From stapp@thsrv.lbl.gov Thu Dec 14 08:53:00 2000 Date: Thu, 14 Dec 2000 08:48:11 -0800 (PST) From: stapp@thsrv.lbl.gov Reply-To: hpstapp@lbl.gov To: Aaron Sloman Cc: Brian Josephson , hpstapp@lbl.gov, Kathy Laskey , Stuart Hameroff , pat hayes , brings@rpi.edu, brucero@cats.ucsc.edu, ghrosenb@ai.uga.edu, Jeffrey M. Schwartz , keith@imprint.co.uk, klein@adage.berkeley.edu, kpribram@runet.edu Subject: Re: Science and the Human Person/Perception of mental states. On Wed, 6 Dec 2000, Aaron Sloman wrote: > > 6 Dec 2000 (02:13 GMT) > Henry, > Just before I go to bed, this caught my eye in your latest: > > > 2. The scientific data pertaining to the human person is > > basically of two kinds: > > > > 1) Reports about the body/brain of the person: > > e.g., reports about readings of instruments that probe > > the person's physical body/brain: > > These are reports about spacetime locations, motions, > > and shapes of observable physical objects and systems. > > > > 2) Reports by the person about what he or she is experiencing: > > e.g., personal reports about feelings, perceptions, etc. > > There's lots more. E.g. > > 3) I can tell that someone is angry, or jealous, or obsessive, > or ambitious, even though the person does not report these things and is > possibly even totally unaware of them. (Shakespeare knew that well.) > > In describing them thus I am NOT talking about, or trying to explain, > spacetime locations, motions, shapes. etc. etc. in category 1. I am > talking about states, processes entities within the information > processing architecture that constitutes that person's mind, the vast > majority of which he is completely unaware of. These descriptions are > not derived by some inductive or deductive process from observations > of data, any more than descriptions of the atomic structure of matter > are derived deductively or inductively from observations of pointer > readings, etc. > Workers in any line of work develop a language and conceptual structure for organizing and communicating their thoughts about the aspects of their experience with which they are jointly dealing. Thus a TV repairman might speak of condensors and resisters and wires, etc., and be very effective without ever talking about, or knowing anything about, germanium and its properties. A biologist in 1940 could do a lot of biology, and be content, without knowing anything about DNA and the expression of genes. But we understand, now, how high-level theories and concepts have limitations, and that more comprehensive understandings can arise by tieing different levels of description together. Descartes provided a useful scheme for tieing the different levels together: one conceives the world to have these two types of aspects, physical aspects, characterized by spacetime properties---he invented analytic geometry---and mental/experiential aspect, with interaction between the two aspects localized in the brains of human beings. This is an exceedingly useful way of organizing one's thinking/theorizing/communicating: it lead to the classical physical theory of Newton and Maxwell. Even though one often cannot DEDUCE the necessary emergence of the various useful higher-level properties that workers in specialized fields have identified, and named, one can understand how spacetime shapes are consequences of the elementary properties of the properties postulated in classical physical theory, and use these properties to link together different level of description. So you and I are both interested in "> processes entities within the information > processing architecture that constitutes that person's mind, the vast > majority of which he is completely unaware of." But there is an interesting question of how this structure is connected, on the one hand, to the aspects of the human person studied by neuroscientists, and, on the other hand, to the person's conscious thoughts. Although the language and concepts that you use are not immediately the language of the neuroscientist or the phenomenologist, there is the possibility that understanding the process of immediate to you will be broadened by imbedding it in the concepts of neuroscience and psychology. > I have friends here in neuroscience who study effects of brain damage. > They too do not seem to need to bother with whether the damaged > components use classical or quantum mechanisms. But when the components > stop working normally they can have some bizarre effects that can be > described without mentioning anything a physicist would know about, and > in many cases neither do the poor individuals concerned. > Call this type 3) or type 4). It's neither 1) nor 2). > The bizarre effects are behaviors of the person, and hence of type 1. > > 3. To discuss the internal dynamics of a human person > > two physical theories can be considered: > > Classical (physical) theory. (CT) > > Quantum theory. (QT) > > Both are completely irrelevant when we are trying to understand the > information processing architecture referred to above. I spend most of > my life thinking about the latter and in doing so I completely ignore > both classical and quantum physics. I do not believe this hampers me in > any way at all. > > But you must admit that you have not yet achieved a good solution of the problems you face, so your belief could