INTRODUCTION !ARU TO RUN THE MODEL AFFECT PRINCIPLES REASON PRINCIPLES UTILITY PRINCIPLES TO AMEND !ARU OPTIONS
INTRODUCTION !ARU TO RUN THE MODEL AFFECT PRINCIPLES REASON PRINCIPLES UTILITY PRINCIPLES TO AMEND !ARU OPTIONS


O', dthat Mary.

Australian Philosopher Frank Jackson, in his often cited 1982 essay Epiphenomenal Qualia, presents us with a thought experiment which invites us to examine our intuitions about what it might be like to have learned absolutely everything even remotely possible to learn about the physical structure of the world. What would it be like to have such perfect knowledge? Well it's quite a daunting prospect. I imagine even the most well versed scientists will be humbled by the thought of it. As a philosopher, it seems to me that the more knowledge attained, the larger seems the portion yet to know. In a sly move, Plato's Theatetus, has Socrates deny that he knows anything at all. Jackson creates the fictional Characters Fred and Mary to help pump up our intuitions.

We'll concentrate our interest on Mary. Mary, according to Jackson's tale, is a sort of super scientist who does manage to learn it all - the totality of it! What a phenomenal accomplishment. The quirky thing about Mary, 'though, is that most all she knows is by way of 'book learnin' that is to say via the relations of ideas. Contrast this with learning about the world through experimental means through the manipulation of things, trial and error, and so forth. Jackson highlights this contrast by asking us to imagine limits on Mary's learning environment. Lock Mary in a monochromatic room while she's learning everything. Mary, after all, would require some manipulation of her environment even to read the books or to view a tele-conference. So, it won't do to deprive her of all contact with the world; the world of which she is to gain expert knowledge.

As if Mary's life wasn't already a living hell, Jackson proposes a further torture. Once the poor woman performs this incredible feat, she is released into the colorful world with which we are all so familiar. Happily, (unless the USA uses this technique in their renditions) there is no real life Mary, although there are medical cases of color vision restored in color vision impaired individuals.

Here is the point in the story at which we are asked a crucial question. Does Mary now know something new that she did not know before leaving the room? An affirmative answer here produces a philosophical storm that has lasted fifteen years.

Recent work to explain the problem of Mary's seeming new knowledge relies upon a semantic theory of indexicals. !ARU and the problem of "rock paper scissors" provides a distributive computational approach to understanding the role of indexicals in the knowledge argument.

In addition to levels of explanation (see !ARU 5 ), considerations of simplicity (see !ARU 3), explanatory breath (see !ARU 1 & 2), and affective impacts (see !ARU 6), this example of !ARU considers Utility of action and implements categorization via a semantics of indexicals and ostension. HARRY_LEARNS_ROCK_2 provides a platform for understanding a rational decision process for preferring some single set of behaviors/methods for solving a best action puzzle given an array of neural stimulations. The original ROCK PAPER SCISSORS example of !ARU (see !ARU 20), we noticed, was a neutral platform in that its default behavior when run exhibited no prejudice in favor of any particular representation of methods or theory representations. HARRY_LEARNS_ROCK_2 is no longer a neutral platform because the model uses its system of ostensive indexical identification of objects/processes. It exhibits a preferred behaviors/method. The model tends to settle on acceptance of rocks and rock breaking behaviors.

On the ROCK PAPER SCISSORS example of !ARU, if the inputs to the system are put out of balance through the use of the "be forgetful" button, (without the intervention of a system for identification of indexicals) it will settle on a random preferred behavior/method contingent upon what environmental influences happen by chance to dominate. On the HARRY_LEARNS_ROCK_2 example of !ARU, however, having merely 'learned' to pick out a rock tips the balance of the network in the direction of rocks and rock related behavior.

The Difference between HARRY_LEARNS_ROCK_2 And HARRY_LEARNS_ROCK_3

HARRY_LEARNS_ROCK_3 differs from other HARRY_LEARNS_ROCK models in that HARRY_LEARNS_ROCK_3 moves explaination from a relationship between concepts and evidence in the world to a relationship between concepts and the 'signals'(whatever these may be?) arrising in sensory apparatus. We find that (in this example) there is no difference in way the network settles.

Later we can extend the system to simulate environmental conditions for Mary inside the room vs Mary outside the room. We can use this system to suggest how differences in learning conditions interact with an indexical semantics and influence what Mary has the possibility to know. We can also analyze the system to arrive at a criteria for "individuating" and "counting" Mary's knowledge and determining how much and which of it is new knowledge. We can then use that to suggest whether this adds support to the abilities hypothesis the the qualia hypothesis or some other/new? hypothesis.

