THE GENESIS OF MICROVIEWS
A Compact Sketch of Microviews
The aim of this research has been to apply ideas from structuralist descriptions of knowledge to the interpretation of learning-related behavior observations. The objective is theory development. I make no claim to propound a complete theory of mind. I struggle to describe coherently some of the processes that may obtain in the development of mind. In its weakest description, this “theory” is a cluster of ideas, richly exemplified, exhibiting what it might mean if one saw mind as a system of disparate, active structures. I call such structures microviews.
The important idea encoded in the term “micro” is the centrality of fragmentation in processes of knowing: the external world can only be experienced as a collection of disparate microworlds; the challenge of learning has as an essential ingredient coming to understand the unities behind what we experience in so fragmentary a fashion. Microviews are internal, cognitive structures built through interacting with such microworlds and reflecting that fragmentary process of knowing [note 1].
Three outcomes summarize how the microviews formulation proves valuable. The first is that certain kinds of insight can be understood as the experienced correlate of modifications of the organization of microviews. A second is the description of some forms of guidance available to microviews from which effective procedures may develop and be fixed as well known results or procedures. A third outcome is some examples of how particular experiences interact with specific cognitive structures at the moment of insight. In these examples, the insights which occur are seen to derive from a specific configuration of the problem encountered and the control structure of the mind.
There are four primary characteristics of microviews. They are disparate, in the sense of being essentially different, one from another. Microviews are active cognitive structures. They have a genetic history. Finally, they have an internal structure evolved from this history of their genesis.
The internal structure may be divided first into a perspective and functions. I use the term perspective in a secondary sense of the term, that is, as the aspect of an object of thought from a particular standpoint, as in “a historical perspective.” A PERSPECTIVE is the recognition mechanism by which a microview analyzes a problem into elements its functions can cope with. It determines the aspect of a problem from the standpoint of a particular microview. As the historical perspective is owned by the historian, so the microview’s perspective is of the microview. A perspective is composed of SLOTS, which may be thought of as variable names to which are assigned values from the problem context by procedures associated with each slot. Microviews differ from each other because their perspectives are comprised of different slots bearing different functional relations to one another.
Functions and Primitives
The FUNCTIONS of a microview relate slots to one another. They may be either well known results, procedures, or primitives. Well known results are things remembered. Procedures relate slots of the perspective to one another through simple transformations.
Primitives are the operations from which procedures develop. ANCHORING WITH VARIATION is an example of a primitive — but one that in turn needs explaining. Anchoring with variation is a basis for progressive achievement of a result [note 1]. In counting a pile of coins, when the value of a second coin is added to some first, the calculation is anchored at the intermediate result and varied by the value of the next coin to be added. The history of prior, intermediate steps is irrelevant and is obliterated in anchoring. The variations possible from an anchor are a direct consequence of the quiddity, the essential character, of the values apposite for specific perspective slots. That is, one can add number or coin values, but one can not rotate them as one can with shapes drawn on a computer graphics display. There are at least three forms of this primitive operation, each form reflecting a specification of the dominant source of guidance in the variation: difference reduction, bricolage, and detour [note 2].
Trying to prove a theorem, mathematicians ofttimes work backward from the result to the assumptions they need in order to prove the theorem. This “backward chaining” still exhibits anchoring with variation. A common, powerful technique for solving complex problems doubles up forward and backward chaining of anchors and variations, then uses difference reduction in searching for bridges between nearest anchors. This is problem solving by NEGOTIATION among all constraints that actually apply in a given context. These complications are introduced here only to indicate how the elaboration of this procedure germ, anchoring with variation, can lead first to procedures and upon reflection, to methods of solving problems.
The Filiation of Microviews
Some knowledge is a pre-requisite to other knowledge. This empirical observation, which is obvious in simple cases, is also true in more complex ways in more complex situations [note 1].
I will treat the observation as a theoretical assumption and relate it to the internal structure of microviews. An ANCESTRAL microview (or ANCESTOR) is one whose perspective functions as a template for the perspective of another, which is defined by this relation as the DESCENDENT of the former. This filiation of microviews creates the control structure of the mind through specific learning processes. Notice that this filiation is epistemic and not temporal. That is, the logical ancestor in a more mature mind may be different in fact from that first genetic ancestor of the original microview [note 2]. It is clearly possible for a microview’s perspective to be inadequate for the problems it confronts and to be later modified so as to create better registration between the cognitive structures and the problem situations of which it attempts to make sense.
