Sent Tue Feb 13 09:49:54 GMT 1996 Newsgroups: sci.psychology.consciousness References: <960206232153.236070a4@theorm.lbl.gov> Subject: Re: Tucson II [Function and Experience] Henry Stapp and Jeff Dalton rightly rebuked me for making over-confident assertions. I'll try to clarify. [HS (commenting on Tom Clark)]: > The issue is the difference between "A is B" and "A arises whenever B > happens, and only then", where A and B are empirically characterized > differently: A is a `personal experience' , B is a functional activity ^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^ > that is implemented in a brain. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ We may think we know exactly what we mean when we use the underlined phrases, and be deceiving ourselves. I have often tried to ask a question and then later realized that I was mistaken in thinking I understood what I was asking. I assume others have had similar experiences. A common example is starting by thinking you understand what free will is and after much thought and argument finding that the concept has evaporated. Sometimes a new, much more clearly defined, concept replaces the abandoned mush. A less common example is wondering whether (a) mathematical truths exist independently of us and are discovered, or whether (b) we invent or create them. I used to believe (a) and not (b), but the closer I looked the harder I found it to make any sense of the difference. Many mathematicians have not yet had the pleasure of that conceptual liberation, and still beat their breasts about it. [HS] > Let us suppose that A and B are empirically found to co-occur. Contrast that with discovering that two descriptions A and B don't describe anything different. E.g. your attempts to explain (even to yourself) what the difference is between A and B become incoherent and unconvincing. A could be "X is conscious" and B could be "X has the following co-existing, interacting functional states ......" (A very long description would be needed.) [HS] > ..But the claim that > nature simply makes one what conceptually and physically could be two > seems to need some explanation. It may not be nature's achievement. If you try VERY VERY hard to specify the differences between your concept of having experiences and your concept of being in a very precisely defined sort of functional state (using deep functional relationships, not just shallow behavioural dispositions) you may find that the difference evaporates. My conjecture is that as people learn more and more about what a deep functional specification might be, many will gradually come to realise that they were previously confused, as they find it harder and harder to specify anything that's in their intuitive concept of experience that's missing from the newly specified functional concept. > .... [HS] > ...there is no > justification for categorically asserting that ... Agreed. I apologise for categorical assertion. The exposition of a conjectural theory may look like categorical assertion if not every clause is appropriately qualified. Re-iterated disclaimers can get pretty boring, and we can all sometimes get carried away in the cut and thrust of heated debate. So please ignore my frequent and lamentable lapses into expressions of misplaced confidence, for which Jeff Dalton rightly chastised me. It's all still conjecture: an attempt to define a research programme, and an invitation to people to try thinking in new ways. But feel free to attack the content of what I say. [HS] >.... > theory must reduce to a structure that makes experience something else: Again, I can't speak for others, but I am not primarily attempting reduction of one thing to "something else". The question is whether we really understand experiences as anything other than functional states. If they are different, what exactly is the difference? Nothing I've read about the alleged difference adds up to anything that I can understand, except comparisons with shallow functionality that would clearly fail as a specification for really capable robot. A really capable robot is not just one that will answer mathematical questions, or compute trajectories for your starship, or assemble Fiats, but one that you would trust to look after your children. What are the functional requirements for that? That's a good test case to consider in trying to understand what functionalism has to be about. The evolutionary process seems to have included a similar criterion! For such a robot you'll not merely wish to specify its external behaviour but also many of its internal states, i.e. what it should attend to, how it should feel about things, what it should care about, what kind of learning it can do, how it can deal with situations not anticipated by its designers. How to *implement* all that is another question. [HS] > A. Sloman has sent a posting supporting Clark's message. >.. > But he seems instead to undermine it. He speaks of peoples impoverished > grasp of the problem, and his difficulty in convincing people that the > conceptual gap is illusory; The fact that it's difficult to convince people of a deep identity does not of itself undermine the claim. There are all sorts of deep things that people fail to grasp. [HS] > ...that they are unable to grasp relationships > between something ill-defined, and something they barely understand > because it has not been specified yet; that functionalism is hardly > a theory yet: it's a research program that's still in its very early > infancy; that it took him ten years to believe it; ^^^^^^^^^^ "Believe" is too strong. While trying to see clearly how experiences, qualia, etc. were conceptually distinct from what could possibly occur in a machine built entirely from physical components, I gradually found that as more and more sophistication was specified, i.e. more and more varieties of sophisticated causal powers (mainly *internal* powers concerned with information processing and control of information processing) were specified, it became harder and harder for me to discern any difference between what I was used to in myself and what I was talking about in such a sophisticated (hypothetical) machine. So I became attracted to the *conjecture* that by continuing the process of functional specification I could incorporate *everything* I knew about minds, and the conceptual difference would vanish. The more I examine requirements for visual perception, generation and management of motives, decision making, organisation of thinking processes, learning, etc. in an intelligent machine, the more I find that what is required turns out to include things that lead people to say that qualia, or experiences, exist in humans. I.e. the conceptual gap seems to be vanishing. In particular, a really