HOME  |   ENGINEERING  |   RESEARCH  |   CONTACT
Society of Mind

MARVIN MINSKY
1985

The Society of Mind is a scheme in which each mind is made of many smaller processes called Agents. Each mental agent by itself can only do some simple thing that needs no mind or thought at all. True intelligence comes from joining these agents in certain special ways to create societies.

Minds are simply what brains do.

  • To understand any large and complex thing we need to know:
    • How each separate part works
    • How each part interacts with those to which it is connected
    • How all these interactions combine to accomplish what that system does – as seen from the outside
  • Reductionists are those people who prefer to build on old ideas. Novelists are the ones who like to champion new hypotheses.
  • Psychology need not reject the laws of physics, but use additional kinds of theories and principles that operate at higher levels of organization. It must not conflict with our knowledge of “lower-level” agents but build on them.
  • In physics we’re used to explanations in terms of a dozen basic principles. For psychology, our explanations will have to combine hundreds of smaller theories.
  • “Subjective” reactions are based on how things interact. The difference is that here we are not concerned with objects in the world outside, but with processes inside our brains.
  • It is mainly when our other systems start to fail that we engage the special agencies involved with what we call “consciousness.”
  • It is assumed that conflicts between agents tend to migrate upward to higher levels.
  • The Principle of Noncompromise: The longer an internal conflict persists among an agent’s subordinates, the weaker becomes that agent’s status among its own competitors. If such internal problems aren’t settled soon, other agents will take control and the agents formerly involved will be dismissed.
  • A hierarchical society is like a tree in which the agent at each branch is exclusively responsible for the agents on the twigs that branch from it. But hierarchies do not always work.
  • Pain is powerful because it is hard to think of anything else. Pain simplifies your point of view. It can disrupt our concerns for long-term goals thus forcing us to focus on immediate problems. It is the same for pleasure.
  • In order to appear opposed, two things must serve related goals – or otherwise engage the self-same agencies.
  • A person’s self-image is a set of beliefs about what we are. Our self-ideals are our ideas about what we ought to be. We find these hard to express because they’re inaccessible to consciousness.
  • Sometimes we feel single minded, other times that we are in two minds.
  • People ask if machines can have souls. Can souls learn?
  • To understand what we call the Self, we must first see what Selves are for: one function of the Self is to keep us from changing too rapidly. Each person must make long-range plans in order to balance single-purposeness against attempts to do everything at once.
  • Many of the schemes we use for self-control are the same as those we learn to use for influencing other people. We make ourselves exploit our own fears and desires, offering ourselves rewards, or threatening the loss of what we love. These are short range tricks.
  • Character is shaped by our slowest-changing hidden agencies, that are systems concerned not merely with the things we want, but what we want ourselves to be – that is, the ideals we set for ourselves.
  • In childhood, our agencies acquire various types of goals. The older agencies can influence how our later ones will behave.
  • Without enduring self-ideals, our lives would lack coherence. A working society must evolve mechanisms that stabilize ideals.
  • Circular Causality: There need be no first cause as both goals may be on equal ground, then a loop of circular causality ensues, in which each goal gains support from the other until their combined urge becomes irresistible.
  • To straighten out a path that contains a loop may require ignoring important interactions and dependencies that run in other directions.
  • All human cultures evolve institutions of law, religion, and philosophy, and these institutions both adopt specific answers to circular questions and establish authority-schemes to indoctrinate people with those beliefs. They may substitute dogma for reason and truth, but in spares populations from wasting time in fruitless reason loops.
  • Circular thinking can lead to growth when it results in deeper more powerful ideas.
  • The idea of a single, central Self doesn’t explain anything. This is because a thing with no parts provides nothing that we can use as pieces of explanation.
  • Reasons why we think we are a single self include:
    • The physical world: it appears as though we are one person
    • Personal Privacy: we assume our thoughts are our own, it gives us also a sense of responsibility.