More Ideas on Internal Structure
In response to questions at lecture advocating his production systems as a representation of mind (MIT, 1978), Alan Newell noted there was no such thing as a satisfactory learning theory, that every one he had examined contained a “big switch” which, when thrown, caused learning to happen. My focus on how interaction between internal structure and external circumstances can be seen as an attempt to move the “big switch” out of the mind as a closed system and into the arena of interaction between a system and an external context. The word microviews names structures for which this empirical study has indicated a need. It would be possible to create production systems which would function as the microviews I have described. But such models would be inadequate because the structures would exist only as emergent phenomena of interacting rules.
Constraints on the Representation
All workers in the field hope for the simplest adequate representation scheme, but I require one that is designed to permit the straightforward and explicit representation of learning. Because the genetic filiation of cognitive structures is an important idea for understanding learning, a system of microviews should be able work by “analogy” both in deforming problems (assimilating them to internal structures) and in constructing new component microviews. Similarly, they must be locally complete and self contained because the disparateness of structures representing fragmentary knowledge is an important idea. Further, the analyses of Chapters 2 and 4 have shown how valuable can be the notion that self-constructing systems should complete an output-proposing cycle in parallel even when behavior is serialized (as in moving in a single game cell in Tic tac toe or speaking a single answer to an addition question); unusual output conditions are the primary cue that a problem solving situation is novel in an interesting way. More, microviews must be such as can permit the growth of organization because dealing with the issue of emerging coherence is central to understanding mind. Finally, ones needs a kind of structure that can begin simply enough to represent Miriam’s first use of the Tic tac toe “three corners” strategy(initially no more than an imitation of an action coupled with a conviction that the action would lead to a specific result). But microviews must also have the potential to develop continually over a long time into something as complex as frame systems.`
Elements For a System (GACs: Goal, Action-plan, Constraint)
The logical structure embodied in the most common kinds of rules used in cognitive modelling are condition action rules, of which the productions of Alan Newell may be taken as the most widely used type. A formulaic expression of such a rule might be: IF condition-x then DO action-y.
Demon procedures mentioned through the text as the essential components of microviews can be contrasted with such IF-(then) DO rules in a comparable formula: WHENEVER condition-x DO action-y. Cognitive models using demon procedures depend upon the simultaneous condition-testing of many such demons for their effects. Parallelism of demons is the most salient aspect of such systems. A final contrast may be made with an element which one might call the ACTION-PLAN, represented by the formula: FOR purpose-x DO action-y. Here is a simple summary table to highlight what is most salient about each of these entities:
|production||IF x DO y||conditionality|
|demon||WHENEVER x DO y||parallelism|
|action-plan||FOR x DO y||active purpose|
The aspects of purpose, conditionality, and action execution sequence are engaged by system of each of these kinds of elements. Purpose is usually implicit in rule-based models. Conditionality needs to be handled by any system of action-plans that can be resource limited. The fact that each of these elements is a 2-part structure does not mean that system elements must continue to be so limited. Indeed, one should expect that a system constructed out of slightly more complex elements could have significantly more developmental potential.
I propose one consider constructing learning oriented systems based on a three-part element, which I call the GAC (an acronym for Goal Action-plan Constraint). The GAC can be contrasted simply with these other structures by summarizing the which elements are explicit in the element or implicit in the architecture of a system made up of the elements:
|production||implicit||explicit||only at activation|
|GAC||explicit||explicit||through output [note 1]|
GAC’s can be described most simply as demons with explicit purposes.
GACs permit learning separately about goals, action-plans, and constraints
For an aspect of a structure to be changable, either it must be explicitly represented or else there must exist processes within the organization of the system for the modification of that organization itself. I believe developing models built around elements such as GACs will permit a s significant advance in cognitive simulation through separating out components of what must be learned for effective action. More specifically, one can recognize a new goal without knowing how to achieve it through a specific plan. Similarly, one may learn a procedure by some form of imitation before developing a well-articulated understanding of the goals it will achieve. Finally, constraints upon the achievement of goals with given procedures can be a later acquisition, as in Chapter 4, Miriam learned how to achieve a 3 corner fork and only later can to appreciate her plan’s vulnerability to counter-move [note 1].
Learning separately goals, action-plans, and constraints should prove to be a much more tractable challenge for computer modelling than learning such relatively complex things as if-then rules.`
GACs and Microviews
Such a simple structure as the GAC can by itself be the beginning of a microview. For example, one could represent Miriam’s initial strategy for achieving a fork in Tic tac toe by these three elements: a set of three cells numbers which comprise her markers in the fork; a list of three cell numbers which comprise her plan for achieving the fork; an empty list of constraints. As it becomes more complex through variabilities introduced by self-generated elaborations or by interactions with the environment, two of a GACs separate parts come to function as the perspective of the microview: the goal becomes a demon procedure looking to satisfy its conditions for activating the plan; constraints become demon procedures looking to inhibit goal demons. Since the functioning of a microview is an activation cascade, what originally was a purpose driven GAC may cross over in situations of under-constrained activity to function as a stimulus responding structure: this is how BRICOLAGE emerges as a consequence of the internals of a purpose driven structure. With a multitude of such GACs competing, purposes would be quickly selected by and appear to be driven by conditions and possible actions. Response to a stimulus can thus seen to be a highly developed reaction by a complex structure — not the primitive thing itself from which structures have to be built up.
