Technology and History

Summary

IfWizard Corporation is a Delaware C Corporation developing the world’s most advanced human analysis software, Kotodama. This system draws from the latest developments in motivation research, behavior analysis, psychometrics, and machine learning to create a highly useful portrait of the psychological and behavioral characteristics of a person. These personal models are automatically formed from a unique and diverse set of channels including free-form video interviews, timed behavior recording, open-ended decision trees, and visual belief/value mapping.

Models are collectively analyzed in conjunction with outcomes to train automatic classifiers with which other people may then be matched or measured. For example, reliable employees, compliant medical patients, gifted students, and likely criminal re-offenders may be respectively analyzed to form trained models that then help classify new individuals. Longitudinal features allow for analyzing and alerting to trends in a single person’s behavior that could alert observers of likely oncoming changes in their behavior or emotional well-being.

Beyond the analysis of existing populations and the classification of new individuals, Kotodama was explicitly designed to also be able to experimentally direct or optimize parameters to encourage beneficial outcomes. Kotodama does this by measuring model changes in relation to input parameters and output results, then algorithmically altering experiment variables to improve results.

The benefits of Kotodama include: (1) scientific analysis of existing behavioral human factors and their dynamics, (2) data-driven classification of people and their point-in-time activity, (3) discovery of opportunity for improving results generally and for cohort clusters, and (4) the interactive guidance of human behavior towards desirable outcomes.

IfWizard’s Kotodama technology is protected by multiple granted patents in the United States, China, Japan, Canada, and Korea, with additional patents pending throughout the world.

Details

IfWizard’s Kotodama technology is a unique system for personal modeling and analysis, featuring built-in experiment design, behavior guidance, and classification. Kotodama has an unprecedented diversity of data channels each providing a special lens with which an aspect of human behavior and identity may be measured and related. This synthesis of different human dimensions provides a uniquely powerful and versatile tool for analyzing behavior and psychometrics.

Kotodama’s evolution happened over multiple development stages by IfWizard Corporation, all geared towards a vision for bridging computer science and psychology to create flexible, powerful, and easy-to-use tools for interactive analysis and optimization.

Originally created as a computer tool designed for ordinary users to create abstracted interactive content automatically optimized in its presentation to users, the first code was written in 2012 just prior to the founding of IfWizard Corporation. From this initial work, a patent was granted in China and is currently pending in the United States and elsewhere.

The principle application of this system was online education, given several special properties of the invention. Although a large body of psychological research on creating effective online courses existed, it was being almost entirely ignored by the systems and tools for course creation which often actively encourage “anti-patterns” that actually inhibit learning. IfWizard’s content-focused designed allowed authors to provide evidence-based, flexible presentation regardless of their technical or graphical ability, allowing the continuously improving application of best practices and flexible presentation on a multitude of current and future devices.

The software presented a knowledge-focused paradigm of content authoring whereby the user is tacitly encouraged to focus on the central elements of the experience including text, multimedia, relational organization, navigation rules, and result evaluation. Removed from the author’s attention while creating the content were design particulars that place the emphasis on questionable subjective aesthetic or ludic preferences.

As the authoring process was focused on the articulation and relation of educational knowledge, the remaining task of converting this now-codified information into a multimedia course was handled automatically through a set of heuristic algorithms and professionally designed visual templates. By automating this publishing process, several significant benefits resulted, including design time saving, technological future-readiness, and increased accessibility. But the most significant benefit was the opportunity to automatically apply evidence-based methods, techniques, and patterns in the course publishing heuristics.

When compared with conventionally administered (i.e. in person) courses, online education varies widely in its relative pedagogical effectiveness. For example, a meta-analysis of 318 studies was performed comparing controlled conventional and online course outcomes. While a minority of online courses produced better educational results than their conventional equivalent, about 1/3 of the online courses had inferior results when compared with conventional methods. As these courses were largely designed by careful and knowledgable researchers, it is likely that online courses developed by a more general audience even more frequently have less effective relative outcomes.
The observation of variance and interest in improving electronic teaching methods has resulted in many years of research and hundreds of suggestions for quantifiable improvement. Dr. Richard Mayer of UCSB, one of the leading online education researchers, expresses the situation thusly:

E-Learning courses should incorporate instructional methods that have been shown to be effective based on high-quality research. Over the past twenty years, hundreds of research studies on cognitive learning processes and methods that support them have been published. We know a lot about how learning occurs. Much of this new knowledge remains inaccessible to those who are producing or evaluating online learning because it has been distributed primarily within the research community.