    • Mental Activity: we find it hard to think two things at once.
  • Our mental processes often seem to flow in “streams of consciousness” perhaps so that we can simplify what’s happening and feel in control.
  • Some reasons why we make choices that have no reason at all include:
    • Recognizability: familiar styles make it easier for us to recognise things
    • Uniformity: we protect ourselves from too much distraction
    • Predictability: rules can help make complex things work more efficiently
  • Fredkin’s Paradox: The equally attractive two alternatives seem, the harder it can be to choose between them – no matter that, to the same degree, the choice can only matter less.
  • Some simple reasons why we assign Traits to people:
    • Selectivity: our images of other minds are often falsely clear. We tend to think of another personality in terms of what we can describe.
    • Style: to escape the effort of making decisions we consider unimportant, we tend to develop policies that become so systematic that they can be discerned from the outside and characterized as personal traits.
    • Predictability: we try to conform to the expectations of our friends to maintain trust. Then, to the extent that we frame our images of our associates in terms of traits, we find ourselves teaching ourselves to behave in accord with those same descriptions.
    • Self-Reliance: imagined traits can make themselves actual. To make our own plans we need to be able to predict what we ourselves are likely to do – and that will become easier the more we simplify ourselves.
  • We find it almost impossible to separate the appearances of things from what they’ve come to mean to us in previous times.
  • In every normal person’s mind there seem to be some processes that we call consciousness. Conscious thoughts reveal to us some little of what gives rise to them.
  • Our conscious thoughts use signal-signs to steer the engines in our minds, controlling countless processes of which we’re never much aware.
  • Whenever a new thing’s internal workings are too strange or complicated to deal with directly, we represent whatever parts of it we can in terms of more familiar signs. The use of signals, symbols, words, and names let our mind transform the strange into the commonplace.
  • We tend to think of knowledge as good in itself, but knowledge is useful only when we can exploit it to help us reach our goals.
  • No supervisor can know everything that all its agents do. There’s simply never time enough.
  • We’ll conjecture that your brain contains a host of agents called K-lines, which you can use to make records of what some of your brain-agents are doing at a certain moment. Later when you activate the same K-lines, this restores those agents to their previous states. This makes you remember part of your previous mental state, by making those parts of your mind do just what they did before. Memories will always be incomplete – nothing could have the capacity to record every detail of its own state otherwise it would be larger than itself.
  • It is simply impossible for any agent to know for certain what another agent is doing at precisely the same time. Each agency must have at least a slightly different sense both of what happened in the past – and what is happening how.
  • Our memories are only indirectly linked to physical time. We have no absolute sense of when a memorable event actually happened.
  • Nothing can have meaning by itself, but only in relation to whatever other meanings we already know.
  • The secret of what anything means to us depends on how we’ve connected it to all the other things we know. That’s why it’s almost always wrong to seek the “real meaning” of anything. A thing with just one meaning has scarcely any meaning at all.
  • There is a sinister way to make the world seem orderly, in which the mind has merely found a way to simplify itself. This what be what happens in some of those experiences that leave a person with a sense of revelation. One can acquire certainty by amputating inquiry.
  • To get a good idea, one must engage huge organisations of submachines that do a vast variety of jobs.
  • The smaller two languages are, the harder it will be to translate between them because there are too few meanings. If two agents have nothing in common, no translation is possible.
  • If we could deliberately seize control of our pleasure systems, we could reproduce the pleasure of success without the need for any actual accomplishment. And that would be the end of everything.
  • Paradoxically, it is smart to realize that one is confused, as opposed to being confused without knowing it.
  • Our minds contain processes that enable us to solve problems we consider difficult. Intelligence is our name for whichever of those processes we don’t yet understand.
  • To be considered an expert, one needs a large amount of knowledge of only a relatively few varieties. In contrast, an ordinary person’s common sense involves a much larger variety of different types of knowledge – and this requires more complicated management systems.
