Synthetic Teachers

We all want our children to be taught by the best expert in a field. In the cog future we imagine, cognitive systems will possess expert-level domain, problem-solving, and task knowledge about virtually any domain of discourse. Therefore, it is natural to think about cogs becoming our teachers. In fact, the teach skill is one of the fundamental skills of an expert identified in our Model of Expertise discussed in Chapter 7.

We envision a future in which our children go through their K-12 school years (roughly ages 6-18), college, and even through their entire life with access to a collection of synthetic teachers they have learned from and worked with their entire lives. Teacher cogs will have tremendous advantages over human teachers. They will never tire, never get frustrated, and will be able to discuss how to teach us with the millions of other teacher cogs in existence. As a result, it is possible for new teaching methods to evolve no human could ever have developed.

Intelligent Tutoring Systems

Mechanical teaching aids have been in existence for decades and intelligent tutoring systems (ITS) have been an active area of research since the 1960s. In 1924, Sidney Pressey created a typewriter-like machine capable of presenting questions via a window to a student and scoring their answers (Pressey, 1926; 1927). In 1958, B. F. Skinner, developed a teaching machine with the ability to display a question and allow a response from a student. The sequence of questions could be programmed thereby controlling the learning process of the student (Skinner, 1958). Skinner recognized "machines have the patience and energy for simple exercise and drill" (Skinner, 1965).

With the rise of computers, many computer-aided instruction (CAI) projects were undertaken in the 1960s and 1970s (Hartley and Sleeman, 1973). An early CAI system named SCHOLAR engaged in dialog with

Fig. 8-1: Traditional intelligent tutoring system architecture.

students both questioning the student and allowing the student to ask questions (Carbonell, 1970). SCHOLAR represented domain knowledge as a network of associated facts, in many ways similar to modern hyperlinked documents on the World Wide Web, and guided the student through the material in nonlinear fashion based on the student's responses. Intelligently coaching students and customizing content and delivery for the student is a central feature of modern intelligent tutoring systems (ITS). Anderson and colleagues developed the LISPITS system in 1983 at Carnegie Mellon. LISPITS taught the programming language LISP and incorporated many features seen in today's ITS such as providing feedback to students and engaging in dialog with students (Corbett and Anderson, 1992).

As shown in Fig. 8-1, an ITS is made up of four parts: the user interface model, the student model, the domain model, and the tutor model (Freedman et al., 2000; Nwana, 1990; Al-Emran and Shaalan, 2014; Nkambou et al., 2010). The user interface model facilitates interaction with the student. The student model tracks and depicts the status and progress of the student. The domain model, contains the body of knowledge being taught to the student. By comparing the student model with the domain model, the tutor model determines the next actions to take. The tutor model conducts the training by recognizing the strengths and weaknesses in a student and customizing the instruction. Intelligent tutoring systems provide training and guidance, facilitate practice and exploration, exhibit patience, and answer questions from the student just as any human teacher (Clancey, 1986; Anderson et al., 1985).

Teaching Styles and Pedagogy

A vigorous area of research for decades has been teaching styles and pedagogy. One survey lists 71 different learning styles (Coffield et al., 2004). Kolb's experiential model is based on students' experience, observation, conceptualization, and experimentation (Kolb, 1984). Students tend to favor one of the following learning styles:

• • Accommodator: Concrete Experience + Active Experiment
• (physical therapists)
• • Converger: Abstract Conceptualization + Active Experiment
• (engineers)
• • Diverger: Concrete Experience + Reflective Observation
• (social workers)
• • Assimilator: Abstract Conceptualization + Reflective

Observation (philosophers)

Other researchers focus on how students receive educational instruction, identifying seven modalities (Gardner, 2011; Fleming and Baume, 2006; Conway, 2019):

• • Visual (spatial): pictures, images, and spatial understanding
• • Aural (auditory-musical): sound and music
• • Verbal (linguistic): words, both in speech and writing
• • Physical (kinesthetic): body, hands and sense of touch
• • Logical (mathematical): logic, reasoning and systems
• • Social (interpersonal): groups or with other people
• • Solitary (intrapersonal): work alone and use self-study

We expect teacher cogs to employ any number of modalities to deliver educational material to a student. In fact, it is entirely possible for different teacher cogs to specialize in one or two modalities over the others. Thereby, we see a competitive industry arising where teacher cogs specializing in different styles compete with each other for market share. Of course, in a mass-market environment, this gives rise to the endorsement side of the industry. We can imagine celebrities and other prominent figures endorsing a particular teacher cog. The same will be true for large companies. One can easily imagine Microsoft, Apple, and Google brands of teacher cogs.