sophisticated robot's self-monitoring and self-controlling capabilities appear to be indistinguishable from whatever I could observe in myself by introspecting. NB This is not an empirical discovery. It's a conceptual change arising from an philosophical/engineering activity: specifying requirements for a design. It's a re-crystallisation in a new sharper form of concepts that were previously very mushy, but apparently without losing anything of importance. That process also led (as explained in my previous message) to the notion that *maybe* there are plenty of experiences in me of which I am unaware because I can't access them, though other things in me can access (some of) them. (Isn't that close to what Freud said?) But I don't yet know what the range of functional requirements for a human mind really is (e.g. a mind that can perceive fast flowing rapids, that can enjoy the taste of new potatoes, that can be fascinated by a mathematical problem about infinite sets, etc.), so I am not in a position to prove (or even believe) that nothing more is required for human experiences to exist than satisfaction of some set of functional requirements. I don't know how long a full functional specification would be. Maybe close to the full Encyclopaedia Britannica? Maybe longer? And I still don't know what kinds of mechanisms are capable of supporting the implementation of such functionality. As far as I know, no existing AI systems, including connectionist AI systems, or hybrid digital-analog robot AI systems, get anywhere near it. (They have some tiny fragments.) [HS quoting AS] > ...and that "Until > we have our theory of deep functionality we may not have a good basis > for saying why some of the things that it refers to are identical > with what we previously called experiences, or consciousness." [HS] > But why should we be convinced by a counter-intuitive claim I don't expect anyone to be convinced so easily. It's no trivial matter, and certainly one or two messages from me could hardly produce a conceptual change as deep as that. I merely invite people to try imagining that there is no gap. It requires conceiving of some new possibilities. It's not easy. (Practice six times every morning before breakfast, and don't let initial failure put you off.) [HS] > ....that leads > only to a research program that after so many years is still in its very > early infancy. So many years? about 50? Compare the age of physics, chemistry, biology, geology, and the numbers of people who have worked in those fields. I suspect understanding what minds are is going to be much harder than any of these, partly because of the variety of types of minds and partly because of their intrinsic complexity: the two reasons are related. In addition, we are designed (for good reasons) to have partial and shallow insight into ourselves which leads us into deep muddles about what we are. [HS] > ....A more plausible position is that the very real > problems dealt with in connection with digital computers may not ^^^^^^^^^^^^^^^^^ > be directly related to consciousness: Information processing does not have to be digital. I don't have any views as to whether digital mechanisms suffice for the production of experiences or whether machines with continuously variable states will be needed. For all I know, it may need a mixture of digital fine structure manipulation under the more global control of continuously varying chemical soups or electrical fields, etc. I think digital mechanism may be required for reliable long term storage of some kinds of intricate structures. (E.g. memorising poems, nursery rhymes, mathematical formulae, grammatical rules) But that does not imply that everything has to be digital. Some aspects of experience SEEM to involve continous variation, e.g. watching a swan gliding down to a pond. Whether that's just continuity in a high level virtual machine implemented in lower level digital processes or whether some physical continuity is required I don't know (compare watching TV). If continuous variation at a physical level is needed for some human mental processes then poor old Turing will have to come up with a new machine. I have not yet found conclusive reasons to doubt that we can replicate everything relevant using purely digital machines, but it's still an open question. I do see reason to doubt that classical physical mechanisms will work within requirements of size, reliability, memory capacity, and energy consumption. I've also argued that parallel implementations may differ from serial ones from engineering viewpoint even when there's no mathematical difference. [HS] > ..maybe such machines are not > conscious. If you are talking about existing computing systems, I'd agree. There's a very long way to go. > Should not the research efforts pertaining to consciousness > begin with systems that can reasonably be assumed to be conscious? Yes. That's invevitably where we all start. But if we want to ask what a collection of functional mechanisms is capable of explaining you have to get into design mode. You can interrogate, look at, measure, video-tape, scan, or physically probe people, animals, brains as much as you like, but that in itself will not tell you the functional requirements they meet nor their design specifications. (Similar remarks apply to observing comptuers.) [HS] > ---systems that naturally and normally give reports that supply *some* > information (certainly not infallible) about the subject under > consideration? Yes. Certainly. But I would urge enormous care (even cynicism) in accepting anything *general* they say about their own mental states. What people say about PARTICULAR states (where their itch is, what they can see, whether they are feeling humiliated, etc.) is normally reliable, for that's something they were designed to be able to report on (though there are cases of brain damage and self-deception where caution is needed.) But as soon as people make any general statement about consciousness or experience, of the form "All Xs are Ys", "No Xs are Ys", I would not believe a word without independent corroboration, preferably in the context of a deep theory. [HS] > Sloman says "I think most or all people (e.g. Henry Stapp?) who say > experiences cannot play that role..." > > I have never said that. Quite the opposite. Apologies. I misread the following: | Date: Thu, 1 Feb 1996 15:51:35 -0800 | | ...it would seem that | the fact that certain functional structure are also experiential | structures makes no difference at all in the dynamical development. I read this as saying that experiences (qualia) could have no causal role. I should have realised you would not say that because of your view that experiences enter into the dynamics of quantum physics. Aaron