One new possibility that is being explored here is that Mary will be shown to have less knowledge after leaving the room.

For those not familiar with the story:

According to Frank Jackson (2004) "Mary is a brilliant scientist who is, for whatever reason, forced to investigate the world from a black and white room via a black-and-white television monitor. She specializes in the neurophysiology of vision and acquires, let us suppose, all the physical information there is to obtain about what goes on when we see ripe tomatoes, or the sky, and use terms like 'red', 'blue', and so on . . . .

"What will happen when Mary is released from her black-and-white room or is given a color television monitor? Will she learn anything or not? It seems just obvious that she will learn something about the world and our visual experience of it. But then it is inescapable that her previous knowledge was incomplete. But she had all the physical information. Ergo there is more to have than that, and physicalism is false" (pp. 42-43).

The Chalmers Formalism

David J. Chalmers, as quoted below, provides one of the most recent attempts to formalize the knowledge argument presented by Jackson above.

(1) 'PT has proper subset Q' is a posteriori.

(2) If 'PT has proper subset Q' is a posteriori, 'PT has proper subset Q' is 1-contingent.

(3) If 'PT has proper subset Q' is 1-contingent, 'PT has proper subset Q' is 2-contingent or

panprotopsychism is true.

(4) If 'PT has proper subset Q' is 2-contingent, Physicalism is false.

(5) Physicalism is false or panprotopsychism is true.

"Let P be the complete micro-physical truth about the world, and let Q be a truth stating that phenomenal redness is instantiated, deploying a pure phenomenal concept of phenomenal redness.

"...straightforwardly ... conjoin[ing] to P a "that's-all" claim T, saying that our world is a minimal world satisfying P (roughly, a world containing no more than it needs to in order to satisfy P).. .

"Let us say that a sentence S is 1-necessary when its epistemic intension is true at all centered metaphysically possible worlds, and that it is 1-contingent when its epistemic intension is false at some centered metaphysically possible world. Let us also say that a sentence S is 2-necessary when it's subjunctive intension is true at all worlds, and that it is 2-contingent when it's subjective intension is false at some world.

"2-D thesis: If S is a posteriori, S is 1-contingent" (D.J. Chalmers 2004, pp. 269 298).

A system of Ostensive Indexical Identification of Objects/Processes

Chalmers considers:

“beliefs and concepts as psychological entities rather than as semantic entities. Beliefs and concepts have contents, but are not themselves contents. . . . .

“ Mary looks at a red apple, and visually experiences its color.  This experience instantiates a phenomenal property R, which we might call phenomenal redness. . . . Phenomenal redness (a property of experiences, or of subjects of experience) is a different property from external redness (a property of external objects), but both are respectable properties in their own right.” according to Chalmers (2004 p. 270).



Do these assumptions rig the game?  A correct understanding of semantics will uniquely map the psychology of belief and concepts. To hold semantic and psychological entities separate is simply to admit that your semantic theory is problematic, perhaps, exactly at the points that bear upon the problem of Mary. To hold, unargued, that phenomenal redness is ontologically distinct from external redness seems question begging. Why can’t property R be an aspect of external redness? It is one thing to use a distinctness assumption to argue that the thesis of distinctness is indispensable or should be preferred. But, to present the ontological distinctness of the two, properties, psychological-entity-R and external-redness, as brute fact would be simply dogmatic. Chalmers, though, is not being dogmatic. He is entitled to hold that concepts, as psychological entities, are not identical to the external properties of the world they are responsible for representing. There is space for the logical possibility of such non-identity even if it turns out that the referents of R and external-redness overlaps. Moving on:

"The reference of the concept expressed by ‘red’ or ‘phenomenal redness’ is fixed via a relation to red things and a relation to paradigmatic red objects ostended in learning the term. . . .The phenomenal concept plausibly designates an intrinsic property rigidly, so that there are counterfactual worlds in which red experiences are never caused by red things.

 

“One can distinguish at least two relational phenomenal concepts, depending on whether reference is fixed by relations across a whole community of subjects, or by relations restricted to the subject in question.  The first is what we can call the community relational concept, or redc.  This can be glossed roughly as the phenomenal quality typically caused in normal subjects within my community of paradigmatic red things. The second is what we can call the individual relational concept, or redI.  This can be glossed roughly as the phenomenal quality typically caused in me by paradigmatic red things.  The two concepts redC and redI will corefer for normal subjects, but for abnormal subjects they may yield different results.  . . .

