Skill and Skill Learning - Machine Learning Perspective

Human skill is the ability to apply past knowledge and experience in performing various given tasks. Skill can be gained incrementally through learning and practicing. To acquire, represent, model, and transfer human skill or knowledge has been a core objective for more than two decades in the fields of artificial intelligence, robotics, and intelligent control. The problem is not only important to the theory of machine intelligence, but also essential in practice for developing an intelligent robotic system. The problem of skill learning is challenging because of the lack of a suitable mathematical model to describe human skill. Consider the skill as a mapping: mapping stimuli onto responses. A human associates responses with stimuli, associates actions with scenarios, labels with patterns, effects with causes. Once a human finds a mapping, intuitively he gains a skill. Therefore, if we consider the ‘ stimuli” as input and ‘ responses” as output, the skill can be viewed as a control system. This “control system” has the following characteristics: It is nonlinear, that is, there is no linear relationship between the stimuli and responses. It is time-variant, that is, the skill depends upon the environmental conditions from time to time. 0 It is non-deterministic, that is, the skill is of inherently stochastic property, and thus it can only be measured in the statistical sense. For example, even the most skillful artist can not draw identical lines without the aid of a ruler. It is generalizable, that is, it can be generalized through a learning process. 0 It is decomposable, that is, it can be decomposed into a number of low-level subsystems. 

The challenge of skill learning depends not only upon the above mentioned inherent nature of the skill, but also upon the difficulty of understanding the learning process and transferring human skill to robots. Consider the following: A human learns his skill through an incrementally improving process. It is difficult to exactly and quantitatively describe how the information is processed and the control action is selected during such a process. 0 A human possesses a variety of sensory organs such aa eyes and ears, but a robot has limited sensors. This implies that not all human skills can be transferred to robots. The environment and sensing are subject to noises and uncertainty for a robot. These characteristics make it difficult to describe human skill by general mathematical models or traditional AI methods. 

Skill learning has been studied from different disciplines in science and engineering with different emphasis and names. The idea of learning control presented in article is based on the observation that in machine learning, actions of learning machines being subject to "playback control mode", repeat their motions over and over in cycles. The research on learning control have been reviewed anf for a repeatable task operated over a fixed duration, each time the system input and response are stored, the learning controller computes a new input in a way that guarantees that the performance error will be reduced on the next trial. Under some assumptions, the P-, PI- and PD-type learning laws have been implemented. This approach is based on control theory, but the problem is certainly beyond the domain. According to the characteristics we discussed previously, it is obviously insufficient to approach such a comprehensive problem from only a control theory point of view. The concept of task-level learning can be found in related studies. 

The basic idea is that a given task can be viewed as an input/output system driven by an input vector responding with an output vector. There is a mapping which maps task commands onto task performance. In order to select the appropriate commands to achieve a desired task performance, an inverse task mapping is needed. Task-level learning has been studied in great deal for "trajectory learning" to provide an optimum trajectory through learning and has been successful for some simple cases. For a more complicated case which is realistic in practice, the inverse task mapping is too difficult to obtain. Both learning control and task-level learning emphasize achieving a certain goal by practice, and pay no attention to modeling and learning the skill. From a different angle, a research group at MIT has been working on representing human skill. The pattern recognition method and process dynamics model method were used to represent the control behavior of human experts for a debugging process. In the pattern recognition approach, the form of IF-THEN relationship: IF(signal pattern), THEN(control action) was used to represent human skill. 

Human skill pattern model is a non-parametric model and a large database is needed to characterize the task features. The idea of the process dynamics model method is to correlate the human motion to the task process state to find out how humans change their movements and tool holding compliance in relation to the task process characteristics. The problem with this approach is that human skill can not always be represented by the explicit process dynamics model and if there is no such model, or if the model is incorrect, this method will not be feasible. Considerable research efforts have been directed toward learning control architectures using connectionist or Neural Networks. Neural Network (NN) approaches are interesting because of the learning capacity. Most of the learning methods studied by connectionists are parameter estimation methods. In order to describe the input/output behavior of a dynamic system, NN is trained using input/output data, based on the assumption that the nonlinear static map generated by NN can adequately represent the system behavior for certain applications. Although NNs have been successfully applied to various tasks, their behaviors are difficult to analyze and interpret mathematically. Usually, the performance of the NN approach is highly dependent on the architectures; however, it is hard to modify the architecture to improve the performance.