be quite wrong: a biologist who does know about DNA may believe that he can understand development, or some other biological feature, without considering DNA, but he could be quite wrong. More generally, the various levels of description of a complex system such as the human person may be in need of mutually support. Indeed, let me skip ahead to your later message: Subject: Re: Perception of mental states [Was Science and the Human Person (Revised)] On Sun, 10 Dec 2000, Aaron Sloman wrote: > > [AS] > I agree that most of the processes in perception are completely > unconscious. > > However there is lots of work in philosophy (starting with Kant, or > maybe Plato?), neuroscience, psychology, and AI showing that data-driven > processes whether inductive or deductive cannot account for perception. > > I think it was von Helmholtz who summed it up over a hundred years ago > by saying that perception is controlled hallucination: a far cry from > any normal understanding of either of the terms "Induction" and > "Deduction". Pat used to work on finding ways to combine deduction > with *control* mechanisms in order to address this problem. (Mid 1970s). > I think the project petered out when he emigrated to the USA. > > If you look in detail at the locally highly impoverished and ambiguous > information available to sensory transducers (e.g. photons hitting > retinal cells, acoustic stimulation of auditory nerves) you'll find that > the processes of interpretation involve enhanced versions of all > the problems I previously described regarding the interpretation of the > necker cube (many alternatives leading to explosive combinatorics). > Real-world perception includes many additional problems, e.g.: > segmentation is highly ambiguous (which bits should be grouped with > which), many levels of abstraction are required in order to get a handle > on important aspects of the scene, and in general the available data are > mathematically incapable of determining a unique interpretation. > > Simple line drawings are bad enough, but even with line drawings the > combinatorics can explode as the number of line segments goes up. Look > at a typical Dennis-the-menace cartoon drawing, with tens or hundreds of > straight and curved line fragments instead of the 12 (maximal) straight > lines in the necker cube. Look at small (5mm square) regions with > a magnifying glass to see how many ambiguities there are about which > lines belong with which. > > It's worse if the optic array mostly lacks sharp edges and instead you > have to perceive mostly smoothly changing gradients in texture, colour, > intensity (e.g. seeing a naked human torso, sand-dunes, clothing, > cloth-covered furniture, bananas viewed close up). > > Binocular vision adds to the complexities, since you have to measure > discrepancies between the two optic arrays, but you can't do that until > you've worked out which bits correspond, which is not easy because they > are not completely in registration (if they were, binocular vision would > not give depth information). Finding the correspondences is not too bad > if the images are made of discrete lines: handling smooth curved 3-D > surfaces is another matter. > > Then there's the higher order grouping problem in seeing motion, > including motion of superimposed scenes, e.g. trees waving in the breeze > seen through a dirty window as you walk past it, especially if the > window also has reflections. > > Animals with deliberative capabilities need to see not only what's there > but also what's possible (possible changes of location, shape, > orientation, relationships, causal influences, etc.). I am generalising > Gibson's notion of "affordance" here. > > And finally, social animals need to be able perceive states of other > animals that are socially relevant, including fear, anger, submission, > etc. More on this below. > > Mechanisms for coping with all this are required that are FAR more > subtle and complex than either induction (looking for correlations in > the data, forming generalisations) or deductive inference (using logic > or mathematics to infer conclusions from some combination of general > and particular premises). > > It seems that in animals (and robots) perceptual mechanisms have to make > use of variable proportions of: > > sensory data, > > prior (implicit) theory > (including an appropriate presupposed ontology, which may > need to be different in different contexts of goals, needs > and current environment), > > prejudiced processing mechanisms abstracting possibly useful > groupings, features, properties, relations, histograms > from the data, (probably also switchable/tunable according to > context) > > a variety of forms of representation for use in different > stages or levels of the interpretation process > > subtle context-sensitive, dynamically modifiable, optimisation > mechanisms controlling the combinatorics of the search > for good global interpretations. (Maybe lots of different > such mechanisms operating concurrently at different levels of > abstraction, using different principles to suit the different > sub-ontologies.) > > and probably more besides, depending on the current function of the > perceptual process (e.g. whether it is part of a tight feedback control > loop involving external action, or