NOTE-348: The Emotion Machine

Contribution

The draft is the late statement of the Society of Mind programme, and it changes three things in it.

• It renames the earlier book’s “agents” as “resources” and gives the reason. Readers took an agent to be “a personlike thing (like a travel agent)”.
• It puts emotion inside the machinery of thinking. An emotional state is a large-scale selection of resources, of the same kind as an “intellectual” Way to Think.
• It organises the resources into a Critic–Selector loop running at six levels of reflection.

The rest is the earlier programme carried further: K-lines, frames, panalogies, censors and negative expertise, the Single-Self as fiction. The draft sends the reader back to The Society of Mind (LIT-429) for the details.

Key insight

The mind has no one way of thinking, so it needs ways to notice that the current way has failed and to switch. A Critic recognises a type of obstacle: “I’ve made many attempts with no success”, “Achieving goal A made goal B harder”. A Selector then turns on a different set of resources: a different Way to Think. When that switch is large enough to spread into a “cascade” that changes goals, perception and priorities together, we call the result an emotion. “Resourcefulness” is having many such ways and good Critics for choosing among them. “Intelligence” is a suitcase word for it.

Assumptions

• Method. The arguments are thought experiments, everyday vignettes (Joan’s knee, Carol’s mud cake, Charles in love), dialogues with imagined objectors, and appeals to AI programs Minsky’s group built (Builder, Evans’s analogy program, GPS). There is no experiment, and the neural claims are few and hedged (“I expect that in the next few years, we’ll discover structures like those in this diagram”).
• Functionalism. What matters is “what each separate part does, and how those parts are connected up”, not what it is made of (§1-5).
• Complexity first. The draft rejects parsimony for psychology on purpose: “this book will embark on the opposite quest: to find more complex ways to depict mental events that seem simple at first” (§1-2; also §5-6, “No theory should have more parts than it needs” is answered as having “retarded Psychology”).
• Conjecture, marked as such. The Imprimer theory is “only theories of mine… These ideas might be right but they also might not” (chapter 2 summary). Disruption cascades in suffering are something brain scans “should” one day show (§3-2).
• Levels are a modelling convenience. “I agree that those boundaries are indistinct” (§5-6). The Critic–Selector model and Model Six are two views to switch between, not one architecture (§5 summary).

Key results

The draft proves nothing; these are its main proposals, by chapter.