Another issue is the real-time learning, i.e., dynamically updating the model to achieve the most likely performance. In real-time circumstance, we need to compute the frequencies of occurrence of the new data and add them to the model. The procedure is the same as that used to cope with multiple independent sequences. In this study, we have shown the fundamental theory and method that are needed and the preliminary experiments for real-time learning. However, various issues on real-time learning have not been discussed extensively. For example, what happens if the measured data fed in the learning process represents the poor skill, Le., unskilled performance. Using the current method, the model will be updated to best match the performance of the operator, not to best represent the good skill. This is because we have a criterion to judge the model of the skill, but do not have a criterion to judge the skill itself. In other words, it is possible to become more unskilled in real-time learning. This is a common problem in other recognition fields such as speech recognition. One way to minimize the problem is to ensure the feeding data always represents the good performance. This again needs criterion to describe how good the skill is. We will look at this issue in the future.

In this article I presented a novel method for human skill learning using HMM. HMM is a powerful parametric model and is feasible to characterize two stochastic processes - the measurable action process and immeasurable mental states - which are involved in the skill learning. Based on “the most likely performance’! criterion, we can select the best action sequence out from all previously measured action data by modeling the skill as HMM. This selection process can be updated in real-time by feeding new action data and updating the HMM, and learning through this selection process.

The method provides a feasible way to abstract human skill as a parametric model which is easily updated by new measurement. It will be found useful in various applications in education space, besides tele-robotics, such as human action recognition in man-machine interface, coordination in anthropomorphic master robot control, feedback learning in the system with uncertainty and time-varying, and pilot skill learning for the unmanned helicopter. By selecting different units for the measured data in different problems, the basic idea is applicable for a variety of skill learning problems.

Learning Trajectories

Wood, Bruner, and Ross (1976) and Bruner (1986) developed the concept of leading children’s learning forward through “scaffolding”. This involves the teacher providing a pedagogical trajectory to support children’s movement into new territories. In articulating Bruner’s notion of guided participation, Rogoff (1991) argued that the teacher’s main role is to “build bridges from children’s current understanding to reach new understanding through processes inherent in communication” (p. 351). Later, Bruner (1996) drew on Vygotsky’s notion of the “zone of proximal development” (Vygotsky, 1978, p. 86) when he further defined scaffolding as a logical structuring of ideas to be understood in an order that leads children to develop further and faster than they would on their own. A variety of images for teachers’ roles in scaffolding learning have been presented. 

In describing quality teaching, Wood (1991, p. 109) used the term “leading by following”, noting that the most effective scaffolding draws on the interests and understandings of the child. Cobb and McClain (1999) described an instructional sequence that follows a conjectured learning trajectory that “culminates with the mathematical ideas that constitute our overall instructional intent” (p. 24). Hiebert et al. (1997) used the term “residue” to describe the knowledge that children gain from teaching that may be used as a basis for further planning of sequences of tasks aimed at the development of further particular residues over time. Scardamalia, Bereiter, McLean, Swallow, and Woodruff (1989) portrayed learning trajectories as social phenomena, with teachers employing scaffolding to create more general pathways of potential development of mathematical concepts and procedures. Lerman (1998) discussed the teachers’ roles in setting up loci of development—social interactions with mutual appropriation by teachers and students. Simon (1995) demonstrated how the continually changing knowledge of the teacher creates change in expectations of how students might learn a specific idea. 

A hypothetical learning trajectory provides the teacher with a rationale for choosing a particular instructional design; thus, I (as a teacher) make my design decisions based on my best guess of how learning might proceed. This can be seen in the thinking and planning that preceded my instructional interventions … as well as the spontaneous decisions that I make in response to students’ thinking. (pp. 135-136) Simon used the word “hypothetical” to suggest that all three parts of the trajectory are likely to be somewhat flexible, with teachers changing the learning goals and adapting aspects of planned activities in response to (a) their perceptions of students’ levels of understanding and (b) their on-going evaluations of students’ performance of classroom tasks. Thus actual learning trajectories cannot be known in advance. Further, Simon noted that such a trajectory is made up of three components: the learning goal that determines the desired direction of teaching and learning, the activities to be undertaken by the teacher and students, and a hypothetical cognitive process, “a prediction of how the students’ thinking and understanding will evolve in the context of the learning activities” (p. 136). 