When necessary for an action to be modified because it doesnt work a primitive form of learning through debugging can occur. A local constraint can be thought of blocking one path of the net. Others become possible to try; they are tried because they are still driven by an integral purpose. Under such conditions, the inversion of order of subgoals becomes conceivable without remote, complex, and reflexive interventions [note 1].
The Mind: Interacting Purposeful Demons
If we discuss the activity of microviews in terms of the question “What sort of computer is the mind?” we conclude that the mind is not “a computer,” but a network of microviews. The activity of microviews implies that they are networks, each node of which is an independent, purposeful demon, always ready to perform its function whenever it can.
If we think of these nodes as little men, we can imagine microviews as little societies of some sort, with the diversity of objectives and negotiations such a name implies. It is here that we make contact with the idea of problem solving in the greater world as a negotiation between goals and givens through permissible functions; in such a circumstance, actions, conditions, and purposes are all modifiable through local interactions.
More Complex Structures
There appear to be two paths to greater complexity, different in principle.
Complexity can develop through elaboration of structures based on those sorts of experiences which permit the more refined specification of what circumstances facilitate the activation and completion of information processing functions in the mind.
The second path to greater complexity of structures is the one this study has pursued, exploring how learning can be seen as the discovery of coherence through achieving more effective organization of prior structures.
The following discussion is a scaffolding for thinking about such issues.
a Cluster of Hetararchical Microviews
The main tactic of my research has been to correlate particular experiences with changes in how disparate microviews relate to each other while problems are being solved. The issue of control structure is central: some basic learning processes derive directly from the control structure of the mind [note 1].
The common, usually implicit assumptiond is that the problem-solving mind has an executive control structure. There is some decision-making process (call it “I”) that selects one method or another for trial application to a problem at hand. Such an assumption fits well with the surface coherence and permanence of the individual. One quintessential novelty in the Freudian theory of mind was the vision of an apparently coherent personality emerging out of the contention of three psychic entities: Ego, Id, and Superego. With structures of such a type, the central issue is how control emerges from the interaction of parts.
What sorts of things need we assume to make a computational explanation go? We begin with demonic activity and genetic filiation of microviews. How might these microviews interact to solve a problem? The assumption of demonic activity leads us to assume a fundamentally competitive interaction obtains between microviews. Such a system is described as having a HETERARCHICAL control structure. When confronted with a problem, the entire mind may be alerted, but I don’t insist on that. A CLUSTER of microviews is the set whose disparate knowledges might apply to the problem.
Microviews Solving a Problem
Here is a sketch of how a three-microview cluster might respond to an addition problem with SIMPLE CONTENTION. Let us propose the problem “How much is 75 plus 26?” [note 1]. One way of answering the question is with counting operations, e. g. representing the quantities by hashmarks and reciting the number names. These sorts of mental operations are characteristic of the counting microview discussed in Chapter 2. A second answer might come from applying the standard algorithm of vertical-form addition: that is, “5 and 6 are 11. Put down the 1 and carry. 7 and 2 are 9 and the carry makes 10. The answer is 101.” A third answer might come from money-based calculations: “Four quarters make a dollar, and a penny’s 101.”
If the control structure were simple contention, the “fastest” microview would produce the answer. Which microview wins the race depends on the problem. For 75 plus 26, a MONEY world wins in my mind, but for 36 plus 59 other worlds of mental computation come to the fore. The state of knowledge in the microview obviously determines the outcome; for example, you can not use the standard algorithm if you do not know it. How such disparate, competing microviews can be integrated to form a coherent understanding of number is addressed in Chapter 2 and described in terms of the general processes set out in the following.
Here is a general scheme for describing the splitting apart and joining together of microviews. Call such processes CLEAVAGE processes [note 1].
The two sub-classes of cleavage processes are insulation and amalgamation. The latter implies a mixing together of elements which maintain their identity. Insulation implies the separation of things which should still be thought of as joined at a level below the surface. (A pen-insula becomes an island through insulation.) Cleavage processes generate new microviews.