The Synthetic Teacher Model

The goal in this chapter is to introduce a model of a synthetic teacher, we call Synthia, based on our Model of Expertise. As shown in Fig. 8-2, we can map the traditional ITS architecture shown in Fig. 8-1 to our Model of Expertise as shown in Figs. 7-4 and 7-5 and include popular ideas on pedagogy and learning styles. Critical to teaching is understanding the condition of the student (modeling the student, M). Synthia maintains an overall model of the student (M_sludent) containing details of the current state of the student. A collection of generic student models allows Synthia

Fig. 8-2: Synthetic teacher cog (Synthia).

to determine the student's current state by matching perceptions (T) of the student with these models:

Synthia maintains a collection of domain-specific models in M_D describing various learning styles and teaching pedagogies such as those discussed earlier:

Synthia uses these models to tailor delivery of course material to the student based on the student's natural inclinations and the method best suited for the material. The body of knowledge to be taught to the student is K_D where the domain is broken down into a number of fields (M_fjdd) and topics (M_topic). Synthia possesses both a collection of generic tasks (L) and a collection of domain-specific tasks related to teaching (L_D) as well as a collection of generic and domain-specific problem-solving skills (P and P_D).

The goal of the ITS is to evolve a student from an initial state to a goal state—mastery of the subject matter—as contained in the M_mastery model. Therefore, a goal in G is established at the outset of training to make M_student equivalent to M_masterv. The teach skill analyzes and evaluates the state of the student, M_student against the goal M_masteir Evaluating the different and understanding how and why the student is lacking allows Synthia to create and apply strategies to determine the next course of action, A, to take in the teaching process. In doing so, Synthia recalls domain-relevant knowledge, including course materials.

The perceive and act skills effect the user interface for the student. As the teaching process proceeds, the tutor's goals, G, and utility values, U, will change to customize learning for the student. For example, if the tutor, while teaching calculus, observes mistakes made involving algebra, it might alter the utility values of the goal "teach solving equations" thereby adjusting the flow of the teaching experience to accommodate the student's needs.

Episodic memory is important for Synthia (K_E). Synthia remembers every interaction with the student. This not only enables Synthia to track student progress but also is used to tailor teaching style and delivery. For example, imagine a student having trouble learning a certain topic. Synthia is able to match and extract from episodic memory knowledge and solutions from instances of overcoming similar difficulties previously. Synthia can then apply this experiential knowledge to the current situation and create modifications to the teaching process to accommodate the student.

Episodic memory also makes Synthia useful for years to come. Because Synthia remembers every interaction with the student, it can recall and use this at any time in the future. For example, imagine a student having learned algebra from Synthia who has now graduated high school, college, and has been working in a professional job for several years. Imagine the student encounters a problem he or she needs to use algebra to solve. The student can ask Synthia for help. Even though it has been years since the initial schooling, Synthia is able to step in and assist with anything within the body of knowledge. For the student, it is like having his or her best teacher from school sitting on their shoulder all the time.

Synthetic vs. Artificial Teachers

The right-hand side of Fig. 8-2 is a cog (an intelligent agent capable of tutoring a student in a particular domain of discourse). Figure 8-2 represents an artificial teacher in which all tutoring functions are performed

Fig. 8-3: Synthetic teacher.

by an artificial system (Level 5 cognitive augmentation). While we certainly expect this to eventually evolve—this has been the goal of ITS researchers for decades and work continues—in the near term, the right-hand side will be a human/cog ensemble involving a human component and one or more artificial components as shown in Fig. 8-3. This is a synthetic teacher achieving Level 3 or Level 4 cognitive augmentation.

As with all human/cog ensembles, the synthetic teacher is the emergent result of biological activity combined with artificial activity (Synthia). As discussed in Chapter 7, medical doctors using cogs perform at a higher level. Likewise, human teachers working with teacher cogs will perform at a higher level. One way in which a synthetic teacher is better is the number of students one can teach. Alone, a human teacher might be able to teach only a couple of dozen students. However, with the aid of cogs, a synthetic teacher might be able to teach thousands of students. In fact, in recent years, Georgia Tech has created a virtual teaching assistant known as Jill Watson to help with routine student interactions involving questions about a course (Goel and Polepeddi, 2016). As a result, the online master's program is able to sustain an enrollment of several thousand students. Jill Watson is a cog. However, Jill Watson is not able to teach students alone, Jill Watson works with human teaching assistants and human teachers in collaborative effort.