In discussion of Simon’s paper, Steffe and Ambrosio (1995) described teachers’ working hypotheses of what students could learn as being determined by the teacher as she interprets the schemes and operations available to the student’s actions in solving different tasks in the context of interactive mathematical communication. The anticipation is based on the teacher’s knowledge of other students’ ways of operating, on the teacher’s knowledge of the particular mathematics of that student, and on results of the teacher’s interactions with that student. (p. 154) In this discussion, Steffe and Ambrosio raised an important question—one that is at the heart of this paper: “How does a teacher modify a task that fails to activate certain schemes?” (p. 155). This is a question that we return to below. Throughout these varied discussions about scaffolding of learning via the creation of specific learning trajectories, the general picture is one of a teacher planning to create a context in which the class will follow one learning trajectory.

The concepts of “remedial work” and “extension work”, as well as teachers’ everyday experience of some students being more successful than others, suggest that actual learning trajectories are likely to take a shape. “Ability” grouping, setting or streaming present a further model, with teachers aiming to lead groups or whole classes of students to different learning goals (see Figure 3). However, the literature on the negative effects of such grouping in primary and lower secondary schools is extensive (see for example Boaler, 1997; Gamoran, 1992; Ireson, Hallam, Hack, Clark, & Plewis, 2002; Mousley, 1998; Zevenbergen, 2003).

Negative effects of differentiated learning expectations can include lowering for some students of teachers’ expectations, self- and peer-expectations, self concepts, opportunities for positive modelling and mentoring, and student motivation. In fact, this solution to diversity has the potential to exacerbate disadvantage due to self-fulfilling prophesy effects (Brophy, 1983). In our experience, teachers who use groups in this way are aware of these potential effects but also want to set achievable goals for all students. They realise that most classes have pupils with sufficiently divergent needs that any one task may not be appropriate for all. Clearly there is a need to research forms of pedagogy that may help teachers to adapt classroom tasks to the needs of the range of individual pupils in their classes. Thus it is important to research how teacher may modify tasks that fail to enable some students to meet specific learning goals.

 “Clearly … the trajectories followed by those who learn will be extremely diverse and may not be predictable” (Lave & Wenger, 1991) 

In choosing to focus on learning trajectories, we embrace a metaphor that, for all its appeal, implies that learning unfolds following a predictable, sequenced path. Everyone knows it is not that simple; researchers and educators alike acknowledge the complexity of learning. As Simon (1995) emphasized, learning trajectories are essentially provisional. We can think of them as the provisional creation of teachers who are deliberating about how to support students’ learning and we can think of them as the provisional creation of researchers attempting to understand students’ learning and to represent it in a way that is useful for teachers, curriculum designers, and test makers.

I firmly believe that a critical part of our mission as researchers is to produce something that is of use to the field and serves as a resource for teachers and curriculum designers to optimize student learning. No doubt this includes creating, testing, and refining empirically based representations of students’ learning for teachers to use in professional decision-making and, further, investigating ways to support teachers’ decision-making without stripping teachers of the agency needed to hypothesize learning trajectories for individual children as they teach. This focus would add a layer of complexity to our research on learning and invite us to think seriously about how to support teachers to incorporate knowledge of children’s learning into their purposeful decision-making about instruction.

Further, I suggest we consider, in the end, “Whose responsibility is it to construct learning trajectories?” (Steffe, 2004, p. 130). If we researchers can figure out how to supply teachers with knowledge frameworks and formative assessment tools to facilitate their work, teachers will be able to exercise this responsibility with increasing skill, professionalism, and effectiveness. Because of the growing popularity of learning trajectories in education circles, it is worth thinking hard about the role of learning trajectory representations in teaching, and in particular, whether a learning trajectory can exist meaningfully apart from the relationship between a teacher and a student at a specific time and place. Simon’s (1995) perspective on teaching and learning suggests not. As the field moves forward with research on learning trajectories and strive for coherence in learning across the grades, I would like to remain mindful of both the affordances and constraints this particular type of representation offers for teachers and students alike.