Genetic Linkage and Analogy
INSULATION is the creation of a new microview through a splitting off of a new perspective through penetration of a prior perspective by a distinction. The ancestral perspective is SYNGNOSTIC with respect to the descendent, i. e. it views as wholes entities which are analyzed into parts in the descendent perspective [note 1]. The formation of the DECADAL microview (detailed in Chapter 2) exemplifies the process. Let this summary suffice here. One may know that 50 plus 50 is 100 and that 3 plus 3 is 6, yet be confounded by the problem of 53 plus 53. To appreciate that one may catenate the two results of 100 and 6 requires a perspective which divides 53 into decadal and units parts and which implies a specification for the re-integration of the part-sums into the compound entity 106.].
The new knowledge that this additional perspective indicates does not render the ancestral and prior knowledge obsolete. Because there are indications that microviews become distinct, we CHOOSE to view this process of distinction as the genesis of new structure rather than as the complication of a prior structure [note 2].
How does a descendent structure relate to its ancestor? THE GENETIC LINK BECOMES THE PATH OF ANALOGY. Let me be more precise.
Supportive and Constructive Analogy
A descendent microview may directly invoke the knowledge of an ancestor, and this in an essential, supportive role. Confronted by a query “How much is 27 plus 46?”, a counting microview would not respond quickly. A newly developed microview, in which decade numbers (30, 70, etc.) were the salient objects, would not contain as a well known result “20 plus 40 equals 60.” But if its perspective breaks down the original problem to a subordinate one, “How much is 2 plus 4?” (which this new microview likewise would not know), that subordinate query could readily be answered by its ancestor, the counting microview [note 1].
SUPPORTIVE ANALOGY names the case where the perspective analysis of one microview reduces a query into components which may be resolved by the response of an ancestral microview. CONSTRUCTIVE ANALOGY names the case where supportive analogy creates results which have such salience for the applications of the new microview that they become well known results of that world.
The general problem addressed by constructive analogy is the question “Where does guidance come from in hypothesis formation?” Genetic filiation is the primary answer I propose. Supportive analogy provides a weakly held, speculative method of solution which is then subject to confirmation or refutation by trial. An example of procedural analogy may help here [note 1]. A year or more after the end of The Intimate Study Miriam, angry with me for some reason, challenged me with a difficult calculation problem, writing 1916 and 9232 on my chalk board.
When I scoffed that the sum was easy, she had her revenge: “It’s not plus,” she said bitingly, “It’s TIMES.” I refused to do the problem. The next day Miriam returned, entirely without the intervention of anyone else, and worked the problem thus:
' ' | 1 | 9 | 1 | 6 | x | 9 | 2 | 3 | 2 | ------------------- 1 | 0 | 8 | 4 | 2 |
That is, she multiplied within columns and accounted for columnar interaction by ‘carrying’ as in addition. This invention is illuminating because it is so wrong. The perspective of this multiplication microview witnesses its descent from addition knowledge. The interaction procedure, carrying, derives from the ancestral procedure knowledge of addition microviews. When supportive analogy creates a procedure (correct or not) which dominates microview functions, it is constructive analogy.`
Learning Processes of Insight and Construction
With the examples of insulation and constructive analogy behind us, we can now discriminate TWO DIFFERENT KINDS OF LEARNING: processes of insight and processes of construction. The former generate new microview perspectives; the latter fill out the functional capabilities of microviews. Let me describe groping, another function-filling process, before returning to the processes of insight.
GROPING is a more general term for learning by trial-and-error, more general in that it covers trial-and-success. Given the preceding description of anchoring with variation, groping can be described as the kind of behavior that results from the absence of a procedure effective to achieve an objective. If there is strong goal commitment, difference reduction guides the path of problem solution (or failure). If the objective is inspecific, bricolage will lead to its negotiated adjustment of means and ends. If there is strong goal commitment, and the process meets an insurmountable obstacle, the detour will take guidance from the obstacle which blocks direct achievement of the goal.
Since the variation required by any specific problem depends upon the quiddity or “what-ness” of the objects of thought, the groping process is responsive to genetic filiation through the descent of the perspective. With that qualification, groping can be seen as an intra-world process for problem solution and for the establishment of microview-specific procedures with whatever mnemonic reinforcement repetition provides.`
Amalgamation: The Elevation of Control
The second category of insight processes, AMALGAMATION processes, bring formerly unrelated microviews into relation with one another. The two to be discussed, the elevation of control and the correlation of perspectives, both indicate a knowledge of other microviews, but of different kinds.