"Phenomenal properties can also be picked out indexically. When seeing the tomato, Mary can refer indexically to a visual quality associated with it, by saying ‘this quality’ or ‘this sort of experience’.  These expressions express a demonstrative concept that we might call E.  E functions in an indexical manner, roughly by picking out whatever quality the subject is currently ostending.  Like other demonstratives, it has a ‘character,’ which fixes reference in a context roughly by picking out whatever quality is ostended in that context; and it has a distinct ‘content’, corresponding to the quality that is actually ostended—in this case, phenomenal redness.  The demonstrative concept E rigidly designates its referent, so that it picks out the quality in question even in counterfactual worlds in which no one is ostending the quality.

 

“The three concepts redC, redI, and E may all refer to the same quality, phenomenal redness. . . . There is another crucial phenomenal concept in the vicinity, one that does not pick out phenomenal redness in terms of its relation to external objects or to acts of ostension, but rather picks it out in terms of its intrinsic phenomenal nature. This is what we might call a pure phenomenal concept.  . . .Mary may learn (or reasonably come to believe) that red things will typically cause experiences of such-and such quality in her, and in other members of her community. She may learn (or gains the cognitively significant belief) that the experience she is now having has such-and-such quality, and that the quality she is now ostending is such-and-such. Call Mary’s ‘such-and-such’ concept her R.” (p. 271-272)..

Mary might come to believe:

redC = R

redI = R

E = R

Chalmers maintains that these are not a priori for Mary.

 

 

How Harry's brain is organized.

We can understand Harry's brain/world as consisting in six interacting modules, a motor activation center, an affective center, a sensory processing center, a language center, a perceptual interface and a world of possible experience.

Harry receives input into his 'brain' via a special node with links to nodes that represent his environment. Harry's world thus consists in three possible objects - examples of rocks, paper or scissors. Initially, each of the examples provides equal input into the system since the special node is locked at 1 and the links there from at .04. We can introduce variability into this world by simulating variability in how often the possible objects of Harry's world are presented for his attention. This is accomplished by randomly resetting the value of each link between the special and the nodes representing the objects of Harry's world using the 'Be Forgetful' button. We may interpret that what Harry experiences more often he remembers more clearly:  what he experiences less often tends to be forgotten or even rejects. This will be reflected in the size shape and color of these exemplary nodes and in the size shape and color of their corresponding concept nodes, etc.

In the model Rock Paper Scissors, no distinction is made between the possible objects and the sensory apparatus needed to experience the objects because the focus of that model was the ability of the model to simulate a subject predicate semantics and to entertain multiple equally reasonable courses of action. In the initial version of HARRY LEARNS ROCK, we see the necessity to distinguish a perceptual interface to provide a mechanism through which ostensive definition may operate. In this case the interesting contact with the world is on the output side. This is accomplished via a 'soft feedback loop' rather than back propagation.

To begin the soft feedback loop, a new element is added to Harry's world consisting of a demand that directly addresses Harry's language center [the sensory apparatus for the demand is omitted for simplicity] and contributes to excitation of the concept nodes for rock.  We may conjecture that Harry has previously acquired a lexicon through an appreciation of the possible interconnection of terms. He may have a theoretically based knowledge, e.g., rocks break, prior to and independent of his experience with rocks.

Harry now rigidly designates his paradigmatic sample through his ostending and touching a rock. Thus, the necessity of a node representing motor and sensory output, i.e., the 'that's a rock! Touching/pointing' node. Simultaneously with the action of picking out this particular rock as a sample of the direct reference of rock, Harry also indicates that he is picking out this particular something as essentially that very thing and not another thing. That is, he is asserting the haesity of the object named.  Dubbing dthat through the mechanism of an ostensive indexing procedure requires an action and thus is linked to the vspecial through the name-that valenced node. It is conjectured that this will involve Harry in an ontological commitment. It is not clear that ontological commitment begins as a concept in the sense required for linguistic manipulation. However, we postulate that some neural population computes a vector that at some point becomes activated in thoughts and activities requiring, or coinciding with ontic commitment.

When we run the model we find that the methods/actions, methods/concepts, and object concepts associated with rocks all get a boost. The end result is acceptance of the rock concept and along with it the concept for actions associated with rocks, as well as a propensity for rock related behavior. Harrys new visual system seems to amplify Harry's resolve and we now find that only the Theory that Rocks break is accepted.

