MARVIN MINSKY, THE EMOTION MACHINE , chapter 2 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ (Simon & Schuster, Nov. 2006) http://web.media.mit.edu/~minsky/ [Comments in square brackets are my own thoughts.] 2. Attachments and Goals ~~~~~~~~~~~~~~~~~~~~~~~~ He again distinguishes between 3 sorts of things that can be learned (in one's adaptation to the world), with the intention of focusing on the third: i. learning methods/tools -- what works, what doesn't (child making mudpie with bucket, fork & spoon) ii. learning the value or risk of places, situations, or activities (reproach from a stranger) [cf. Rolls' identification of the orbitofrontal cortex as the locus of associations between particular situations and positive/negative experiences; cf. also Fellous & Ledoux's "contextual fear", which in addition involves the hippocampus.] He presumably has in mind both unconditioned stimuli (like the pleasure of eating) and conditioned stimuli (like the potential rewards, and possible hazards, of being in a playpark). iii. learning goals or goal avoidance (mother's rebuke) "It's one thing to learn how to get what you want [where trial-and-error works well] -- and another, to learn what you *ought* to want [where pride and shame, induced by praise and blame, are crucial]". [But it seems to me he is making a sharp distinction where there is, in fact, a great deal of similarity: in all three cases above, we can say that a causal connection is learned between certain kinds of activities or situations, and ensuing rewards or punishments: (i) the use of the instrument leads to success or failure of an intention, and that seems (by our very nature) to lead to good or bad feelings; (ii) certain situations/places tend to lead to consequences perceived as threatening, which induces bad feelings; (iii) pursuit of certain goals or certain activities seems to lead to reproach from a caregiver, which (by our natures) makes us feel bad. So ultimately it is rewards or punishments that are paramount -- the type-(iii) goals that Minsky sets up as free-floating, with no superordinate goals, in fact *do* have superordinate goals, which is to gain rewards or avoid distress.] Goals ~~~~~ "You 'want' to achieve a situation G when some active mental process works to reduce the difference between G and your present situation". [Well, I think `wanting' *causes* such mental processes (in most circumstances), rather than *being* these mental processes -- Minsky's "proceduralist bias" is showing here. E.g., a man locked in solitary may *want* to be elsewhere, but if he knows there's nothing he can do, he is unlikely to be exerting himself mentally towards the goal of being elsewhere.] [His discussion of goal regression, and trying to close the distance to a goal is *very* much in the spirit of the General Problem Solver (GPS, until this started meaning soething else) of Newell, Shaw and Simon (1957)!!] Imprimers and Impriming ~~~~~~~~~~~~~~~~~~~~~~~ Imprimers are individuals to whom we become sufficiently attached so that *their opinion of us* matters to us. "Impriming" is the process (involving praise and blame, leading to pride and shame) whereby the imprimers shape our goals. The preliminary "theory" is depicted as |\ Praise detectors --->|A\ |N \ "Elevate the current goal" Imprimer detector --->|D / | / |/ [As I've indicated, I think that the learning here is just as in the case of "mud pies": we discover, more or less by accident, that pursuit of certain goals, or certain activities, brings rewards -- in this case, social ones -- and once we've learned that, we'll be more inclined to pursue those goals.] Learning and pleasure ~~~~~~~~~~~~~~~~~~~~~ Minsky points to the complexity of remembering what actions serve what goals -- you need the right sorts of descriptions, and you need to be able to retrieve them as-needed. He suggests that pleasure, grief, etc., are so hard for us to put into words because they involve very complex processes. [Contrast this with McDermott's view that we have trouble describing them because they are primitives of our *symbolic representations* of properties of ourselves. Minsky seems to prefer speaking exclusively of processes here, with representations getting short shrift.] Conscience, values, and self-ideals ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ He continues to dwell on the theme of goal acquisition through "impriming". In the absence of the imprimers later in one's life, these goals become "conscience", our "sense of right and wrong", etc. [But why are ethical precepts so widely shared by human societies (a point that McDermott makes much of)? Isn't there good evidence for altruism, or inhibitions against deadly violence, as a genetic trait of a species? E.g., why do fights between wolves rarely lead to serious injury, especially given that the loser will expose the side of his neck, or belly, to the winner as a signal of surrender? Minsky writes here much as if we were born with an ethical blank slate, which I don't think is true for either humans or wolves.] He suggests that a mechanism for becoming independent of human imprimers, as children become grownups, is to set up internal or other nonhuman surrogates (fictional heroes, even dogs -- is he perhaps thinking of "Son of Sam" David Berkowitz, who said he was commanded by a demon dog to commit his murders?). [What is not explained is why the need for some sort of imprimer persists.] He does make some comments on genetic factors, and makes the (frequently voiced) observation that children's love of play is serious business, because it promotes acquiring skills and knowledge about the world. Attachments of infants and animals ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ He goes further into this topic here, citing not only Konrad Lorenz but the more recent work of John Bowlby on what factors determine whom children attach to (viz., not necessarily a parent or other adult but possibly a child who is the same age or slightly older, the key factor being how quickly and forcefully that individual responds to actions or utterances of the child. Acc. to Bowlby, the