The ELEVATION OF CONTROL names a process where the creation of a new microview perspective subordinates other microviews (for some range of problems) through supportive analogy. This subordination is functionally equivalent to direct invocation of those microviews. This process has been exemplified in detail in case studies of Chapters 2 – 5. What is most striking in the elevation of control is that a significant increase in power (in terms of calculation range and completeness of coverage, for example) results without an increase in the abstractness of the descriptions involved. The “knowledge about knowledge” implicated in the elevation of control is purely functional in character. Contrast this aspect with the more declarative character of the knowledge implicated in the following.`
The Correlation of Perspectives
The CORRELATION OF PERSPECTIVES names a process which provides guidance for the refinement of the perspective of an imperfectly functioning microview from slot correspondences between it and a second microview perspective. That is, there exists no supportive analogy in the sense of an invocable function, but there exists an analogy of signification, a semantic analogy specifying relations between parts and the wholes of the two perspectives [note 1] . The knowledge of the guiding world is reflected into the second for refining its perpective. “Reflection” is meant to suggest (though the suggestion be equivocal) that such correspondence knowledge without direct function is a rudimentary form of structures which eventually permit the reflexivity of mind.
It is tempting to squeeze these processes into a uniform and balanced framework for their better comprehension. (Should there be two forms of insultation, for example, to balance the two forms of amalgamation?) To do so would be unwise at this point, for several reasons. There is no reason to believe the processes are uniform or balanced. Insulation involves two microviews, and amalgamation, three. The processes go on simultaneously, and both insulation and amalgamation may be implicated in an insight. Finally, this sketch is intended as scaffolding for analysis and interpretation, not as a cage.
Re-Learning: Cognitive Reconstruction
After cleavages of microviews, groping and supportive analogy result in the construction of new procedures. But RE-construction also occurs in this circumstance. For example, before she understood place value, Miriam invented an idiosyncratic carrying procedure, one giving incorrect results, which was subsequently displaced by a standard procedure [note 1].
Of more concern here is the reconstruction of mind in the specific sense of changes in the organization of microviews. The descriptions of the cleavage processes all concerned knowledge that reflected what it was about, accurately albeit imperfectly. In such a case the organization of microviews is basically progressive even if within a microview erroneous procedures develop from a syngnostic perspective. It is a different question to ask, “How does the structure of mind get corrected after it has gone wrong?”
Misunderstanding is essential
It is very difficult, if not impossible, to understand a problem which is unlike anything you have ever encountered. If it is a law of mind that there can be no genesis of new knowledge without attachment to prior cognitive structures, that attachment may frequently be made in error. It is essential, then, to address those misapprehensions deriving from the inappropriate attachment of one microview to another. The clearest example appears in Miriam’s computer-focussed experiences and is analyzed in detail in the “Two Geometries” section of Chapter 5. There I argued that she made specific mis-interpretations through seeing the problems as like those she encountered in turtle geometry BECAUSE she had no ancestral microview built on experiences with the appropriate epistemological character. Only after I later provided a specific such experience was Miriam able to cope with this class of problems; further, my instruction in this case was most effective precisely when I called her attention to the similarity of the problem domain to that prior experience.
The process of reorganization, involving a later group of experiences and an insight of the relevance to a long-standing misapprehension, I name THE CLOSURE OF A GENETIC GAP.
The Closure of a Genetic Gap
In biological filiations, ancestors are temporally antecedent to descendants. In cognitive organizations, the genetic filiation is epistemologically and fundamentally atemporal. The closure of a genetic gap is achieved by the retro-active insertion of a POST-CEDENT ancestor in the filiation between the microview of the originally misapprehended problems and the prior structures to which it becomes connected.
There is a more general form of closure of genetic a gap. Imagine there is a microview of primary importance to its descendants, as is the COUNT microview of Chapter 2. COUNT and the set of its descendents comprise a CLUSTER of microviews. Now imagine further such a cluster without the COUNT world in it. One can think of Miriam’s computer based experiences during The Intimate Study as having generated such a chaotic cluster of microviews without a coherent core.
If there were a specific experience she had which in turn created a microview which could serve as a post-cedent ancestor to those microviews, such would be an example of what I speculate is a major process of learning, which I named the nucleation of microview clusters in “Converging Interpretations” of Chapter 5.
If one asks “What makes it possible for one microview and not another to serve as the nucleus of a cluster ?” an answer of a general sort must depend more on the character of the domain than upon that of the person (as was the case with the mediating microviews discussed in Chapter 5). In One Child’s Learning (Lawler, 1979), I argued that Miriam’s learning about the relation of computer designs to her microviews of navigation exemplified nucleation process. Recall that at the end of The Intimate Study Miriam, her brother, and I assembled a a square spiral design, familiar from her computer experience, but made in that case with cuisenaire rods.