If Harry's education is complete and he learns all the objects of his world in this ostensive way simultaneously, we might find that Harry returns to the balance exhibited when Harry knew about his world sans ostension.  But, Harry like us will not likely experience nor learn everything there is possible to learn or experience about the world.

The original HARRRY LEARNS ROCK model, however, was still not detailed enough. On that model Harry's sensory system was limited to 'hearing' a command and something like a 'tactile' system which he utilizes to name the various possible inputs. Harry's sensory system must be enriched beyond its capability to reach out to contact his world. The reach of the world is far too intimate on that model. As things stood in the initial HARRY LEARNS ROCK it appeared that the possible objects of the world deliver input directly to Harry's language center. Here in this model of ARU, HARRY LEARNS ROCK 3, we want to provide an input interface between the nodes representing possible world objects and the network of nodes responsible for making sense of that world. To this end, we now enrich the model to provide a rudimentary sensory input system based on the outlines of human vision.

In HARRY LEARNS ROCK 3, as in HARRY LEARNS ROCK 2, we provide Harry with a rudimentary visual input system.

"The receptive cells for vision are in the retina of the eyes and are of two types. One type are the rods, which can detect the intensity of light by the activation of a molecule called photopigment when light strikes it and it responds to any wavelength. The second type are cones, and each cone has a photopigment that responds mostly to a given wavelength of light, thus allowing color vision.

"The retina of the eye is a thin film covering the inside of the back of the eye and is conformed by three distinct layers. The first layer is made of the receptors, and their distribution is not homogeneous on the retina. Rods predominate the periphery of the retina while cones abound in the center, reaching a maximum at a point called the fovea, that is directly behind the pupil. The fovea is a specialized area for fine visual discrimination and is almost completely made of cones. The second layer is made up of small parallel bipolar neurons, that radiate perpendicular to the retina plane. Bipolar cells are entwined with interconnecting neurons that are called horizontal cells. The third level is made up by ganglion neurons, which are big cells with lots of dendritic ramifications. The axons of ganglionar cells all converge in one point to form a bundle and forming the optic nerve. In the second and third levels there are medium sized neurons called amacrine cells, that spread their processes (dendrites and axons) in a horizontal way to the retina plane, thus interconnecting bipolar and ganglionar cells among them and with each other. Each receptor (rod or cone) makes a synapse with a bipolar cell and it in turn synapses with a ganglion cell. Only one receptor connects to a bipolar cell, but several spatially contiguous bipolar cells synapse with a single ganglion cell. This produces the effect that a ganglion cell will respond to a small restricted receptive field of the whole visual field. This is important since these receptive fields are still found even in the higher neural relays of the visual pathway.

"The axons of the ganglion neurons of the retina form the optic nerves that go into the cranium and are joined to each other at a point called the optic chiasm. Here the fibers that come from the nasal part of the retina cross and project to the opposite side, while the fibers from the temporal (lateral of the head) side of the retina go on the same side. This produces the effect that whatever stimuli is presented in the right side of the visual field, will go to the left hemisphere of the brain, and viceversa. The new bundle of fibers made up of axons from the temporal retina of the same side and the fibers from the nasal retina of the other eye is called the optic tract.

"The axons of the optic tract divide in three distinct visual pathways. The first pathway ends in the lateral geniculate nucleus [LGN] of the thalamus, and it processes the visual information necessary for perception. The other two pathways continue into the midbrain. One ends in the pretectum, synapsing on neurons from which fibers go back to the eye to the cilliary ganglions that control pupilary movements. The other midbrain pathway synapses at the superior colliculus, and it controls the visually guided eye movements, which are the adjusting mechanisms that are not voluntary, such as saccadic movement and focus of the eyes. From the LGN, neurons that carry the visual information project to the primary visual cortex, which is located in the back of the brain, in the occipital lobe. This cortex is organized in vertical columns, and each column has a receptive field, analogous to the ganglionar receptive fields. The difference is that here not only detection of the presence stimuli is made, but each column may code for such complex attributes as sideways movement, verticality or horizontality, contrast, etc. The visual information then carries on to higher order sensory cortices and association cortices" (H. Gutierrez and C. 2007).

Churchland refers to this later organization of the visual sensory system in presenting what he considers a viable physicalist interpretation of color vision.