EXPANSIVE LEARNING AND ACTIVITY THEORY

Another social learning model which has been expounded in a rather profound, dialectical, and somewhat philosophical way, is Yrjö Engeström’s expansive learning theory (Engestrom, 1987). Viewing psychology to be “at the limits of cognitivism” (ch. 2, p. 1) Engestrom took upon himself the challenge to construct a “coherent theoretical [instrument] for grasping and bringing about processes where ‘circumstances are changed by men and the educator himself is educated'”(ch 2., p. 8). Although the following summary of his theory is rather brief a more detailed reading can be found in Learning by Expanding (Engestrom, 1987), the publication in which the theory was first introduced, or in one of Engstrom’s more recent articles (such as,  Engestrom, 2000a; 2001; 2009; 2010).

It should be noted that Engestrom’s target was not merely a theory of learning but something much more comprehensive, i.e., “a viable root model of human activity” (Engestrom, 1987, ch. 2, p. 8). To guide him toward this objective, he set for himself some rather stringent initial criteria: (a) “activity must be pictured in its simplest, genetically original structural form, as the smallest unit that still preserves the essential unity and quality behind any complex activity ” (ch. 2, p. 8);  (b) “activity must be analyzable in its dynamics and transformations [and] in its evolution and historical change…no static or eternal models ” (ch. 2, p. 8); (c) “activity must be analyzable as a contextual or ecological phenomenon [concentrating] on systemic relations between the individual and the outside world” (ch. 2, p. 8); and (d) “activity must be analyzable as culturally mediated phenomenon [sic]…no dyadic organism-environment models will suffice [he insisted upon a triadic structure of human activity]” (ch. 2, p. 8).

To find his theoretical starting point, Engestrom identified three previous lines of research that met his initial requirements (Engestrom, 1987, ch. 2, p. 9):

1. Theorizing on signs – consisting of research beginning with the triadic relationship of object, mental interpretant, and sign by C.S. Pierce, one of the founders of semiotics, down through Karl Popper, who posited a conception of three worlds (physical, mental states, and contents of thought)
2. The genesis of intersubjectivity – the continuity studies of infant communication and language development, founded by G. H. Mead
3. The cultural-historical school of psychology – consisting of ideas that began with Vygotsky and reach maturity with Leont’ev

The first line of research, theorizing on signs, he rejected as a model because it “narrows human activity down to individual intellectual understanding [and provides] little cues for grasping how material culture is created in joint activity” (Engestrom, 1987, ch. 2, p. 15). The second—though it includes the social, interactive, symbol-mediated construction of reality—he also rejected, because its construction “is still conceived of as construction-for-the-mind, not as practical material construction” (ch. 2, p. 22). The third, he accepted as a starting point, because it “gives birth to the concept of activity based on material production, mediated by technical and psychological tools as well as other human beings” (ch. 2, p. 32). On this premise he erected what he referred to as the third generation of cultural-historical activity theory, starting with Vygotsky’s “famous triangular model in which the conditioned direct connection between stimulus (S) and response (R) was transcended by ‘a complex mediated act’…commonly expressed as the triad of a subject, object, and mediating artifact” (Engestrom, 2001, p. 134). This common expression of Vygotsky’s model is referred to by Engestrom as the first generationof activity theory (Engestrom, 1999, pp. 1-3; 2001, p. 134).

Engestrom considered the insertion of mediating cultural artifacts into human action to be revolutionary, providing a way to bind the individual to his culture and society to the individual:

The insertion of cultural artifacts into human actions was revolutionary in that the basic unit of analysis now overcame the split between the Cartesian individual and the untouchable societal structure. The individual could no longer be understood without his or her cultural means; and the society could no longer be understood without the agency of individuals who use and produce artifacts. This meant that objects ceased to be just raw material for the formation of logical operations in the subject as they were for Piaget. Objects became the cultural entities and the object-orientedness of action became the key to understanding human psyche….The concept of activity took the paradigm a huge step forward in that it turned the focus on complex interrelations between the individual subject and his or her community. (Engestrom, 2001, p. 134)