Because most of the techniques for improving online courses are connected with how the course is presented rather than what information is to be taught, IfWizard’s automatic course generation had a multitude of opportunities to incorporate effective, evidence-based practices into the published course. The ongoing generation of courses allow for regular updates and improvements to the generative heuristics as new evidence is discovered and integrated.

But what was most intriguing in our research was the software’s potential to collect and correlate student performance in connection with courses and use this statistical feedback to improve the course generation through machine learning. While the courses created by the original IfWizard system could work within a third-party LMS (Learning Management Systems), the special analytical and machine learning benefits could only be fully realized through the creation of a proprietary LMS, which IfWizard Corporation then created.

The LMS depended upon the use of relational database base management systems (RDBMSs) and during the development of the LMS an innovative technique for data transportation between relational databases was invented. This data transportation technique was extended into a general data migration library and a patent application was written and submitted with the USPTO to cover the new approach, resulting in two granted United States patents, and patents in Japan, Korea, and Canada.

Although the RDBMS Kotodama used was sufficient for LMS performance, in order to fully realize the deep analysis potentials of the IfWizard data abstraction concept, IfWizard developed a unique high performance analytics server system which serves as the internal engine of Kotodama. The analytical server is a high-performance analytical database that can construct, record, analyze, and optimize data-driven experiments.

With extremely efficient online analytics system, the analytical server is a technology suitable for integration not only with e-learning objectives, but also with a wide variety of client technologies and business domains. Extending the original IfWizard software concepts, the analytical server includes a client that may be used to visually create experiments and analyses while achieving performance comparable to that of dedicated, professionally engineered “machine code” executable applications.

The innovative and unique system of the analytical server database, structured for performance and abstraction, is the basis of an additional granted patent. This patent covers a new approach to structuring analysis which not only allows effective optimization routes, but also other benefits such as abstracted interaction at the user interface and developer level.

In complex data sets, many opportunities for optimization are hidden within layers of relationships between multiple variables, often with counterintuitive dynamics. Determining causes and correlations within these data sets is often very difficult, with many possibilities for confounding variables, researcher bias, and poor experiment design to lead to incorrect or inconsistently applicable models. While such complexity will always pose a difficulty for analysis, the IfWizard analytical server’s design makes it easy to apply techniques for automatic optimization including evolutionary or genetic algorithms (GA), support vector machines (SVM), artificial neural networks (ANN), and other machine learning patterns.

IfWizard’s analytical server database has an internal format of Structures, Mappings, Generators, Events, Filters, and Analyses, uses a proprietary “trans-compiled” language named Arya, and is outlined in a sub-joined patent text, but a high-level description of the modeling structure Kotodama creates on top of these elements is:

This unique operational model provides a flexible but relationally consistent arrangement for describing and processing the incoming channel information, which was initially of a decision tree-type structure articulated through the LMS interactivity technology.

Recognizing the new potential of the analytical database for far more complex human information relationships, IfWizard expanded its scope from conventional education to what it termed “educational diversity,” a fundamental shift from conventional education focused on mastery of a predetermined knowledge set to the discovery and sharing of a plurality of divergent knowledge sets.

The new pedagogical model, which could be described as being more classically “philosophical,” contrasted with the conventional educational model in several ways, including the following:

Canonical Education Diverse/Personal Education
Behavioralist, mimetic approach Epistemological, rational approach
Teaching to a test of pre-determined facts Exploration, discovery of facts
Correctness is canonical; single correct answer Correctness is logical consistency within individual perspective/subjectivity
Goal: conformance to an a priori axiomatic set Goal: personal evaluation of experience and belief

 

To achieve this goal, a new software module was created that allowed multiple users to simultaneous construct, critique, and explore their knowledge, potentially guided by a moderator. Key to its design was a proprietary system for modeling knowledge and, crucially, belief.

It is this personal element of belief that allows for assertions about the same experience, presumed to be shared by all users, to be evaluated differently by different individuals. This is of especial significance because conventional education provides a single canonical representation of knowledge/truth, even if within that representation multiple views or “sides” are presented. IfWizard’s new approach effectively ignored the question of what is ultimately true and, instead, encouraged the personal exploration of the participants’ models and beliefs rather than conditioning towards a single understanding.