  • Each type of knowledge needs some form of representation. One that investment is made, it is relatively easy to add additional expertise.
  • Puzzle Principle: We can program a computer to solve any problem by trial and error, without knowing how to solve it in advance, provided only that we have a way to recognize when the problem is solved.
  • The Generate and Test method consists of two parts. The first process produces every possible arrangement, and the second part examines each arrangement to see whether the problem has been solved.
  • The Progress Principle: any process of exhaustive search can be greatly reduced if we possess some way to detect when progress has been made.
  • Goals and Subgoals is the most powerful way we know for discovering how to solve a hard problem – find a method that splits it into several simpler ones, each of which can be solved separately.
  • Using Knowledge is the most efficient way to solve a problem – we can avoid search entirely.
  • In order to solve any hard problem, we must use various kinds of memories. At each moment, we must keep track of what we’ve just done. Also, we must somehow maintain our goals. Finally, once our problem is solved, we need access to records of how it was done, for use when similar problems arise in the future.
  • A goal-driven system does not seem to react directly to the stimuli or situations encounters. Instead, it treats the things it finds as objects to exploit, avoid, or ignore, as though it were concerned with something else that doesn’t yet exist.
  • A process inside a machine that could give the impression of having a goal are general problem solvers or Difference Engines:
    • They must contain a description of a desired situation.
    • It must have subagents that are aroused by various differences between desired situation and the actual situation.
    • Each subagent must act in a way that tends to diminish the difference that aroused it.
  • Genius: in order to accumulate outstanding qualities, one needs unusually effective ways to learn. But also, one has to manage what one learns. Those masters must have some special knacks of higher-order expertise, which helps them organise and apply the things they learn.
  • K-Lines: A theory of memory: Whenever you “get a good idea” or solve a problem, you activate a K-line to represent it. A K-line is a wirelike structure that attaches itself to whenever mental agents are activate when you solve a problem or have a good idea. When you activate that K-line later, the agents attached to it are aroused, putting you into a mental state much like the one you were in when you solved the problem.
  • We can not always judge the complexity of our mental states by how easily we can express them in words.
  • We make new ideas by merging parts of older ones.
  • A Total State of Mind is a list that specifies which agents are active and which are quiet at a certain moment. A Partial State of Mind merely specifies that certain agents are active but does not say which other agents are quiet.
  • Conflicts in state of mind can occur when two K-lines activate agents in the same division for two different partial mental states.
  • Level Band Theory: we learn by attaching agents to K-lines, but we don’t attach them all with equal firmness. Instead, we make strong connections at a certain level of detail, but we make weaker connections at higher and lower levels.
  • Weakly activates memories can be called Assumptions by Default. Some of our most valuable kinds of commonsense knowledge is embodies by assumptions by default.
  • Fringing Effects serve to make our memories more relevant to our present purposes. Beyond the Lower Band level of detail, it becomes difficult to match previous situations to new situations. Memories that arouse too high level agents are called Upper Band effects that tend to provide us with goals that are not appropriate to the present situations. The lower levels represent “objective” details of reality; the upper levels represent our “subjective” concerns with goals and intentions.
  • If each K-line can connect to other K-lines, which, in turn, connect to others, then K-lines can form societies. But when making a new K-line memory, do not connect it to all the K-lines active at the time but only those that are active within a certain level-band.
  • The surer you are that you like what you are doing, the more completely your other ambitions are being suppressed. If unconstrained, it can lead to artificial clarity: it does not reflect what liking is but only shows what liking does.
  • To choose between alternatives, the highest levels of the mind demand the simplest summaries.
  • Papert’s Principle: some of the most crucial steps in mental growth are based not simply on acquiring new skills, but on acquiring new administrative ways to use what one already knows.
  • It is fairly easy to resolve conflicts by switching among alternatives. It is much harder to develop mechanisms that can use cooperation and compromise.
  • We never really make direct contact with the outside world. Instead, we work with models of the world that we build inside our brains.