"In creatures with trichromatic vision (i.e., with three types of retinal cone), color information is coded as a pattern of spiking frequencies across the axonal fibers of the parvocellular subsystem of the optic nerve. That massive cable of axons leads to a second population of cells in a central body called the lateral geniculate nucleus (LGN), whose axonal projections lead in turn to the several areas of the visual cortex at the rear of the brain's cerebral hemispheres, to V1, V2, and ultimately to V4, which area appears especially devoted to the processing and representation of color information (Zeki 1980; Van Essen and Maunsel 1983; Hubel and Livingstone 1987). Human cognition divides a smooth continuum of color inputs into a finite number of prototypical categories. The laminar structure at V4 is perhaps the earliest place in the processing hierarchy to which we might ascribe that familiar taxonomy. A creature competent to make reliable color discriminations has there developed a representation at the range of familiar colors, a representation that appears to consist in a specific configuration of weighted synaptic connections meeting the millions of neurons that make up area V4.

"That configuration of synaptic weights partitions the "activation-space" of the neurons in area V4: it partitions that abstract space into a structured set of subspaces, one for each prototypical color. Inputs from the eye will each occasion a specific pattern of activity across these cortical neurons, a patttern or vector that falls within one of those subspaces. In such a pigeonholing, in now appear, does visual recognition of a color consist. (see P. M. Churchland 1989a, chapters 9 and 10, for the general theory of information processing here appealed to). This recognition depends upon the creature possessing a prior representation - a learned configuration of synapses meeting the relevant population of cells - that antecedently partitions the creature's visual taxonomy so it can respond selectively and appropriately to the flux of visual stimulation arriving from the retina and LGN." (Churchland 2004 pp 165-166).

According to Churchland, persons with impoverished learning environments may fail to develop a sufficiently rich interconnection of neural activation space to taxonomize experience in the same way that would be natural to someone with a more enriched learning environment. This, according to Churchland, explains in physical terms the difference between learning about color in a monochromatic environment vs. a polychromatic environment.

(Note: Some, including Churchland {reference?} have argued against the incommunicability feature of incommensurability thesis on the basis of a claim that there is always the possiblility of an inter-translation of all the sentences of competing theories. This, on its face, seems inconsistent with the crippled-neuro-taxonomy account Churchland offers here. Kuhn counters that differing taxonomic hierarchy is precisely the point at which incommensurability and is manifest. More about this later.)

We've grouped and routed all the nodes representing this limited possible world through three nodes that together play the role of the lateral geniculate nucleus. This small module encodes and passes the sensory information (analogous to vision) on to the language center module.

An Interpretive Narrative

We can suppose that Harry's existence is cartoonishly simple. Yet, he is the conceptual master of his universe. For the moment consider what Harry's existence is like just before the Tutor directs Harry to name something. We come to know Harry at that most opportune moment when he has a priori the potential knowledge of every possible state of his world in virtue of the interconnections of his neural net. You might say that we have pre-equipped Harry with conceptual representations of everything there is to know and provided him with interconnections that will give him the ability to consider every possible truth about the world available to him.

But those pre-experiential concepts will sit idle with out some job to do. It would be as if Harry sat in a vacuum with no care or concern. If we were all in a situation similar to Harry's predicament; where every allowable combination of concepts tended to equally valuable outcomes, would we have any reason to formulate any preferences? Would reason or knowledge matter? We would be, perhaps, 'comfortably numb'.

This idyllic state of poise has been altered, however, by Harry's tutelage and his attempt to name something. Based upon his new experience and the constraints imposed by principles of the theory of explanatory coherence (and its extension to affect and utility), it is plausible to describe Harry as justified in his belief that 'Rocks break'. Not only is the proposition internally coherent: the process of naming has produced some sensory validation that picks out a single unique possibility of his universe. But, knowing 'Rocks break' has some, perhaps, counter intuitive results. Harry does not accept it that 'scissors cut' nor that 'paper covers' nor any other combination of possibilities. Perhaps, we might conjecture, this is why he is also disinclined to behave in any other respect than through rock breaking behavior.

Ought we to say that Harry knows that "Scissors cuts." is false? No. This would be premature. We will say only slightly more modestly that until Harry meets some further conditions yet to be specified it is only that Harry withholds assent rather than that he asserts falsity. Harry simply isn't focusing his attention properly. But, let us boldly conjecture that provided with the appropriate further neural complexity Harry's prejudice in favor of 'Rocks break' would result in Harry's assent to the query, "Is 'Rocks break.' true?" or more directly the positive assertion, 'Rocks Break'.