Let’s assume that through this activity she was able to appreciate how the repetition of the basic operations produced the result because of the special character of the domain [note 1]; the microview embodying knowledge of such a domain I call an ARCHETYPE. What makes an archetype special is the degree of transparency relating the objects which can serve as slots’ potential values, the operations performable on them, and the results of those operations. The results can be appreciated as the necessary outcome of performing those operations on such objects. For example, if one arranges increasing lengths of cuisenaire rods at right angles, it is obvious when a “square maze” appears where it came from. This sense of accessible necessity is the hallmark of the archetype. In other microviews, conclusions are hard won by chains of inference. In the archetype, the conclusions possible (which may be few) are “intuitively obvious.”
A second characteristic of archetypes is that relations in them are capable of ramification. That is, whatever is obvious in that particular microview instantiates an epistemologically more general idea whose exemplary archetype is, in fact, capable of illuminating problems proper to other microviews in the cluster. This is the sense in which an archetypical microview embodies a powerful idea. `
The Nucleation of Microview Clusters
The NUCLEATION OF MICROVIEW CLUSTERS names the retro-active insertion of an archetype in a cluster of microviews. When such an event occurs, the re-constructions of individual microviews within the cluster need not be immediate following the reorganization. If that reconstruction occurs, i. e. if the problems involved are encountered again (either in the world or in thought) and re-solved, the effect of nucleation is profound, for the post-cedent archetype becomes the common ancestor of a cluster of microviews and thus renders them all coherent and comprehensible.
The conclusion to which I drive is that the nucleation of microview clusters describes the process through which order emerges from the variously aggregated experiences of life. This is process through which specific microviews come to serve as the “geographic capitals” in Minsky’s metaphor for the structure of mind. The idiosyncratic system of archetypes develops, atemporally, as the skeleton of the mind. Coherence of mind is an achieved articulation of archetypical microviews. What permits communication, uniting the experience of one individual with another, is the epistemological generality of the ideas instantiated in the various archetypes of diverse minds.`
Confronting the Issue of Motivation
The quintessential commitment of the cleavage theory is characterizing the mind as fundamentally active. This previously proposed formulation (I will refer to it as the QUERY formulation) suffers two apparent flaws. In the first place, it portrays the mind as re-active, not active; that is, the mind merely responds actively to queries put upon it without generating queries itself. Secondly, the activity is located in the slots of microview perspectives, but the competition is portrayed as between microviews; this description either is vague or contains an implicit assumption about how active slots interact, about how their activity is integrated proximately. In neither case can the issue of integrating demon activity be ignored. We will turn first to how questions might arise within the mind.
What follows is unusual and may even strike some readers as bizarre. Therefore I want to be especially clear about what the aim is and why the attempt is necessary. I must now confront the issue of motivation in problem solving. Without resolving this issue, I can not hope to achieve a general theory of the active mind. My uneasiness at this obligatory task may inspire in me some caution and in you, I hope, some extra willingness to suspend your disbelief while the analysis and explication proceed.
Grand theories of motivation reduce man’s needs to essentials (think of the famous four F’s: feeding, fighting, fleeing, f…) and derive behavior by long chains of inference from them. Justification is by appeal to man’s continuity with other orders of life. In depending on what is common to man and animals, we risk attending too little to what is distinctive in the human mind. My approach is more differential than axiomatic. I do not ask “What comes first?” but rather “What comes next, and how does that relate to what just went before?” From the assumption of fundamental activity, and for the sake of coherence, I strive for explanations in terms of demons and anchoring with variation. In these terms we will try to explain the DEVELOPMENT OF OBJECTIVES.
The Development of Objectives
The basic idea is this. When a problem fails to be solved that “should” have been solvable (most likely created by elaboration as described in Chapter 1), a demon is created; call it a PUZZLED demon. This demon proposes objectives for mental activity at later times. Because the perspective under which the solution fails is syngnostic relative to what is required for the problem’s solution (and possibly because the specificity of the original problem is incompletely preserved), the slot values necessary at a later time for the problem to proceed further are underdetermined. This underdeterminacy of slot value assignments permits the appearance of symbolic realization of an objective with respect to the original problem inspiring it.
The collection of puzzled demons within a microview is the AGENDA of that microview; the agenda is not a list but a pandemonium of things-to-be-done. Agenda is a term at the same level of generality as perspective and functions; thus it names a third major microview component. THE FLOWING INTERESTS AND CURIOSITY OF A MIND ARE THE OBSERVABLE COUNTERPART OF INTERACTING AGENDAS OF PUZZLED DEMONS.`
Man is Continually Driven — but not Continuously Driven
One can not escape the problem of drives and the mind’s being driven, but it is very much an open question of how important, quantitatively, drives are. One is not always hungry, frightened, angry, or lustful. How important are drives? A simple mechanical model helps here: think of a freewheeling transmission between the other systems of a person and the mind. When the going is tough, the motor controls the machine’s progress, but when the going is easy, the motion goes on regardless of what the motor is doing. However much life is unpleasant and filled with nasty surprises for a man, there is little reason to believe that life is hard — in the sense of so subjugating body and mind that man is a driven creature. This stance argues that the human mind is fundamentally freewheeling – A formula for this position might be that man is continually but not continuously driven.