For Engestrom there was still one important limitation of Vygotsky’s model; it focused on the individual. Engstrom overcame this by drawing on Leont’ev’s famous example of the primeval collective hunt which “showed how historically evolving division of labor has brought about the crucial differentiation between an individual action and a collective activity” (Engestrom, 1999, “Three Generations of Activity Theory”, para. 3). Beginning with a ” general mode of biological adaptation as the animal form of activity may be depicted” (Engestrom, 1987, ch. 2, p. 33), Engestrom applied Leont’ev’s ideas to complete a “derivation…[by] genetic analysis” (ch. 2, p. 33) and demonstrate evolutional “ruptures” in the three sides of the biological adaptation triangle.Individual survival is ruptured by the emerging use of tools.  Social life is ruptured by collective traditions, rituals, and rules. And collective survival is ruptured by division of labor.[1]

Through further derivations in line with Leont’ev’s differentiation of the individual action and the collective activity, he took “what used to be separate ruptures or emerging mediators” (Engestrom, 1987, ch. 2, p. 35) and converted them to “unified determining factors” (ch. 2, p. 35), thus completing a graphical representation of what he referred to as the second generation of activity theory (1999, pp. 1-3; 2001, p. 134). This model accounted not only for individual actions, but for collective activity of a community.

Note that in the second generation model what used to be biological adaptive activity has been transformed into consumption and placed in subordinate relation to three dominant aspects of human activity: (a) production, (b) distribution, and (c) exchange (Engestrom, 1987, ch. 2, p. 36). Marx (Marx, 1973, p. 89 as cited in Engestrom, 1987) explained the relationship between these three dominant aspects of human activity and the individual aspect of consumption as follows:

Production creates the objects which correspond to the given needs; distribution divides them up according to social laws; exchange further parcels out the already divided shares in accord with individual needs; and finally, in consumption, the product steps outside this social movement and becomes a direct object and servant of individual need, and satisfies it in being consumed. Thus production appears to be the point of departure, consumption as the conclusion, distribution and exchange as the middle (…). (ch. 2, p. 36)

Two examples of how this model might be instantiated in the representation and analysis of a specific activity are given in Engestrom (2000a, p. 962). The first example, represents the subject—in this case a physician—engaged in the activity of reviewing patient records prior to meeting with the patient. The object of this activity is the patient records. The expected outcome is an understanding of the patient’s history and the purpose of the visit. Notice that the interaction between the subject (the physician) and the object (the patient records) is mediated by the physician’s medical knowledge, a tool which he leverages to interpret the records and formulate an understanding of the patient’s general health condition. Continuing the scenario, the second example represents the activity of examining and diagnosing the patient, in which the patient becomes the object and his preliminary assessment the intended outcome.

Building on the second generation triangular model of human activity, Engestrom described “the minimal model for the third generation of activity theory” (Engestrom, 2001, p. 136) as requiring at least two interacting activity systems. An example of this model can be found in Engestrom (2010, p. 6). This example depicts the activity of a home healthcare care worker engaged in completing a list of routine tasks while visiting the client’s home, and this in relation to the client’s activity of “maintaining a meaningful and dignified life at home while struggling with threats such as loneliness, loss of physical mobility and the ability to act independently, and memory problems commonly known as dementia” (p. 6). This model of two activity systems in relation to one another is the minimal model for third generation activity theory.

Engestrom (2001) summarized the following five principles of his revised activity theory as follows:

1. Prime unit of analysis: “A collective, artifact-mediated and object-oriented activity system, seen in its network relations to other activity systems, is taken as the prime unit of analysis” (p. 136).

2. Multi-voicedness: “An activity system is always a community of multiple points of view, traditions and interests” (p. 136).

3. Historicity: “Activity systems take shape and get transformed over lengthy periods of time. Their problems and potentials can only be understood against  their own history” (p. 136).

4. Contradictions: Contradictions play a central role as “sources of change and development…[They] are historically accumulating structural tensions within and between activity systems” (p. 137).

5. Possibility of expansive transformations: “An expansive transformation is accomplished when the object and motive of the activity are reconceptualized to embrace a radically wider horizon of possibilities than in the previous mode of activity” (p. 137).

Expansive learning theory is different from all other theories previously reviewed in three significant ways. First, it is concerned with the learning of new forms of activity as they are created, rather than the mastery of putative stable, well-defined, existing knowledge and skill:

Standard theories of learning are focused on processes where a subject (traditionally an individual, more recently possibly an organization) acquires some identifiable knowledge or skills in such a way that a corresponding, relatively lasting change in the behavior of the subject may be observed. It is a self-evident presupposition that the knowledge or skill to be acquired is itself stable and reasonably well defined. There is a competent ‘teacher’ who knows what is to be learned.

