International Cybernetics Newsletter
THE CYBERNETICS OF GORDON PASK[1]
Bernard Scott
Cranfield University
Defence Academy
Shrivenham
Wilts SN6 8LA
Scott, B. (2007). “The cybernetics of Gordon Pask”, in Gordon Pask, Philosopher Mechanic: An Introduction to the Cybernetician’s Cybernetician, R. Glanville and K.H. Müller (eds.). edition echoraum, WISDOM, Vienna, pp. 29-52.
Overview
This review of the work of Gordon Pask is in two parts, ordered chronologically. The first part concludes with reference to a seminal paper (Park 1969a) in which Pask argued the need for a cybernetics which could successfully address itself, in a full-blooded sense, to the problem of human cognition and consciousness. The requirements to be met by such a theory were spelled out; their statement represented the distillation of Park’s work as a cybernetician which, at that stage, already spanned nearly two decades.
Pask responded to his own challenge and in the following years (in association with Kallikourdis and Scott) produced what he, himself, recognised as his major work: conversation theory and its several applications in education and decision making. Conversation theory and its chief areas of application are addressed in Part 2 of this article.
Part 1 – Genesis of Theory
1.Introduction
In Pask (1966), the author reports an “emotional impact” on first reading Weiner’s (1948) classic Cybernetics. To the young Cambridge medical student (his degree, when it came, was in physiology), with a diverse background and interests in chemistry, geology and psychology, the book brought fully to consciousness a sense of unity in nature and man’s endeavours that, thus far, had been latent in his own eclecticism. Here was the vision and the final justification for the generalist: the twentieth century version of Renaissance man was born.
In those early years in Cambridge, Pask developed his interests in the work of the information theorists (Hick, Crossman, Broadbent) who, following the pioneering spirit of Bartlett were to lay the foundations for the contemporary dominance of the “information-processing” approach in experimental psychology. At the same time, he began work on the design and development of “learning machines”: the hardware forebears of today’s artificial intelligences (v Pask 1965).
Somehow he found time to continue his interest in the theatre, cross-fertilising his interests with the creation of an adaptive “musi-colour” system that not only anticipated the light shows of popular music concerts by several years but is still, in conception, of greater sophistication. As well as machines that would adapt to a performer or audience and entertain, Pask developed systems that would adapt to a learner and teach.
As early as 1958, he had produced a commercially available system, SAKI, an “adaptive teaching machine” for the instruction of keyboard skills (v Pask 1958). His interest in the fruitful juxtaposition and symbiosis of man and machine, particularly in the area of education, continued throughout his life. It is fair to say that, only now, more than forty five years on, has the evolution of computing power become such that his vision of the automated classroom (v Pask 1961 chapter 6) become a reality.
During these early years, Pask styled himself as a “mechanic philosopher”; he was always at pains to emphasise the role of theoretical insight and speculation in the development and design of new technologies. Indeed, from the perspective adopted in this article, Pask’s primary role was not that of a system builder or inventor but that of thinker and theoretician, who was impelled, on critically thought out methodological grounds, at each stage in the development of radical theory to embody the theory in an artefact.
I think here lies a major problem for many who come to Pask’s work. Their tastes may give them an interest in his systems and in the associated empirical findings when they are employed in education or experimental psychology. Alternatively (and, perhaps, more rarely), their tastes may give them an interest in his work as a theoretician and philosopher of cybernetics. With the former approach, Pask’s published work appears esoteric, pedantic, obscurantist. With the latter approach, the need to come to terms with the working details of new technologies (hardware, software) may lead to a failure to grasp the full power and meaning of Pask’s abstractions.
Since those early years, Pask was prolific as a research worker, writer and international envoy for cybernetics. He published more than a two hundred papers, including six books, as well as handling the report and proposal writing chores of an independent research team (System Research Ltd), holding part-time chairs at Brunel University and the Open University and serving as visiting professor in the Universities of Illinois (at Urban and Chicago Circle) and Atlanta, Georgia.
His course and commitment to cybernetics was set in 1959 on the occasion of the (by now, legendary) conference on “The Mechanisation of Thought Processes” at the National Physical Laboratory in London. That conference brought together workers and thinkers in cybernetics and associated disciplines from all over the world, being both a summation of work to date and the establishment of a platform from which future work was to develop (notable participants were Stafford Beer, Heinz von Foerster, Warren McCulloch, Donald Mackay, Marvin Minsky, W Ross Ashby and Oliver Selfridge).