  • Other things being equal, the apparent similarity of two stimuli will depend on the extent to which they lead to similar mental consequences.
  • The nerve pathways that preserve the physical nearness relations of our skin-sensors can make it easy for inner agencies to discover corresponding nearnesses about the outer world if space.
  • All behaviour cannot be classified into either built-in or learned. Some appear destined to end up with a certain behaviour. We call this predestined learning.
  • Learning is more economical. It would require an enormous store of genetic information to force each separate nerve cell to make precisely the right connections, whereas it would require much less information to specify the construction of a learning machine designed to unscramble what irregularities result from a less constrained design.
  • Dividing things in two (ie concepts and words into opposites or dumbbell theory) is a good way to start, but one should always try to find a third alternative. If one cannot, one should suspect that there may not be two ideas at all, but only one, together with some form of opposite. Whenever any simple idea appears to explain so many things, we must suspect a trick.
  • Learning is making useful changes in the workings of our minds. Some different ways to learn include:
    • Uniframing combining several descriptions into one. Most differences are redundant.
    • Accumulating: collecting incompatible descriptions.
    • Reformulating: modifying a description’s character.
    • Trans-framing: bridging between structures and functions or actions.
  • Uniframers disregard discrepancies in favour of imagined regularities. They tend to be perfectionists but also tend to think in terms of stereotypes. This sometimes leads to recklessness because they have to reject some evidence in order to make their uniframes work.
  • Accumulators are less extreme. They keep collecting evidence and hence are much less prone to make mistakes. But then they’re also slower to make discoveries.
  • Our different worlds of ends and means don’t usually match up very well. So when we find a useful, compact uniframe in one such world, it often corresponds to an accumulation in our other worlds.
  • The Exception Principle: it rarely pays to tamper with a rule that nearly always works. It is better just to complement it with an accumulation of specific exceptions.
  • To know the cause of a phenomenon is to know, at least in principle, how to change or control some aspects of some entities without affecting all the rest.
  • What people call meanings do not usually correspond to particular and definite structures, but to connections among and cross fragments of the great interlocking networks of connections and constraints among our agencies. Because these networks are constantly growing and changing, meanings are rarely sharp, and we cannot always expect to be able to define them in terms of compact sequences of words.
  • It can require more skill to produce what we regard as a simple copy or imitation than to produce what we consider to be an abstract representation.
  • Our systematic cross-realm translations are the roots of fruitful metaphors; they enable us to understand things we’ve never seen before. When something seems entirely new in one of our description-worlds, it may turn out that when translated to some other world it resembles something we already know.
  • The Investment Principle: our oldest ideas have unfair advantages over those that come later. The earlier we learn a skill, the more methods we can acquire for using it. Each new idea must then compete against the larger mass of skills the ideas have accumulated.
  • In order for a mind to think, it has to juggle fragments of its mental states.
  • To outgrow infancy, you have to sacrifice your memories because they’re written in an ancient script that your later selves can no longer read.
  • Memories are processes that make some of our agents act in much the same ways they did at various times in the past.
  • The Immanence Illusion: whenever you can answer a question without a noticeable delay, it seems as though that answer was already active in your mind.
  • A brain has no single, common memory system. Instead, each part of the brain has several types of memory-agencies that work in somewhat different ways, to suit particular purposes.
  • The Recursion Principle: when a problem splits into smaller parts, then unless one can apply the mind’s full power to each sub-job, one’s intellect will get dispersed and leave less cleverness for each new task.
  • No matter how neutral and rational a goal may seem, it will eventually conflict with other goals if it persists for long enough. No long-term project can be carried out without some defense against competing interests, and this likely to produce what we call emotional reactions to the conflicts that come about among our most insistent goals.
  • The question is not whether intelligent machines can have any emotions, but whether machines can be intelligent without any emotions.
  • A fantasy need not reproduce the fine details of an actual scene. It need only reproduce that scene’s effect on other agencies.