The problem is that much of the most intriguing kinds of learning in work organizations violates this presupposition. People and organizations are all the time learning something that is not stable, not even defined or understood ahead of time. In important transformations of our personal lives and organizational practices, we must learn new forms of activity which are not yet there. They are literally learned as they are being created. There is no competent teacher. Standard learning theories have little to offer if one wants to understand these processes. (Engestrom, 2001, pp. 137-138)

Engestrom voiced a rather strong view against a notion of learning “limited to processes of acquisition of skills, knowledge and behaviors, already mastered and codified by educational institutions” (Engestrom, 2000b, p. 526), arguing that such a perspective makes learning irrelevant to the discovery and implementation of novel solutions:

If our notion of learning is limited to processes of acquisition of skills, knowledge, and behaviors already mastered and codified by educational institutions and other accepted representatives of cultural heritage, then finding and implementing future-oriented novel solutions to pressing societal problems has little to do with learning.

I have proposed that a historically new form of learning, namely expansive learning of cultural patterns of activity that are not yet there, is emerging and needs to be understood (Engestrom, 1987). (p. 526)

He further argued that the traditional view of learning is a perpetuated relic of the enlightenment era, and called for a shift of focus toward emergent learning processes from below as a necessary alternative in order for education to maintain relevance:

Give people facts, open their minds, and eventually they will realize what the world should become….I would call this an enlightenment view of learning. Learning is a fairly simple matter of acquiring, accepting, and putting together deeper, more valid facts about the world. Of course, this tacitly presupposes that there are teachers around who already know the facts and the needed course of development. Inner contradictions, self-movement, and agency from below are all but excluded. It is a paternalistic conception of learning that assumes a fixed, Olympian point of view high above, where the truth is plain to see. (Engestrom, 2000b, p. 530)

If education is to remain relevant, educators need to study carefully these changes and build on their internal contradictions and emergent learning processes from below, rather than continue preaching the right answers from above. (Engestrom, 2000b, pp. 533-534)

Second, expansive learning theory is concerned with collective transformation, rather than individual learning. Although changes in the collective are initiated by individuals within the community, the transformation itself is a change in the collective system:

The object of expansive learning activity is the entire activity system in which the learners are engaged. Expansive learning activity produces culturally new patterns of activity. Expansive learning at work produces new forms of work activity. (Engestrom, 2001, p. 139)

Although change originates with individual participants in the collective, the effective change takes place in collective activity system as a whole:

Human collective activity systems move through relatively long cycles of qualitative transformations. As the inner contradictions of an activity system are aggravated, some individual participants begin to question and deviate from its established norms. In some cases, this escalates into collaborative envisioning and a deliberate collective change effort from below. (Engestrom, 2000b, p. 526)

In fact, in his original presentation of expansion learning theory, Engestrom actually reformulated Vygotsky’s conception of zone of proximal development (Engestrom, 1987, ch. 3, p. 27) in terms of collective activities. Although he indicated it to be provisional at the time, he is still using the same reformulated definition:

Vygotsky’s concept of zone of proximal development is another important root of the theory of expansive learning. Vygotsky (1978, p. 86) defined the zone as “the distance between the actual developmental level as determined by independent problem solving and the level of potential development as determined through problem solving under adult guidance or in collaboration with more capable peers.” In Learning by Expanding, Vygotsky’s individually oriented concept was redefined to deal with learning and development at the level of collective activities:

“It is the distance between the present everyday actions of the individuals and the historically new form of the societal activity that can be collectively generated as a solution to the double bind potentially embedded in the everyday actions.” (Engestrom, 1987, p. 174)

In effect, the zone of proximal development was redefined as the space for expansive transition from actions to activity (Engestrom, 2000). (Engestrom, 2010, p. 4)

Third, expansive learning theory focuses on horizontal development rather than vertical. Although it acknowledges a vertical dimension, it emphasizes a focus on the horizontal dimension:

We habitually tend to depict learning and development as vertical processes, aimed at elevating humans upward, to higher levels of competence. Rather than merely denounce this view as an outdated relic of enlightenment, I suggest that we focus on constructing a complementary perspective, namely that of horizontal or sideways learning and development. Both dimensions are involved in expansion. (Engestrom, 2000b, p. 533)