To construct a belief model, participants start with one or more raw experiences, considered the objective “sensed reality” responsible for the phenomena of participants. In other words, the experience is presumed to be perfectly identical for all participants, perfectly reproducible, and perfectly objective. Of course, participants actually experience these phenomena differently, due to perceptual differences and subjectivity, but the experience itself is maintained as constant.

Experiences may be related together through focuses. A focus provides a parent-child relationship between one experience and another, allowing for better granularity and clarity in the knowledge dialogue generated by the system. For example, a entire book may be one experience, but a particular chapter could be another, focused experience.
Focuses have different types depending on the kind of parent experience. For example, topic focuses have a question or topic as their delineation relative to the parent experience and may be applied to any type of parent experience. A span focus, on the other hand, is limited to experiences with a range or duration such as text, audio, video, and so on. Other focus types include index, position, and region focuses.

Associated with experiences are sets of assertions. An assertion is simply a textual statement and there are no enforced rules about the content of an assertion, but typically assertions take the form of a claim or proposition regarding the experience.

Assertion sets may collectively reflect a certain philosophy (i.e. several competing views or theories could be represented as several different assertion sets), a participant’s own understanding, or any other designation. An assertion may be objective, such as “there were twelve apples in the bowl,” or subjective, such as “the apples look delicious,” but all assertions are subject to consideration by participant.

The principal form of this consideration is the belief that a participant chooses to associate with the statement. Beliefs have levels of agreement and confidence, from complete agreement to complete disagreement and complete confidence to complete uncertainty respectively. In addition, beliefs allow for a note to be provided by the participant to explain or argue the belief. Beliefs are always associated with a particular participant and each participant’s beliefs form a belief set that may be examined by other participants.

By combining the belief-value mapping with the earlier LMS content creation tools and feeding the results into the analytical database, this new model allowed a co-existing ecosystem of different philosophies and perspectives, each of which could be considered a “school of thought.” Each participant’s views formed their own school reflecting their specific beliefs, principles, and ideas. In addition to the personal schools of individual users, users and organizations could create “group schools” which were not a mirror of a single individual’s knowledge and philosophy but were those of an organization, creed, party, school of thought, etc.

Open discussion areas around this content were topical and took place as commentaries about either “original” or “derivative” content depending on whether the content refers to (i.e. originates from) other content. Commentary was more than just a social experience; it provided the fundamental mechanism for demonstrating understanding or “mastery” of a school. Each user has an individual level of mastery of a school’s various conceptual areas and a user who has mastery could act as a teacher of less masterful users for that specific school’s teaching by “grading” their commentary. By successfully demonstrating mastery, the user earned a higher level of mastery up to the mastery of the user grading them and then, themselves, could teach others.

Grading, which was explicitly explained and related to the canon by the grader, was itself a form a gradable commentary and could act to increase the mastery of the user grading. Depending on the preferences of a group school, some levels of mastery could only be reached through teaching. A range of game-like rewards, ranks, privileges, and other outcomes such as access to higher canon could be achieved depending on the wishes of a school, providing a highly customizable experience for each school.

This revolutionary and patent pending model of education through personalized analysis and classification relative to a dynamic normal was the technological and philosophical basis of the next major development in Kotodama’s evolution. By applying behavioral and classical conditioning psychology to the personalized models and experimental analysis already developed, a bridge between the canonical and diversified educational approaches was created: the unique, interactively derived behavior and belief model of a person could form the basis for mimetic instruction.

The result was a Kotodama module to entrain conditioned behaviors in subjects in an efficient and undemanding way. The desired outcome was to condition the subject to reflexively behave in a specific way given specific prompting. While this is similar in essence to all education, the key is that the performed behavior should be entirely automatic, not requiring any active thought or, in some cases, even awareness of the subject. To more easily achieve this, the conditioning itself is an entertaining, addictive, and “frictionless” activity, with the objective of effectively automating and self-directing training.

A special methodology called SOPTRON was developed to automatically guide behavior according to the model derived from another person, population, or explicit construct. In combination, the seven steps of SOPTRON form a flexible conditioning program very well suited to entrainment and automated learning via computer technology.