Man’s behavior may be seen as entirely determined (if one is so inclined) but NOT by drives, rather so through the incremental elaboration of his past successes or through plunging into symbolic encounters with his past failures under the guidance of his present ignorance. The agendas of microviews are primary determinants of behavior.
But how, precisely, do these agendas act, and interact among themselves and between themselves and perspectives? If the agendas of microviews propose queries to which the affiliated perspective of each world responds, is it not circular in extreme to claim that their organization is well described as involving the competition among microviews? Further, there is a basic sense in which commitment is a precursor of problem solving at the detail level we shall consider. For example, this succession of narrowing questions represents the decision steps and commitments through which Miriam came to confront problem solving on a typical day of The Intimate Study:
1. Do you want to go to Logo today? Or stay home? Or go elsewhere?
2. [At Logo] What do you want to do? (The basic and obvious choices were from these possibilities: use the computer, do puzzles, socialize, get snacks, run around.)
3. [When using the computer] What sort of activity interests you today? Would you like to use a game, e. g. Shoot, or an interface, e. g. Draw+? Would you like to use the Logo interpreter directly?
If one is interested in a particular kind of problem, is not the competition of microviews a distraction, and even more, a primary source of confusion? Here we seem to have an essential paradox. That very competition of active knowledge which can illuminate learning is simultaneously a source of pervasive confusion. In a profound sense, then, this theory of learning must be coextensive with a theory of confusion. I grapple with that general problem through exemplifying two specific kinds of confusion, genetic and nominal.
GENETIC CONFUSION is possible when one microview derives from multiple ancestors. Each ancestor could provide guidance, and how could a microview decide beforehand which to prefer? The descent of the DRAW+ (see Chapter 1) program from SHOOT (see Chapter 2) and EEL exemplifies this condition. EEL is a shape manipulation Logo interface, with which Miriam was familiar. A child moves a shape by keying direction characters (U for up, D for down, etc.). Shape size is altered by keying B for bigger and S for smaller. Miriam was familiar with EEL. These command characters were incompatible with Logo anguage. In DRAW+, a new and Logo-compatible interface with the same system appearance, shapes were located by presetting the turtle’s location before invoking the shape-drawing procedure by name. Miriam understood my explanation of this difference, but even after assigning input values to specify the shape’s size, she asked, “How do I make it bigger? B?” How does such incorrect advice get suppressed? Just raising that question focusses attention on the assumed impermeability of microview boundaries, even with respect to ancestors. We will return to this question after discussing a second kind of confusion.
This second major confusion, less essential perhaps in respect of structural genesis but nonetheless more pervasive, might be described as linguistic, but I choose not so to describe it. I will call it NOMINAL (by which I mean name-based) confusion.
There are many more things in the world and relations among things than there are words. The mind experiences more things more differently than it can remember words of the common tongue. Because they must represent more than a single meaning, words cannot be unique labels for slots in perspectives. This situation is the source of nominal confusion, but it is more, much more. Let me return to a specific example before we plunge on to the theme of nominal confusion and its relation to concept formation.
The Interplay of Confusions is the ladder for climbing to abstract meanings.
Both the knowledges constructed from Miriam’s use of SHOOT (a game) and MPOLY (a design program) were among the earliest of The Intimate Study. Each was a world disparate from the other. In both, ANGLE was the dominant focus of attention. The turtle was aligned with the target by turning through an ANGLE as a precondition to using the SHOOT procedure. For every use of SHOOT, several turns through angles were common. In the SHOOT world, turning RIGHT 90 meant turning to the right a little bit, 90 times. The question to resolve was one of scale, how much was “a little bit.” In the designs of the MPOLY programs, there was a whole lot of turning going on, but it is beyond imagining that Miriam connected her own RIGHT 90, when we played at SHOOT, with the quickly generated, massively complex, and incomprehensible designs of MPOLY. She sought to establish correspondences between input angle values and specific designs judged attractive.
We call both these things “angle,” but Miriam could not possibly have seen them as “angle” with a common meaning. We ask, “Was Miriam confused by this usage?” “A ridiculous question!” any critic might reply. “Every child solves this puzzle of multiple meanings at an early age and wends his tortuous way through the accidents of language history to the concepts important in his world.” Precisely so — but — each new occurrence of a meaning is another confusion to be puzzled out. We may own a concept of ANGLE which permits us to unify these two diverse experiences of turning into a single abstract description, but a person who owns no such concept can not be other than puzzled by the particular problem.