Another area in which a synthetic teacher exceeds the abilities of a human teacher working alone is "contact time" with the student. A human teacher is not accessible 24 hours a day, 7 days a week. However, a cog can be available whenever the student needs it to be. Furthermore, a human teacher can conduct only one teacher/student engagement at a time. Cogs can be instantiated as many times as needed. Therefore, synthetic teachers achieve a much larger bandwidth for teacher/student interaction.

A third area synthetic teachers exceeds human teachers is breadth and depth of knowledge. As Synthia becomes more capable, it will embody more and more of the body of knowledge being taught—including new and up-to-date knowledge. Over time, cogs will embody the best knowledge in the domain. Today's educational system relies on millions of human teachers. While all teachers are capable and intelligent, some teachers are more knowledgeable than others. In the cognitive system future we imagine, the teacher cogs will become the sink for superlative domain knowledge. Since the teacher cogs will be able to communicate with each other and share knowledge with each other directly, the entire contingent of teacher cogs will increase in domain knowledge and eventually exceed the knowledge of even the best teachers in the field.

Subject-Oriented Teacher Cogs

It is possible a cog capable of teaching any subject will one day be developed. However, especially in the near-term, we expect cogs to be developed able to teach a specific subject matter. Also possible is the development of a teaching cog for a textbook. Today, it is common for a textbook to come with supplemental multimedia materials. Within a few years, cognitive systems technology will be available to read a textbook and then be proficient at answering questions about the content in the textbook. We certainly foresee cog-based supplemental resources for any textbook being available. In fact, we expect competition in the textbook market to evolve whereby publishers compete with each other over the quality of their synthetic teacher components.

As the capability of teacher cogs increases, students will come to work with multiple personal teacher cogs. One can imagine a student having a cog for each subject as shown in Fig. 8-4.

Teacher cogs will adapt to each individual using them. In fact, tailoring style and pedagogy to students has long been a tenet in education. A cog will learn what works best for the student it is teaching. The student/cog relationship will evolve and be unique to that student/cog pairing. Even though millions of students may be using instances of the same algebra cog, for example, each student will have a unique and personalized

Fig. 8-4: Teacher cogs.

experience with their cog. Furthermore, the student/cog relationship will form and persist over a period of time—years even. In the same way people refer back to class notes and textbooks years after taking a course, our personal teacher cogs will always be available to us to turn to when we need help.

Although Synthia will work with us on a personal level, possibly being housed in our smartphone or tablet computer, it will be able to communicate via the Internet to other teacher cogs. This gives Synthia a tremendous advantage over human teachers—the ability to learn instantly from millions of others. One can imagine the future in which millions of people around the world are learning from their own respective teacher cogs. When Synthia encounters a challenge with its human student and successfully overcomes the challenge, it can immediately make all other teacher cogs aware. If Synthia encounters a challenge and is unsuccessful at overcoming it, Synthia can query other teacher cogs for help. Each teacher cog/student instance is unique so knowledge contained in one teacher cog may be critical to solving the challenge experienced by another teacher cog. In this way, the millions of instances of teacher cogs will work together to evolve and enhance the performance of the entire contingent. Not only will teacher cogs' domain knowledge increase rapidly, but the cogs' ability to teach will evolve rapidly as well.

Another likely corporate use for Synthia is professional development and training. It is easy to imagine the development of Synthia for major products requiring training. For example, one can envision the Synthia for Microsoft Excel, or the Synthia for Adobe Photoshop. Employers will come to use and value people's experience with specific teacher cogs much like they value professional certificates today. For example, a hospital may require employees dealing with electronic medical records to use a Synthia from Epic Systems, Inc. as a condition of their employment and compensation. Also likely is people putting virtual certifications via teacher cogs on their professional resumes.

The presence of millions of teacher cogs in use by average people could have quite an impact on scholarly studies of learning by proving or disproving learning theories or discovering new learning theories. Researchers create theories and models then must test them with humans in actual learning scenarios. Often, such studies are limited and invite much debate. However, with teacher cogs out there in use by millions of people, an enormous amount of data on learning will be generated in a very short amount of time. We foresee future studies of theories and models being based on the wide-market response via teacher cogs.

We also would not be surprised to see totally new teaching methods and techniques evolve both from human researchers using expert-level teacher cogs and also from the interaction of the teacher cogs themselves. As stated in Chapter 1, every time new technology is adopted, it brings about changes to the way humans live, work, play, and behave. Just like handheld electronics, the Internet, and social media have revolutionized shopping, news, and entertainment, the availability of personalized teacher cogs to the mass market could bring about changes and upheavals to the educational system.