Pask’s own paper (Pask 1959) established his reputation as a controversial and original thinker. Entitled “Physical Analogues to the Growth of a Concept”, it contained theoretical discussion of the nature of learning and evolution accompanied by the demonstration of purely physical phenomena (growth of crystals at electrodes suspended in an electrolytic solution) that could serve as an adaptive medium for computing equivalences (conditioning) and as a dynamic, regenerative storage medium (memory). Unlike many others of Pask’s visions and insights, this one, of the “chemical” or “organic” computer, has yet to be. One awaits the next generation of computers (after the silicon chip) with interest.
At that conference, Pask established a close relationship with the late Warren McCulloch, which continued until the latter’s death. McCulloch was the chief of Pask’s mentors, as Pask has acknowledged on many occasions. Although their explicit areas of interest and approaches to research differed enormously (Pask, chiefly an experimental psychologist, McCulloch, chiefly a neurophysiologist – v McCulloch 1965) they shared a parallel view of the nature of consciousness and reality and a belief in the power of the abstract concepts of cybernetics to bring some rational order to that view. Telling examples are McCulloch’s own formulation of the concept of heterarchical organisation and the associated “principle of the redundancy of potential command”. He, himself, applied these concepts to the analysis of the workings of the reticular formation in vertebrates. As we shall see, the same concepts have guided Pask’s work on human learning and cognition.
Also at that conference, Pask met Heinz von Foerster and established a similarly fruitful intellectual association. (I referred to this association in an earlier article in ICNL; Scott 1979.)[2]
Pask spent a year with von Foerster at the University of Illinois, Urbana and the two cooperated on research concerning the establishment of cooperation and conflict in small groups (Park and von Foerster 1960, 1961).
On his return to England, System Research Ltd was established as a non-profit research organisation and a programme of research was begun (financed chiefly by grants from the US Army and Air Force) concerning human learning, the development of computer simulations and other models for learning processes, the (continued) study of small group interaction and the development of adaptively controlled systems that, on the one hand, demonstrably facilitated the acquisition of perceptual motor skills and, on the other hand, demonstrably facilitated the efficiency of small group communication, learning, and problem-solving.
In toto, this work spans the decade of the sixties and beyond and has been well summarised by Pask himself (Pask 1975b). His chief collaborators in those years were the psychologist, B N Lewis and a computer scientist, G L Mallen. The author of this article joined the team as research assistant toward the end of this period, in 1967, and, albeit from a lowly position, was able to reconstruct the history to date and to witness and to some extent contribute to the developments that followed.
Given the volume of Pask’s output and the range of his thought, any review has to be selective. In the following sections, I give brief accounts of his work, organised under the headings “Learning as an evolutionary process”, “Teaching and learning by machine” and “Social cybernetics”. As I hope will become clear, the hallmark of cybernetic thought is its unitary nature. In the final section, as the bridge to Part 2, I show, as promised, how Pask himself brought the strands together in his address to the first World Congress of Cybernetics (Pask 1969a).
2.Learning as an Evolutionary Process
As far as I know, Pringle (v Pringle 1951) was one of the first to articulate in some detail the analogy between evolution and speciation and human learning and adaptation. He developed a descriptive model of the brain as a medium for the evolution of more and more complex forms of organisation. His neurophysiological speculations were paralleled by Crossman’s (1959) psychological theory of skill organisation in which stable behaviours are selected from an initially diverse population. From his posthumous writings (Craik 1966), we know that Kenneth Craik, too, had a similar vision of the mechanisms underlying learning and adaptation.
It was Ross Ashby (1956) who clarified the concepts most elegantly: from the perspective of abstract cybernetics, evolution of more complex forms is a necessary consequence of applying a constraint (a rule or principle of selection) to a system with a relatively large number of (initially) “uncoupled” parts.
Pask (for whom Ross Ashby was a constant source of inspiration) understood from Ashby’s abstract theory that the medium of evolution and the nature of the evolving entities are irrelevant. This freed him from a concern with the particulars of brain organisation to concentrate on what he later referred to as the "symbolic evolution” of concepts (Pask 1972a). From the early sixties on, he was concerned to explicate and model the complex processes of mentation (v Pask 1968a) as abstract systems constrained by well-defined properties. In Beer’s terms, these are the “fabric” properties found in any brain-like or, in general, organic system:
- There is a limit on the resources available (be they conceptualised as “storage space”, “free energy”, or “processing time”).
- The basic units or parts out of which a self-organising system is constructed or modelled are themselves just such self-organising systems.
- The system and its parts are active.
Pask’s telling phrase was that “man is a system who needs to learn”; within the limits of “boredom” and “fatigue” humans are always learning, always engaged in a species of symbolic evolution.
As well as the special purpose computers or “learning machines” mentioned above, Pask (in association with Feldman and Mallen) pioneered the use of computer programs to simulate learning and evolution.