part of a planning process, or part > of a recognition process, or part of a learning process, or part of a > communication process, etc.) > > Finding an architecture that can combine all of this (along with other > requirements for mind) in a successfully functioning system is a very > hard research task on which, as far as I can tell there has been much > reporting of detailed facts, many sketchy and inadequate failed > theories, and *very little* real progress. > > (If anyone can explain how quantum mechanisms can make it any easier > I'll start taking claims by Henry, Penrose, etc. more seriously). > ... > I'll apologise even more to Henry if the working system essentially uses > non-local Quantum mechanisms or something else that can be understood > only in terms of quantum physics. > > I'll apologise a bit less if it *merely* uses the same mathematics as > QM, but does not presuppose a QM infrastructure, > > [AS] > > >any more than descriptions of the atomic structure of matter > > >are derived deductively or inductively from observations of pointer > > >readings, etc. I am speaking about using a rational mathematical theory created/invented by quantum theorists to account for empirical mathematical regularities. ... > > (I've just been reading Richard Feynman's "The Character of Physical > Law", a fascinating set of lectures he gave on the BBC in 1965. It's > short, written with brilliant clarity, and generally an excellent read. > > The last chapter is a valiant attempt to say something about how > scientists find new laws, but he confesses ignorance about much of it, > and conjectures that they have to use different thinking strategies at > different stages in the development of science. > > One thing he missed, though it is implicit in much of what he wrote > about, was that the most important developments are not ideas about new > LAWS, but good ideas about how to extend the current ONTOLOGY, for that > opens up a whole new realm of possible laws to investigate. Neither > induction nor deduction (as normally conceived) can produce new > ontologies. However they can usefully point to problems in the current > ontology which then motivate scientists to search for something new, > using something other than induction and deduction.) You have described huge computational problems that the human person faces in arriving, on the basis of meager clues, at an appropriate response. What I am suggesting is that the difficult problem that you face requires for its solution a grounding in a new dynamics rooted in a "new" ontology. The two main ideas of this new ontology are: 1) The basic ontological reality lies midway between the classical conceptions of mind and matter: the basic reality is information, with bits of information injected by experiential events, but instantiated in, and carried forward by, clouds of virtual realities of the classical type. 2) The dynamics involves three processes. The first process is analogous to the the dynamical process in classical mechanics: it is local-deterministic, and involves the usual physical variable (e.g., masses and charges of the electrons, protons, neutrons, etc.). It governs all unconscious processing in a massively parallel way. The second process is an evaluative process that selects a proposed course of action on the basis of its (projected) experiential "feel", The third process either actualizes or vetoes that proposed course of action in accordance with a statistical rule. This ontological structure has, on the face of it, some important advantages in the processing that leads in human beings from meager clues to an actualized course of action. 1) The basic structure is informational, not substantive. 2) Massive parallel processing can be combining the clues in a huge number of different ways, all at the same time, in the milieu of strong environmental decoherence, and constructing, on the basis of these processed clues, corresponding "appropriate" possible courses of action. 3) An evaluative process that looks simultaneously at all of the huge number of possible courses of action and can select or pick-out any one that meets the criterion of "feeling right", and propose that possible course of action to nature for actualization or rejection in accordance with the statistical laws of QM. 4) This evaluation process works on the "feels" of prior experiences in order to construct a feeling for a proposed course of action. There is freedom within the constraints of quantum theory for specifying how this evaluation proceeds. It appears to allow for the rational evaluations called for by Dennett's "Intentional Stance" and Kathy's evaluative procedure. 5) Some such process is needed to complete the quantum dynamics, and the details so far put forward about it seem to be in line with evidence coming from psychology and psychiatry. So although you believe that you can ignore physics, I would venture the opinion that progress on the very difficult problems that you have described, and upon which you say almost no progress has been made, in spite of your lifetime of effort, might be significantly speeded by opening up your thinking to the very-non-classical concepts of the dynamical processing of information by the human brain that arise naturally from quantum theory. Best regards, Henry