An important question to be explored, only slightly aside from our current interest, is how care and concern influence the very possibility of knowledge. What is needed here to spark Harry's learning curve is for something to be at stake.

Now, post naming instance, what can we say about Harry? We have interpreted Harry's attention as focused by repeated encounters with some subset of the three middle sized objects that make up his world together with a very simple teachers command. We present these ways of focusing attention as implemented in the channels of information - sight and hearing. When we consider the impact of dubbing and re-dubbing, through the process of ostension, we get touch as well (or at least the focused and coordinated movement of pointing). These three, in combination with the a priori conceptual representations we've hardwired into the conceptual scheme, produce a minimally motivated Harry inclined to accept the theory 'Rocks break'.

What would happen if our Harry were to undergo surgery that crippled his ability to process incomming visual stimulations? We can model such a case using the lesion-link button to click on links that connect with the LNG. This can also be accomplished with the lesion-node button to delete nodes number 2 through 19. Going further we can also delete the link between World influence and node 33. In the later case we will, also, be crippling the hearing neural center's contact with the world. To interpret this additional loss, we might imagine that Harry obeys an inner voice which directs him to pick out not Rocks in the world, but rather, the place in the visual field that such rocks would appear if there were any. Similarly, Harry's ostensive behavior can be merely a coordinated motor behavior that does not causaly interact with objects in the world. Will the reader be surprised to learn that there is no difference in how the network settles?

Can we assess Harry to be a rational agent?

Mary Inside the Room

To make the move from Harry to Mary,

Mary Outside the Room

Individuation and counting Knowledge

REFERENCES

D. I. Chalmers 2004 Phenomenal Concepts and the Knowledge Argument, in There's something about Mary Essays on Phenomenal Consciousness and Frank Jackson's Knowledge Argument, Eds. P. Ludlow, Y. Nagasawa, D. Stoljar, MIT Press, Cambridge, Mass. pp. 269-298

P. M. Churchland 2004 , Knowing Qualia: A Peply to Jackson, in There's something about Mary Essays on Phenomenal Consciousness and Frank Jackson's Knowledge Argument, Eds. P. Ludlow, Y. Nagasawa, D. Stoljar, MIT Press, Cambridge, Mass. pp. 163-178

H. Gutierrez and C. Ormsby, 2007 the Instituto de Fisiologia Celular, UNAM, in Mexico City. http://www.ifisiol.unam.mx/Brain/vision.htm

F. Jackson 1982 "Epiphenomenal Qualia." Philosophical Quarterly 32: 127-136

F. Jackson 2004 What Mary Didn't Know, in There's something about Mary Essays on Phenomenal Consciousness and Frank Jackson's Knowledge Argument, Eds. P. Ludlow, Y. Nagasawa, D. Stoljar, MIT Press, Cambridge, Mass. pp. 51-56

P. Thagard 2000 Coherence in Thought and Action, MIT Press, Cambridge, Mass.

P. Thagard 1992 Conceptual Revolution, Princeton University Press, Princeton, NJ.

Back to Top of page

.

=====================================================================

= = = = = = = = = = ARU - AFFECT REASON UTILITY = = = = = = = = = = =

= = = = = = = = = = = = = = = = = = BY = = = = = = = = = = = = =

____________________WILLOW ANGELETTE_________________

======================================================================


= = = = = = = = = = = = = ABOUT THE MODEL = = = = = = = = = = = = = =

This model provides the user with the ability to rationally assess competing ideas on the basis of coherence using a connectionist formula to update the activation of interconnected nodes. The network is also a (PDP) Parallel Distributive Processing Model in the sense that each node acts independently and simultaneously* to update its activation with information about only those nodes with which it is immediately connected.

As activation of a node increases, the node size increases. As activation decreases, the node size decreases. Below activation 0.0 the node changes from a circle to a triangle representing rejection. Some models may have nodes that include a valence in addition to an activation. Changes in valence are indicated by color changes of the nodes.

With the exception of orange special nodes set at 1, the default activation of a node is 0.01. Changes of activation and valences is achieved via links between the nodes. Positive links, colored green, are created with a default weight of 0.04 while negative links, colored red, have a default weight of -0.06. The activation and valence of unconnected nodes will decay at a rate of 0.05 per cycle.

* the model mimics simultaneous updating by first updating each node in a network and then beginning it's next cycle.