Several models (v Pask 1969b) simulate the evolution of populations of self-reproducing automata on a tessellation plane, competing for the resource of “food” or “energy”. Their intent was to demonstrate the ubiquity of the evolution of more complex forms of automata (as hybrids or cooperatives) even in such a highly abstracted and artificial universe.
A different class of models (developed chiefly in association with G L Mallen) simulate the processes whereby skilled behaviours are acquired by a human operator.
In these models, Park first fully articulated his conception of learning as a hierarchical process. At level 1, problem-solving procedures compete for execution; at level 2, higher order problem-solving procedures monitor, construct and select among the lower level procedures, they themselves competing for execution. Further levels maybe invoked to account for creativity, insight and the development of cognitive structures in ontogenesis (Pask, 1966a, models the developmental psychologies of Piaget and Vygotsky in this way).
In his several accounts of these models (see also Pask 1963, 1966b), Pask is always at pains to point out that the hierarchy of control processes is an abstraction invoked by the observer in order to give a reasoned account of cognition: the reality of cognition is, in McCulloch’s phrase, heterarchical. The distinction between levels of control is a necessary consequence of the initial decision to distinguish processes (as symbolic or program-like entities) from the processors (brain-like entities) in which they are executed. Without these distinctions, the observer finds that not only has his system no “top” or “bottom”, it also has no boundaries to distinguish it from its environment, which, by the same token, includes the observer himself. This, I think, is perhaps the heart of Pask’s contribution as an epistemologist: his recognition that certain distinctions have to be made by the observer, if she is to break out from her solipsistic silence and that, having made these initial distinctions, certain consequences necessarily follow, having import in a very general sense for the whole methodology of the cybernetician. With these concerns, he clearly anticipated the classic work of Spencer-Brown (1969), The Laws of Form.
3.Teaching and Learning by Machine
As noted in the introduction, in the main core of his work on human learning, Pask continually sought applications in the real world of education and training.
At the time when teaching machines and programmed learning first became fashionable, following the writings of the behaviourist B F Skinner, Pask articulated the cybernetic view that “teaching is the control of learning” and, freed from the ideological requirement to fit his work to the paradigms of operant conditioning and schedules of reinforcement, set about, in a thoroughly pragmatic manner, the design and development of machines that did, indeed, teach in the required sense. At the same time, his systems served as an embodiment (Pask’s phrase) for the theory of learning described (albeit briefly) in the previous section.
The systems in question have a long history which has been well documented by Pask himself (Pask 1972b, 1975). Their development excited international interest and parallel work was carried out by, among others, Stolhurow, Gaines, Briggs, Hudson and Kelley. As Gaines notes in his survey of the field, “Pask has made available a very deep and comprehensive discussion of automated training and has placed it in the general context of interactions between self-organising systems.” (Gaines 1967, p7).
The essential feature of an adaptive teaching machine is that it monitors a learner’s performance so as to pose problems at the level of difficulty most conducive to effective learning: as the student’s performance improves so the level of difficulty is increased. This may be achieved in several, parallel ways by several feedback loops (for example frequency of stimulus presentation, complexity of stimuli presented). Where the skill has a clearly articulated structure, higher order feedback can monitor and integrate distinct subskills (Pask and Scott 1969 contains a description of one of the most sophisticated systems, developed, in this case, for the skill of teleprinter operation).
Pask and Lewis (1968) gives an account of an experimental system where they argue that such adaptive, responsive environments are what is required to observe a self-organising system, referring to the methodology as a “null-point” or “steady-state” approach. The admission of a flaw in the methodology led to the first systems in which the full logic of a conversational interaction was embodied, albeit in embryonic form.
The basic assumption of the earlier systems was that the student would abide by the agreement to attempt to perform the skill effectively at all times. Pask and Lewis noted a tendency for some subjects to override the adaptive controller by selectively making deliberate errors. In this way, the subject himself could determine which component of a skill was rehearsed. Clearly, such subjects were imposing their own learning strategy on the system. A series of studies (v Pask and Scott 1971), investigated the form of learning strategy adopted in a free-learning situation, where the student was free to focus his attention on different components of a signal translation task as he chose. Two classes of strategy were detected: a holistic, many-at-once approach, and a serialist, one-step-at-once approach, either of which could be effective.
A “conversational” teaching system was devised wherein the student’s choices were accepted only if his performance was at an acceptable level, otherwise the teaching system assumed control, basing its decisions on empirically validated diagnostic measures. Such systems were shown to be more effective than either pure free-learning alone or straightforward adaptive machine teaching.
These studies were the inspiration for a long series of investigations of individual differences. At the same time they helped inspire Pask to elaborate his theory of learning and teaching so as to fully allow for the student to act as his own teacher with his own descriptive framework for the structure of a skill or body of knowledge. As already noted, these developments, in the form of conversation theory, are the main concern of Part 2 of this paper.