• Emotional states (ch. 1). The thesis: “Each of our major ‘emotional states’ results from turning some set of resources on and turning another set of them off” (§1-5). Selectors do the switching. When the switched set includes other Selectors, the change spreads into a “cascade” (§1-7). Infants switch whole states at once; adults learn to mix them (§1-6). “Emotion”, “love” and “understand” are suitcase words.
• Attachments and goals (ch. 2). Trial and error teaches means. Attachment-related praise and blame teach ends: “Pride and Shame… help us learn ‘ends’ instead of ‘means’”. The people who can do this are “Imprimers”. Internalised models of them become “conscience”. A goal is defined operationally: you want G when “some active mental process works to reduce the differences between G and your present situation”.
• Pain to suffering (ch. 3). Pain is the first sensation, hurting its early effects, and suffering the large-scale cascade that seizes resources and takes away “your freedom to choose what to think about”. Critics come as Correctors, Suppressors, Censors and Self-Controllers. “Negative expertise”, knowing what not to do, may be much of competence. Creativity, humour, decisiveness, pleasure, beauty and mystical experience are each read as Critic suppression (§3-5). One emotion can be used to switch off another, as when anger is used against sleep (§3-8).
• Consciousness (ch. 4). “Consciousness” is a suitcase word, unpacked into 27 listed activities (§4-2). The sense that it is one thing could come from a detector C that fires when four things run together: recent memory, serial processing, symbolic description and self-models (§4-3). The “Immanence Illusion”: answers that arrive before the question is asked feel instantaneous (§4-3.1). A B-brain watches and corrects an A-brain (§4-9). Chalmers’s hard problem is called a compression of many problems into one word (§§4-8, 9-6).
• Levels (ch. 5). Model Six has instinctive reactions, learned reactions, deliberation, reflection, self-reflection and self-conscious reflection. Deliberation uses If→Do→Then rules chained into plans. Search from both ends cuts a 20-step search “several thousand times”. Imagination works by changing high-level descriptions (“simuli”), not pictures. A prediction machine with “suppressor bands” can stop and think (§5-9). Minsky says this scheme appears in §6-7 of his 1953 PhD thesis.
• Common sense (ch. 6). Knowledge is organised in realms: physical, social, dominion, procedural, economic, conversational and more. Panalogies link the same item across realms (§6-1). A rough estimate puts a person’s commonsense knowledge at “a few dozen millions of items”, leaning on Landauer’s figure of about 2 bits per second (§6-2). “Baby machines” are said to have stalled for want of new representations, under three paradoxes. The Optimization Paradox: the better a system works, the more likely each change makes it worse. The Investment Principle. The Parallel Processing Paradox. A goal is a Difference-Engine (GPS) process (§6-3). “Free will” is “a process we use to stop other processes” (§6-5).
• Thinking (ch. 7). The Critic–Selector model (§7-3). Lists of Ways to Think and of Critics at each level (§§7-4, 7-5). Emotional embodiment, discussed with James (§7-6). Poincaré’s preparation, incubation, revelation and evaluation. A conjecture that everyday thought alternates “micro-manic” generation with “micro-depressive” criticism (§7-7). “Cognitive contexts” hold the state to return to after an interruption (§7-8).
• Resourcefulness (ch. 8). Distance is judged many ways, so failures are tolerated (§8-2). Panalogies explain why rooms and objects stay put as we move (§8-3). A “mental animal trainer” copies short-term descriptions into long-term memory by repeated difference-reduction (§8-4). A K-line remembers a Way to Think by re-activating the resources used (§8-4). Credit should go to the earlier choice that selected the winning strategy (§8-5). “Genius” is an unusual combination of common traits (§8-6). Representations run from numbers, which are “opaque”, to semantic networks, frames with defaults, trans-frames and scripts. “Micronemes” are context signals broadcast widely (§8-7).
• The Self (ch. 9). “Whenever you think about your ‘Self,’ you are switching among a network of models” (§9-1). Sub-personalities, traits as self-trained regularities, and the Barnum effect as matching some self-model (§9-2). Reasons the Single-Self idea is convenient (§9-3). Pleasure as a suitcase of satisfaction, exploration, relief and Critic suppression (§9-4). Why a mind cannot be a company: its parts are too specialised. The Parallel and Pinnacle Paradoxes (§9-5). Qualia as complex reflective descriptions, with the “Easy is Hard Paradox” (§§9-6, 9-7).

Claims

id	claim	strength	support
C1	Each major emotional state is a selection of resources turned on and others off, and is a “Way to Think” of the same kind as intellectual ones	weak: a design proposal	thought experiments (§§1-1, 1-5, 3-8); no measurement
C2	Attachment figures (“Imprimers”) change a child’s goals through pride and shame, where trial and error changes only methods	weak, and marked by Minsky as his own untested theory	Bowlby, Harlow, Lorenz and Goodall on attachment, cited (§2-7); no evidence for the goal-elevating mechanism itself, which Minsky says no brain research has shown (§2-3)
C3	Suffering is a cascade in which prolonged pain seizes resources and removes freedom to choose what to think about	weak	phenomenology of complaints; Melzack and Wall quoted on pain’s wide brain involvement; “when scanners show more… we should be able to see those cascades” (§3-2)
C4	”Consciousness” names many processes, and the sense of one thing could come from a detector of four co-occurring features	moderate as a conceptual argument (suitcase analysis); weak for the detector	§§4-2, 4-3; argued, not modelled
C5	No resource can know what another is doing “right now”, only recent records (“Immanence Illusion”)	moderate: a plausible argument from signal delay and caching	§§4-3.1, 4-7; Dennett and Kinsbourne quoted
C6	Purely numerical representations, as in neural networks and statistical learning, are opaque to reflection and limit a system’s growth	weak, and a 2005 prediction	argued from the number 12 having many histories (§8-7); no test
C7	The body is one more set of resources; bodily expression may sustain emotional states as an external memory, but emotion starts in the brain	weak: argued against James, with no data	§§3-8, 7-6
C8	Elliot’s decision failures might have caused his emotional flatness, not the reverse, through damage to Critics	speculation, offered as an alternative reading	§7-2; no evidence offered, and the gambling-task data are not discussed
C9	A person’s commonsense knowledge is of the order of a few dozen million items	weak: an order-of-magnitude estimate	§6-2, from rough counts and Landauer (1986)
C10	The Self is a network of partial self-models switched among, and a single unified model would be impractical	moderate as an argument	§§1-8, 4-5, 9-1; analogies with physics and car repair