The point is this: before it exists, there can be no influence from an abstract description, even to the essential minimum guidance of proposing what sorts of features of a situation ought to be explored to solve a particular problem or to make sense of experience. I propose, however, that the genetic path of experience and the confusion of multiple meanings of words are fecund and complementary influences through which the language-capable mind raises itself to the plateau of abstract meanings where the language of its culture has committed the vocabulary.
The Substitution of Emphatic Forms for Particular Descriptions
Let me turn a trifle analytic. The nomenclature of the language is not perverse. Usually, if two things you fail to see as related are similarly named, the naming indicates that your perspective relating to both name applications is not so developed as the cultural standard. Of course, occasionally the commonality is an accident. In this latter case, when confusion arises implicating a word of multiple meanings, the accident (that culturally embedded nominal confusion) provides the mental contexts which permit disambiguation. Consider this simple joke, the first pun Miriam claims to have invented[note 1]:
Miriam: What should you do if your toe falls off?
Victim: I don’t know.
Miriam: Call a tow-truck.
Here the disambiguation of meanings in her mind is highlighted by a joke as two uses of /to/ which must not be permitted to interact (except in this new mode of relation, punning). The interaction prohibition of multiple word meanings is an argument why the separation of microviews is necessary. It is also the most pervasive and natural example of a situation calling for that hypothetical process in concept formation advanced by Winston [note 2] , the substitution of emphatic forms for particular descriptions. The initial description is one of an object; the last is one of a concept. The change from one to the other is defined as substituting a proscription or prescription for a description. In Winston’s words:
“…I want to make clear a distinction between a description of a particular scene and a model of a concept. A model is like an ordinary description in that it carries information about the various parts of a configuration, but a model is more in that it exhibits and indicates those relations and properties that must and must not be in evidence in any example of the concept involved.”
Near Misses and Far Misses
The language-capable mind receives significant culturally given guidance through the necessity of applying abstract names to varieties of entities in particular experiences. The nominal confusions derived from abstract nomenclature provide exactly the kind of focus needed for learning concepts based on the modification of descriptions. Further, the embedding of labelled slots in descriptions constructed through specific experience of the application of the label in varied contexts constrains to a manageable number the possible hypotheses of a generalized concept through the requirement that those hypotheses be coherent with the existing microview perspectives.
In contrast with Winston’s focus on learning through a prototype and a series of acceptable variants and “near-misses,” the description of knowledge as disparate microviews of thought focusses on “far-misses”; for example, on ANGLE encountered in SHOOT and ANGLE encountered in MPOLY. Good numbers for one world are rarely good for the other. In effect, the different worlds where the same name would be culturally assigned as a slot label must not relate lest confusion result. But paradoxically, the commitment of the language to a common label argues that there exists some perspective within which the label used in these disparate worlds constructed from experience can
The Tight Binding of Confusion, Learning, and Motivation
The importance of not relating is profound. Winston, focussing on a crisp formulation of learning through near-misses, did not mark the asymmetry of the emphatic and negative emphatic forms. In particular, a MUST-NOT-CONFOUND link between two descriptions distinguishes two entities from each other without the necessity of specifying a relation based on some common abstracted character. This is important in avoiding logical complications. But more important is the illumination the negative emphatic form brings to the question of microview boundaries. It is sufficient for the separation of microviews that all slots which are potentially confusable be blocked by MUST-NOT-CONFOUND links. Such refinement is surely a powerful tool for controlling confusion in exquisitely articulated microviews of knowledge.
With less developed microviews, as in the mind of a child or of an adult beyond his familiar experiences, other means of confusion control must be employed. The most common of these is reasoning through actions performed on specific objects. Undeveloped microviews, without a concrete system accessible for anchoring thought to, will suffer frequent confusion through inappropriate demon interruption. If a microview’s boundary system of MUST-NOT-CONFOUND links is imperfect or not rigid, analogical reasoning will be pervasive. If the boundary system is strong and the microview well articulated, analogical reasoning will be constrained, but it will also be controlled in a second sense; it will appear in the guise of reasoned problem deformation guided by the relaxation of MUST-NOT-CONFOUND links.
A third character of the MUST-NOT-CONFOUND link between confusable entities is that its very existence is a goad, perhaps even a puzzled demon. The fact that one observes an empirical suppression of confusion is an implicit criticism of his mind. Microview boundaries purchase proximate coherence by admitting ignorance, one’s incompetence of mind. Thus learning, as it unifies disparate knowledges, is not merely an increase in power; it is a personal victory.