Back to Top of page

======================== AFFECT =======================

Affective elements can be modeled by !ARU and are represented by nodes that connect with Special Star shaped nodes. In addition to activation these star shaped nodes deliver a valence set at a default of 0.05 for positive valences and - 0.05 for negative valences.



Principles to consider in the construction of coherence networks to model the influence of Affect are:



Elements have positive or negative valences.



Elements can have a positive or negative emotional connections to other elements.



The valence of an element is determined by the valences and acceptability of all the elements to which it is connected," according to Thagard (2000 p. 173).



This particular example models a coherence network with the influence of affect shown as a star.

Back to Top of page

======================== REASON =======================

The theory of explanatory coherence can be summarised in the following principles (Thagard 1992. Conceptual revolutions. Princeton, NJ: Princeton University Press, 2000. Coherence in thought and action. Cambridge, MA: MIT Press.)



Principle E1. Symmetry. Explanatory coherence is a symeteric relation, unlike, say, conditional probability. That is, two propositions p and q cohere with each other equally.

Principle E2. Explanation. (a) A hypothesis coheres with what it explains, which can either be evidence or another hypothesis;

(b)" hypothesis that together explain some other propositon cohere with each other' and (c) the more hypothesies it takes to explain something, the lower the degree of coherence.


Principle E2. Analogy. Similar hypotheses that explain similar pieces of evidence cohere.



Principle E4. Data priority. Propositions that describe the results of observations have a degree of acceptability on their own.


Principle E5. Contradiction. Contradictory propositions are incoherent with each other.



Principle E6. Competition. If P and Q both explain a propostion, and if P and Q are not explanatorily connected, the P and Q are incoherent with eachother. (P and Q are explanatorily connected if one explains the other or if thgether they explain something.)



Principle E7. Acceptance. The acceptability of a proposition in a system of propositions depends on its coherence with them.

Back to Top of page

======================== UTILITY =======================

Thagard (2000) offers the following principles to be considered in the design of PDP models of utility and which principles may be applied to !ARU:



"Principle L1: Symmetry Coherence and incoherence are semmetrical relations: if factor (action or goal) F1, coheres with factor F2, then F2 coheres with F1.

Principle L2: Facilitation Consider actions Ai, . . . , An that together facilitate the accomplishment of goal G. Then

(a) each Ai coheres with G,

(b) each Ai coheres with each other Ai, and

(c) the greater the number of actions required, the less the coherence among the actions and goals.[we may imagine this to be a way to interpert work ]



Principle L3: Incompatibility

(a) If two factors cannot both be performed or achieved, then they are strongly incoherent.

(b) If two factors are difficult to perform or achieve together, then they are weakly incoherent.



Principle L4: Goal priority Some goals are desirable for intrinsic or other noncoherence reasons.



Principle L5: Judgment Facilitation and competition relations can depend on coherence with judgments about the acceptability of factual beliefs.



Principle L6: Decision Decisions are made on the basis of an assessment of the overall coherence of a set of actions and goals (p. 128)."



Actions like hypotheses are evaluated with respect to their coherence with each other and with goals. In addition goals get a degreee of priority since they will connect directly with specials or else Vspecials via a valence node.

Back to Top of page

========================================================================

= = = = = = = = = = = = = USING THE INTERFACE = = = = = = == = = = = = =

Buttons and sliders along the top of the interface provide the user with the ability to load and run the default model and to modify its operation.



Buttons and sliders along the left allow the user to add and remove components of the model.

Networks are displayed in the large center square.

Information about the status of the network is displayed on demand at in the white rectangle at the right of screen.

Back to Top of page

========================================================================



================TO RUN THE DEFAULT MODEL=================

========================================================================

Click the button labeled HARRY LEARNS ROCK
click the button near the top left labeled "balance weights" then
click on "the button labeled "go".

As activation of a node increases, the node size increases. As activation decreases, the node size decreases. Below activation 0.0 the node changes from a circle to a triangle representing rejection. If "valence off" is not selected, color changes will represent the flow of valence through the network.

Click the button at the top right labeled display output to see display of the current status of each node.

Back to Top of page

==========================================================================

= = = = = = = = = TO RUN ADDITIONAL PRE-SCRIPTED NETWORKS = = = = = =



go here



Back to Top of page

==========================================================================

= = = = = = = = = = = = TO MAKE CHANGES TO A NETWORK = = = = = = = = = = =



TO DELETE NODES

Click the control labeled "leison nodes"
Using your mouse, click on each node to delete. All links to the node will be deleted as well.
Click the "leison nodes" control again after having deleted all nodes you selected.