Concepts

• Resource. Any structure or process the mind uses, from perception to bodies of knowledge, used “in a hazy way” (§1-5). It replaces the earlier book’s “agent”.
• Critic / Selector. A Critic recognises a problem type and activates a Selector, which turns a set of resources on or off. At the lowest level they are just an If and a Then (§7-3, note 3).
• Way to Think. A set of active resources suited to a type of problem. Emotional states are a subclass with popular names.
• Cascade. A spreading change in active resources large enough to alter goals and perception (§1-7).
• Imprimer / impriming. A person a child is attached to, and learning values in their presence (§2-3).
• Suitcase word. A word that packs many different processes, such as “consciousness”, “emotion”, “pain” or “pleasure” (§§1-1, 4-1).
• Immanence Illusion. “For most of the questions you would otherwise ask, some answers will have already arrived before the higher levels of your mind have had enough time to ask for them” (§4-3.1).
• Model Six. Six levels: instinctive and learned reactions, deliberation, reflection, self-reflection, self-conscious reflection (ch. 5).
• Simulus. “A counterfeit perception caused by changing a mental representation” (§5-8).
• Panalogy. A “parallel analogy”: one structure linking corresponding roles in several realms, so a view can switch without starting over (§§6-1, 8-3).
• K-line. A Selector that re-activates the resources used when a problem was solved, a “snapshot” of a Way to Think (§8-4).
• Microneme. A widely broadcast context signal, often nameless (§8-7).
• Negative expertise. Knowing what not to do; carried by Censors and Suppressors (§3-5).

Connections

• The Society of Mind (LIT-429). The draft carries that book further and leans on it for details. It cites its sections throughout: K-lines to §8.1, panalogy to chapter 25, verbal expression to §22.10, “Closing the Ring” to §19.10, the self as a friend inside to §17.1. Its §9-5, note 24, says the earlier book was read as a “community view” against its intent. This is the extends: relation on the LIT.
• Singh’s exposition (LIT-444). The draft’s Introduction cites “Push Singh 2003” for how the two books relate, and the draft credits Singh’s “Reflective Critics” for Critics that reflect (§7-3, note 3). Singh’s footnotes describe this book’s selector K-lines, panalogies and six levels before publication, and they agree with the draft.
• The earlier memos and papers, being filed by another reader at the same time. The draft names its K-line idea as first developed in “Plain Talk” (1977) and “K-lines” (1980) (§8-4, note 17). It cites the frames memo for default assignments and for the room example (§8-3, note 7; §6-1, note 2). It cites the jokes essay for jokes as negative expertise (§3-5, note xvi). §9-5 quotes a passage on “the paradox-oscillation of ultimate purpose”, attributed to “Jokes and the Logic of the Cognitive Unconscious”. That is the heading of the memo as Minsky posted it on his home page. The memo (AI Memo 603, November 1980) and its 1981 published version are titled “Jokes and their Relation to the Cognitive Unconscious”.
• Damasio (LIT-383 to LIT-388). The draft cites only Descartes’ Error, which is not in this record. The comparison:
  • What an emotion is. For Damasio it is a body state or its as-if simulation, which marks options (LIT-384). For Minsky it is a selection of resources. The body is “just one more set of resources to exploit” (§3-8). “However, we should not too closely identify these [somatic effects] with how Anger changes one’s Ways to Think.” This is the disagreement, and it is direct. The Introduction announces that chapter 7 “will take the opposite view” to the bodily one.
  • The body’s role. Minsky grants James that bodily expression may keep a state alive as a “dependable external memory device” (§7-6). This is near Damasio’s body loop. But Minsky makes it optional: “exaggerating the body’s role in emotions can lead to serious misconceptions”.
  • Elliot. Damasio infers that reduced emotion degraded decision. Minsky proposes the reverse: lost Critics degraded decision, and lost cascades followed (§7-2). Minsky gives no evidence. The gambling-task result (LIT-383), with anticipatory skin-conductance responses before the advantageous strategy is known, is the kind of timing evidence that bears on the direction, and the draft does not engage it. The record’s reading of that report finds its statistics weaker than its abstract on this point, so neither side’s case is closed.
  • Feeling. Damasio: feelings are the experience of body states and report homeostatic value (LIT-388), and they are the source of consciousness (LIT-387). Minsky: “what we call ‘feelings’ are attempts (by various parts of our minds) to describe large-scale aspects of mental conditions” (§3-3). Their objects are brain cascades, and they need reflective self-models (§9-6). So the dependence runs the other way between the two authors.
  • Agreement. Both reject a separable, emotion-free rationality (“a myth”, Introduction and §7-8). Both put emotion inside decision rather than against it.
  • An unmarked parallel. Minsky’s K-line, re-activating the set of resources that were active, has the shape of the recall-as-reactivation in Damasio’s convergence-zone proposal (LIT-385). Neither cites the other on this; the pairing is mine.
• Global workspace. §4-6 quotes Baars and Newman’s workspace and accepts it as “a good way to start”, then says the mind needs “a great deal more architectural structure”. It is one of the theories Butlin et al. (LIT-056) turn into indicator properties.
• Collective intelligence and group minds. §9-5 argues a mind is not like a company or town, because its parts are too specialised. It asks whether a city could have a Self, and answers that it would seem to have more of a personality only if its departments’ models were linked into panalogies. This bears against reading Pilgrim et al.’s collective intelligence (LIT-046) or Schwitzgebel’s conscious United States (LIT-159) as models of a single mind. The bearing is mine; the draft names neither.
• Others the draft quotes that the record holds or lists. Ryle’s The Concept of Mind (LIT-318, unread here) on unnoticed and mistaken sensations (§9-7). Chalmers on the hard problem (§§4-8, 9-6). Dennett throughout. The draft’s “external representations” Way to Think (§7-4) and the body as external memory sit near the Extended Mind (LIT-097), which it does not cite.