function of attachment is physical protection -- see citation near the end of #2-8. That seems consistent with the selection as "imprimers" of individuals who react quickly & forcefully). He also mentions Bowlby's study of the stages a child goes through when separated for a long time from their parents or other caregiver -- the transition from active agitation, to more subdued unhappiness, to detachment [something observable from recent cases of abducted children who over several years became accepting of the abductor and indifferent to their parents.] He recaps much of common lore about imprinting in infant animals, and speculates about the development of sexual preference -- an intriguing puzzle [but I don't quite get what he's suggesting here. It's intriguing because a baby's innate concept of "human" seems at best to be extremely vague and crude. So it's totally implausible that babies are born with an innate ability to distinguish males from females perceptually, and a sexual preference for one or the other. Of course, they *learn* to distinguish males from females, but what pre-existing brain mechanism could possibly exploit such a learned distinction, and in most cases lead to attraction to the opposite sex? One might suppose that the predisposition lies in picking whichever gender (typically, mother or father) has less perceived similarity with oneself; but that is contradicted by the observation that children raised from babyhood by one parent or same-gender parents are no more likely to develop a homosexual orientation than others. Perhaps, as in moths and many other animals, the secret lies in pheromones (of which we are largely unaware at the conscious level) -- that would be the simplest explanation, because odor-based predispositions can certainly be innate.] Self-models and self-consistency ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ How can we control our actions? "The answer must be that we learn to represent things in extremely simple, yet useful ways." [Representations at last! But in relation to much of what we have read, Minsky seems to make too easy an assumption here that we *do* control our actions -- rather than that the thoughts about our actions that we become conscious of are merely high-level summaries of complex unconscious processes -- which do the actual controlling! Remember Libet's lagging will, as well as the discussion of "who controls the spotlight and action in the theater of consciousness" -- which may well be the (unconscious) "audience".] "Consider how remarkable it is that we can describe a person with words" [indeed! The same goes for our ability to describe our *experiences* with words, at various levels of detail or sketchiness; recall Vere & Bickmore's very preliminary work on this.] Minsky suggests that in turning such descriptions upon ourselves, we make our job easier by actively abiding by our own self-descriptions. In particular, if we abide by our self-descriptions, the future becomes easier to predict, because we are constraining what we might do in future. [An interesting thought, though not entirely new -- much the same is often said in the "BDI" literature (belief-desire-intention), as the chief factors governing the behavior of a human or human-like agent) e.g., by Michael Bratman, who points out that by sticking to our intentions we lighten the cognitive load in predicting the future as it concerns ourselves.] Minsky's added wrinkle is that he attributes this stabilization of the self-model in the early stages of a person's development to the precepts thrust on her by her imprimers... He also suggests that cults gain their converts by systematic destruction of the imprimer-imprimee relationship (e.g., by separating the intended convert from parents, etc., for extended periods of time, putting them in an entirely different environment), and then providing a new imprimer -- the prophet, or whoever, leading the cult, [I would add that they also seek to destroy the intended convert's self-model and self-respect, telling them they have been leading meaningless lives, as have all those around them. This prepares the victim for being handed new, different opportunities for attachment and self-respect.] 2.10 Public Imprimers ~~~~~~~~~~~~~~~~~~~~~ Given John Bowlby's observation about the formation of attachment depending on fast and forceful feedback, Minsky wonders how charismatic leaders manage to gain such large followings -- after all, a speech to a mass of people can't respond to audience members individually. He suggests that the trick lies in making many audience members feel AS IF they were getting direct responses. The skilled orator does this by *raising questions* in the minds of the audience, and then answering the question in a way that is carefully timed to appear like a prompt and apt response to the question raised. [Obviously Minsky has decided to use this technique very blatantly himself, by having "citizens", "students", "philosophers", etc., interject questions into his text -- ones he hopes may have arisen in the minds of the readers -- and promptly responding. He jokes about this, at one point having the imaginary audience accuse him of seeking to persuade them by that technique, rather than by sound reasoning.] Curiously, after talking about fictional attachments -- to characters in books, martyrs, gods, etc., he suddenly says that these attachments later in one's life are not an instance of impriming, but an instance of mentoring, a term not even used up until this point. In any case, his central contention remains that it's our attachments that give us our higher-level goals. [And this, to me, remains an overstatement. Approval and disapproval are innately pleasurable and distressing to us, just as certain physical interactions have those properties. So pleasure and distress are the ultimate driving forces. Morever, we have innate altruistic and violence-inhibiting dispositions, as well as selfish and violent ones. These too, depending on their relative strengths and of course opportunities for expression and development, strongly shape our ideals and characters.]