The process begins by attracting and directing the subject’s focus. At all times, the human mind practices selective suppression of raw perceptual data. This suppression is crucial to accurate perception of reality, as under-suppressed perception leads to overload and hallucination (viz. 5-HT regulation) while over-suppression leads to disassociation and unawareness. While the attention may either be subliminal or supraliminal, depending on the subjects own awareness of the target of attention, all imprinting must occur with the subject focused on the conditioning. Stimulation of the attention is therefore the initial step in SOPTRON.

With the attention acquired, the prompt is given to the subject. This prompt need not be logically related to the desired outcome, but will nonetheless become tightly associated through repeated imprinting. It is crucial that the prompt closely mirrors the real-world prompt anticipated, though complex and varied real-world prompts may be approximate through programmatic variation of the conditioning prompt that maintains the salient facet.

Once the prompt is presented, the correct behavior is immediately and very briefly shown to the subject. Requiring subjects to blindly experiment towards success is both grueling and requiring of a higher cognitive load, both anathema. Priming may be done through rapid explicit demonstration, hinting, indirect association, or subliminal stimulation. It is imperative that the priming phase be as rapid as possible while still sufficiently perceived. This preconditioning makes subject success easier and encourages suggestibility.

In conjunction with or replacing the prompt (depending on desired association), the subject must very quickly select the correct behavior among a very small set of alternatives. Generally speaking, a single correct and single incorrect choice (with details varying on repeated imprints) is preferred as it reduces cognitive load and strengthens the selection/elimination association.

The subject’s reaction is immediately rewarded or punished depending on its correctness. Reward is provided through physiological pleasure (smiling faces, harmonious sounds, symmetric shapes, bright colors, and other deeply seated signals), progress (points and other accumulated “treasure”), and approval (encouraging messaging especially through respected puppets and, more rarely, raised social status, e.g. ranking within a locality). Punishment is provided by inverted versions of rewards. Reward is multi-imprint aware, in the sense that correct “streaks” or completion of lengthy sessions are specially rewarded.

Behind the scenes, Kotodama processes the response time and correctness of subject’s reaction, comparing it with the subject’s accumulated profile and those of similar subjects to optimally plan the next imprint, whether for review, easy success, repetition, or increased challenge. Optimization objectives are, in order of priority, continued engagement, reinforcement of correct behavior, and further conditioning of new behaviors. In addition to the prompt-test steps, this optimization also affects the future stimulation, priming, and reward/punishment steps.

Based on the optimization result, the software advances to the next imprint and begins SOPTRON anew. Note that no perceivable pause between reward/punishment reception and the next stimulation is presented to the subject and certainly no requirement of voluntary “continue” interaction. It is very important that attention gathering stimulation is performed and the next imprint is almost immediately begun to maintain high engagement and efficiency and decrease subject autonomy.

SOPTRON and its technology form the basis of a pending patent which explicitly connects psychological conditioning, gamification, and personalized goal models. After addressing so many structured human-computer interaction scenarios, the final Kotodama informational channel needed was the incorporation of free form analysis.

To do this, IfWizard leveraged the text-processing functionality of IBM’s Watson, which has thoroughly verified training sets or knowledge bases available as a service to other applications. IfWizard created software to record, transcribe, and analyze free-form audio, video, and text using Watson and IfWizard’s own proprietary linguistic algorithms to evaluate a person’s mood, personality, interests, education, and intellectual ability.

Kotodama creates a unique analysis and comparison of individuals and groups both longitudinally and for a specified point in time. Through these analyses, it forms a comprehensive profile of each person based on about 200 quantifiable factors that are deduced and inferred from the collected data. These factors may be further summated and reduced into more readily actionable values.

The free-form profile forms an evaluation that can be used to help determine the suitability of an individual for a given intended purpose. The potential uses are very broad including employment screening, patient compliance, addiction rehabilitation, customer selection, and so on. Using the example of employee screening, Kotodama can use a custom-trained personalized conditioning model to match the characteristics of existing employees who are considered to be successful and valuable by an employer, while at the same time filtering out those applicants who would likely not make good employees. Similarly, Kotodama can flag candidates and existing employees who demonstrate particularly important characteristics for potential promotion or risk management, and provide interactive training towards improving employee productivity.

Kotodama’s free-form analysis is extremely versatile and powerful in creating a personal portrait from the verbal or written expressions of an individual and complete Kotodama’s human analysis spectrum. The multiple information channels combined with Kotodama’s patented tracking, classification, optimization, conditioning, and dynamic analytics create IfWizard Corporation’s central technology: Kotodama, sophisticated and unique human analysis software.