognitive system based on the Recognition / Metacognition model employs several types of visualization aids. First, it provides a graphical representation of the mental models that decision makers create and maintain to understand the situation and arrive at an appropriate plan. Second, it supports critical reflection on the mental models by providing metacognitive annotations that represent different types of uncertainty within the model. Third, it supports and suggest strategies for handling the uncertainty and mitigating the associated problems.

In real-world decision making, people do not reduce all types of belief and uncertainty to a single measure (unlike classical formal models). R/M aids emulate naturalistic strategies for uncertainty handling. They recognize qualitatively different patterns of uncertainty in a belief network, such as gaps, conflict, and unreliable assumptions, and provide a rich capability for supporting different uncertainty-handling strategies.

The Recognition / Metacognition model summarizes a set of insights into proficient decision making that have been gleaned in the course of empirical research. Each of the components of the Recognition / Metacognition model gives rise to a display requirement for critical thinking support in a decision aid:

1. Evidence-conclusion relationships: The decision aid should help users keep track of elements in a mental model that provide support for an assessment or plan, and distinguish them from elements that conflict with the assessment or plan. It should help users recognize and organize information that serves as reasons for or against the assessment and/or plan.

2. Incompleteness: The aid should help users flesh out a mental model that is centered on the assessment or plan of interest to them. By pointing out gaps in the mental model, it helps them identify potentially important but missing evidence for or against the assessment or plan. Such information might influence their evaluation of the plausibility of the mental model, and thus help resolve significant uncertainty about the assessment or plan.

3. Conflict: The aid should help officers recognize conflicting evidence or goals, i.e., mental models in which some elements point toward one assessment or plan and other elements points toward a different assessment or plan. The aid should support a process of elaborating the mental model in order to explain and resolove the conflict.

4. Unreliable assumptions: The aid should help users identify the assumptions that are necessary in order to make sense of an assessment or plan (e.g., by filling gaps or resolving conflicts). It should help them evaluate the assumptions both individually and as a package, replace unreliable assumptions with more robust ones, and determine the comparative plausibility of packages of assumptions associated with alternative mental models.

5. Quick Test: The aid should help users rapidly distinguish items that require immediate commitment or action -- because they are time critical, high confidence, or low stakes items -- and help users prioritize other items that warrant critical thinking -- because the stakes are high, uncertainty is significant , and the cost of delay is acceptable.

6. Relationship to recognitional processing: Critical thinking support should complement rather than interfere with recognitional processing of more familiar situations. Critical thinking is not an analytical method that only permits judgments about component parts while excluding "intuition" about the overall situation or solution. It starts with intuition and then, if time is available, reflects on it, supplements it, and guides it. Its outputs are not assignments of probabilities, but coherent "stories" or situation pictures along with metacognitive annotations about their strengths and weaknesses. Unlike analytical methods, critical thinking in the R/M framework does not demand that users decompose a problem into component parts, make numerical assessments, and then mathematically put the parts back together again. At any given time, the user has a mental model upon which he or she can act, along with enhanced understanding of where it might go wrong and what can be done about it.

Information overload is an increasingly salient byproduct of the explosive growth of sensor, communications and information processing technologies. CTI has recently applied the concepts of Personalized and Prescriptive aiding to the design of an aid for rapid, efficient information retrieval and organization.

The aid utilizes both semantic and inferential tools to support retrieval that is more precise (fewer false hits), more complete (fewer misses), and more useful. Ultimately, the aid will support collaborative development of improved mental models and problem solutions in a community of users.

The aid helps users construct mental models that reflect their current understanding of their problem. The aid uses technologies related to Latent Semantic Analysis to extract semantic dimensions characterizing the user's mental model. The LSA technology developed by CTI is itself Personalized and Prescriptive, in that it integrates automated statistical analysis of text with human judgment, each complementing the weaknesses and preserving the strengths of the other.

The semantically characterized mental models of one or more users can be integrated into a single belief web (retaining authorship identification and control). Rapid parallel inferencing is applied to the integrated web both to establish joint implications and to identify conflicts. Users can annotate their own mental models or the mental models of others in terms of gaps, doubts, or unreliable assumptions.

Information is then retrieved using:

1. the semantic dimensions underlying the user's mental model,
2. its inferential implications, both alone and in conjunction with the mental models of other users, and
3. any conflicts that have been detected, either internally or with the models of other users, and any significant annotations provided by other users.

Documents are retrieved not as a long, unstructured list of "hits," but in a meaningful context. For example:

1. Retrieval hits pertaining to a particular part of the user's mental model will show up next to that component. They can also be displayed in the appropriate location in a graphically represented semantic space. This semantic space is based on the application of LSA to the mental models of the user and to the domain as a whole.
2. Documents that were retrieved based on inferential linkages among models will appear in the context of an expanded mental model that displays the new linkages.
3. Critical thinking issues, such as inconsistencies or conflicts in the belief web or doubts or comments of other users, will be represented as annotations in the appropriate places in the mental model network. Retrieval hits that address these issues will be displayed next to this issues they address.