Bearing on the record

• NOTE-136 glosses “suitcase word” as “Minsky’s term (ref. 22)”. The reference list of Safron’s editorial (LIT-195), checked on Crossref, gives ref. 22 as Minsky, The emotion machine, 2007, 400 pp., the paperback. So NOTE-136 already names this work, and the draft confirms the term (§§1-1, 4-2.1).
• THEORY-045, on felt affect as the body’s regulatory state, names this reading’s account of feelings (states of the brain, not of the body) as its dissent, and THEORY-056 cites its emotions switching one another off as a design-level neighbour. The Elliot case and the gambling-task timing remain the test between the two views.
• Boundary. The draft carries no instruction for machine-learning practice. It does make claims about machine learning: §§6-2 and 8-7 argue that numerically represented learners (“Neural Networks, Perceptrons, Fuzzy Logic systems, and Statistical Learning Programs”) are opaque to reflection and stall. These are 2005 predictions, untested in the draft, and an anthology reader might want them as historical positions on representation. I did not tag the work anthology-candidate: its subject is the human mind, and the ML claims are a minor thread.

Limitations

• It is a draft, and it is not the book. Chapter 9 is partial and full of editorial brackets (“[edit]”, “[This section should begin by…]”). Several ”§§” cross-references point to sections that do not exist in it (§§Embodiment, §§Alarms, §§Duplication, chapter §10).
• The figures are missing from what I read.
• There are no experiments. The neural claims are guesses about what scans will show. Its account of emotion does not engage the affective neuroscience of its time beyond Damasio’s one patient and a footnote on Panksepp.
• It names many mechanisms (Critics, Selectors, panalogies, micronemes) but specifies none precisely enough to implement or test, as Singh (LIT-444) says of the earlier book.

Open questions

• Does the published book keep the draft’s chapter structure? The Introduction’s outline gives chapter 3 as “Emotions are different Ways to Think”, while the draft’s chapter 3 is “From Pain to Suffering”. Its account of Damasio and its §7-6 should also be checked against the book.
• What evidence would separate Minsky’s Elliot reading from Damasio’s? One candidate: patients with intact anticipatory body responses but impaired problem-type recognition, or the reverse.
• Can the Critic–Selector loop be specified tightly enough to fail? Singh’s later work with Minsky on a commonsense architecture is where that was tried. It is not in this record.

Corrections

• none to a seeded skim (there was no seed)
• Footnote dating. §2-7 (note 14) cites Horner and Gorman’s Digging Dinosaurs as “Harper and Row, 1998”. Open Library’s editions give first publication in 1988 (Workman) and a Perennial Library (Harper & Row) paperback in 1990. It lists no 1998 edition. A reading of the published book should check whether the note was corrected.
• Chapter heading slip. The heading of chapter 1 calls it “Part II”, and the heading of chapter 4 calls it “Part IV”.