The impetus for change in expansive learning theory is attributed to inner contradictions from within an activity or between two activities:

Contradictions are not just inevitable features of activity. They are “the principle of its self-movement and (…) the form in which the development is cast” (Ilyenkov 1977, 330). This means that new qualitative stages and forms of activity emerge as solutions to the contradictions of the preceding stage of form. This in turn takes place in the form of ‘invisible breakthroughs’. (Engestrom, 1987, ch. 2, p. 45)

Engestrom developed the concept of the contradiction leveraging Bateson’s description of inner contradictions, which were referred to as the double bind (Engestrom, 1987, ch. 3, p. 4). He, of course, reformulated Bateson’s individual dilemma in terms of a social one:

The type of development we are concerned with here—expansive generation of new activity structures—requires above all an instinctive or conscious mastery of double binds. Double bind may now be reformulated as a social, societally essential dilemma which cannot be resolved through separate individual actions alone—but in which joint co-operative actions can push a historically new form of activity into emergence. (Engestrom, 1987, ch. 3, p. 20).

Engestrom described four levels of contradictions which may appear in the human activity system (Engestrom, 1987, ch. 2, pp. 43-45):

Level 1: Primary inner contradiction (double nature) within each constituent component of the central activity.

Level 2: Secondary contradictions between the constituents of the central activity.

Level 3: Tertiary contradiction between the objective/motive of the dominant form of the central activity and the object/motive of a culturally more advanced form of the central activity.

Level 4: Quaternary contradictions between the central activity and its neighbor activities. (ch. 2, p. 44)

In concert with his redefined zone of proximal development, Engestrom identified the collective generation of solutions to the double bind potentiality (i.e., learning) as occurring in long cycles of qualitative transformations, driving by inner contradictions of the activity system, which causes individual participants to question established norms:

Human collective activity systems move through relatively long cycles of qualitative transformations. As the inner contradictions of an activity system are aggravated, some individual participants begin to question and deviate from its established norms. In some cases, this escalates into collaborative envisioning and a deliberate collective change effort from below. (Engestrom, 2000b, p. 526)

As described in Engestrom (2001, p. 152), the seven steps in the cycle are (a) primary contradiction, (b) secondary contradiction, (c) modeling the new situation, (d) new model, (e) implementing the new model, (f) quaternary contradictions and realignment with neighbors, and (g) consolidating the new practice. Later, Engstrom (2010, p. 8) presented the same seven steps with simpler names that highlighted the major activities at each step. They revised labels are (a) questioning, (b) analysis, (c) modeling the new solution, (d) examining and testing the new model, (e) implementing the new model, (f) reflecting on the process, and (g) consolidating and generalizing the new practice. Repeated iterations of these seven steps form an “expansive cycle or spiral” (p. 7), and facilitate the ascension of the activity patterns from the abstract to the concrete. This ascension is characterized by the following description:

This is a method of grasping the essence of an object by tracing and reproducing theoretically the logic of its development, of its historical formation through the emergence and resolution of its inner contradictions. A new theoretical idea or concept is initially produced in the form of an abstract, simple explanatory relationship, a ‘germ cell’. This initial abstraction is step-by-step enriched and transformed into a concrete system of multiple, constantly developing manifestations. In learning activity, the initial simple idea is transformed into a complex object, into a new form of practice. (Engestrom, 2010, p. 5)

Through process of the cycle, the object and motive of the activity are reconceptualized to allow for greater possibility and flexibility than the previous pattern of activity:

An expansive transformation is accomplished when the object and motive of the activity are reconceptualized to embrace a radically wider horizon of possibilities than in the previous mode of the activity. A full cycle of expansive transformation may be understood as a collective journey through the zone of proximal development of the activity. (Engestrom, 2000b, p. 526; Engestrom, 2001, p. 137)

The steps of the cyclical model, of course, are a heuristic device comprised of an ideal sequence that Engestrom explains is likely never followed exactly:

The process of expansive learning should be understood as construction and resolution of successively evolving contradictions….The cycle of expansive learning is not a universal formula of phases or stages. In fact, one probably never finds a concrete collective learning process which would cleanly follow the ideal-typical model. The model is a heuristic conceptual device derived from the logic of ascending from the abstract to the concrete. (Engestrom, 2010, p. 7)

---