TO ADD NODES

Several types of nodes may be created with default settings and user interpretations. NOTE: Any nodes that are added by the user can only operate correctly if the network is reset to it's initial conditions before making additions and "Accounts" of new relationships among nodes must be done only after all new nodes have been added. To run your newly revised network, click reset, then balance weights, then go.

Look in the left hand column of buttons.

Click the button labeled "AddObervation" to add nodes representing the interface between the world and how our senses interpret observation. A user input box appears allowing you to enter a brief descriptive label of the observation to be interpreted, e.g. "the wind turns me on". ARU will then create a new node of activation 0.01 with your discriptive label and identify the number of the node, the fact that it is to be regarded as evidence, and link it to a special node. At a later stage in developing your network you can use the mouse to link your new node to others.

Click the button labeled "AddHypothesis" to add nodes representing concepts that may account for some bits of evidence. A user input box appears allowing you to enter a brief descriptive label of your hypothesis, e.g., "Because the wind is high" Another box appears allowing you to assign your hypothesis to a collection of nodes representing hypotheses that give a collective name, e.g., "T1". ARU will then create a new node of activation 0.01 with your descriptive label, identify the number of the node, and tag it as a hypothesis. At a later stage in developing your network you can use the mouse to link your new node to nodes representing supporting evidence and other concepts withwhich it may cohere or in-cohere.



ACCOUNTS

Once you have created and linked all addtional nodes you can click the ACCOUNTS button to begin updating any new explanitory relations among nodes. Next, click a node you've newly created that is explained by another node or combination of nodes. An input box will appear which asks "Is this being explained by anything" Click the drop down arrow if the node has no explaination and pick "no" other wise click "OK" and a new box asks "How many other nodes will this node be explained by?" Using the AFFECT REASON UTILITY principles as your guide, count how many other nodes this node is explained by and enter the numeral in the blank provided. A new input box will now appear that asks you to provide a node number of each explaining node. Each time you enter a node number of an explaining node and click ok another input box will appear asking for the next node number until you have entered all of the node numbers that explain the node you at first clicked.

when the explainations are exhausted for the current clicked node no more input boxes will appear and you can click on the next node in need of explaination. When you're done exploring explanations click the ACCOUNTSbutton again to turn it off.

TO ADD LINKS



To create a positively weighted link, Click the button labeled click+linking-nodes

point your cursor at the node to which you'd like to link and click. The reset-click should turn to true if you've been sucessful.

point your cursor at the node it will link with and click. A green link will appear between the two nodes and reset-click will read false.

You may then create more positive links in this manner.

When you're done, click the button labeled click+linking-nodes again.

To create a negatively weighted link, Click the button labeled click--linking-nodes

point your cursor at the node to which you'd like to link and click. The reset-click should turn to true if you've been sucessful.

point your cursor at the node it will link with and click. A red link will appear between the two nodes and reset-click will read false.

You may then create more negative links in this manner.

When you're done, click the button labeled click--linking-nodes again.

TO DELETE LINKS

Click the control labeled "leison links"
Using your mouse point to the midpoint of the link to be deleted, click on each link to delete.
Click the "leison links" control again after having deleted all nodes you selected.

Back to Top of page

= = = = = = = = = = = = = = = = = = = = Options = = = = = = = = = = = = = = =



Be Forgetful: This button changes the character of the special link to evidence from activation from it's default value to random weights. The effect is that evidence gets "learned" at random and under the influence of the rest of the network some evidence may be forgotten.

Randomize activation: Each node will be reset to a random activation between 1 and -1. This can demonstrate whether a network will robustly produce similar results over a broad range of initial conditions.

Activation 0.01 : Returns the activation of all nodes back to the default initial conditions.

reset-click: If you're mouse click refuses to work, clicking reset-click may help.

resetcycle: Resets the count of cycles to zero.

supress valence: This sets the value of the valence to zero at each cycle. The network should respond by ignoring valences.

Move Node: This button appears torward the top right of the interface below the click reset control. Once clicked, hold the mouse pointer over a node you'd like to move and drag the node to a new position. When you're finished rearanging nodes click the Move Node button again to turn it off.

do-animate: When switched to the on position, nodes that cohere will drift together, those that in-cohere will drift apart. This, however, will slow updating considerably and is not recommened for networks larger than 5 or 6 nodes.

Back to Top of page

=======================SCRIPTING RULES====================

go here

==========================================================================