  • A Simulus: a reproduction of only the higher-level effects of a stimulus.
  • One’s present personality cannot share many of the thoughts of all one’s older personalities – and yet it has some sense that they exist. This is one reason why we feel that we possess an inner Self – a sort of ever-present person-friend, inside the mind, whom we can always ask for help.
  • In the case of:
    • Ordinary forms of failure or success signals, the learner modifies the methods used to reach the goal.
    • Fear-provoking disturbances, the learner may modify the description of the situation itself.
    • Attachment-related failure or reward signals, the learner modifies which goals are considered worthy of pursuit.
  • Functional Autonomy in the course of pursuing any sufficiently complicated problem, the subgoals that engage our attentions can become both increasingly more ambitious and increasingly detached from the original problem.
  • Virtually any problem will be easier to solve the more one learns about the context world in which that problem occurs. No matter what one’s problem is, provided that it’s hard enough, one always gains from learning better ways to learn.
  • Stages of development through plateaus maybe necessary to test changes for harmful side effects.
  • A complex system must be grown in a sequence of separate steps.
  • Many people dislike the thought of being dominated from within by the image of a parent’s wish. Yet, in exchange, that slavery is just what makes us relatively free from being forced to obey so many other kinds of unlearned, built-in instinct-goals.
  • The less we base our conclusions on, the fewer the possibilities can exist for weakness in our arguments! This strategy serves mathematics well – but it doesn’t help us much in dealing with uncertainties. We cannot afford to stake our lives on chains that fall apart so easily.
  • Once we find a way to solve a certain problem, logical analysis can help us find the most essential steps.
  • As scientists, we like to make our theories as delicate and fragile as possible. If the slightest thing goes wrong, everything will collapse at once. If a process fails, a new discovery is made! But that isn’t good for psychology.
  • If we are to understand how language works, we must discard the usual view that words denote or designate; instead, their function is control: each word makes various agents change what various other agents do.
  • Two kinds of agents that contribute to the power of words:
    • Polynemes are involved with long term memories. It sends the same, simple signal to many different agencies: each of those agencies must learn, for itself, what to do when it receives that signal.
    • Isonome: controls a short-term memory in each of many agencies. For example, using the word “it” has no particular significance, but refers to a certain recent memory.
  • To understand a polyneme, each agency must learn its own specific and appropriate response. Each agency must have its private dictionary or memory bank to tell it how to respond to every polyneme.
  • Making variations on a theme is the crux of creativity. It is a consequence of the divisibility of concepts into already significant subconceptual elements.
  • Thoughts themselves are ambiguous.
  • Conceptual Dependencies: a way to represent many situations in terms of a few kinds of relations:
    • P-Trans represent physical motion from one place to another
    • M-Trans represent mental transportation.
    • A-Trans like ownership is transferred, not the object itself.
  • Polynemes are permanent K-lines. They are long term memories. Pronomes are temporary K-lines. They are short term memories.
  • An Isonome has a similar, built-in effect on each of its recipients. It thus applies the same idea to many different things at once. A polyneme has different, learned effects on each of its recipients. It thus connects the same thing to many different ideas. Isonomes control how memories are used.
  • No conception or idea could have much use unless it could remain unchanged and stay in some kind of mental place for long enough for us to find it when we need it. Nor could we ever achieve a goal unless it could persist for long enough. In short, no mind can work without some stable states or memories.
  • The Duplication Problem the states of two different agencies cannot be compared unless those agencies themselves are virtually identical.
  • Any agent sensitive to changes in time can also be used to detect differences.
  • Default assumptions fill our frames to represent what’s typical. We adapt our frames to each particular experience.
  • Our vision systems are born equipped with some sort of “locking-in” machinery that at every moment permits each part to be assigned to one and only one whole at the next level.
  • Our sense of constant contact with the world is not a genuine experience; instead, it is a form of immanence illusion. We have the sense of actuality when every question asked of our visual system is answered so swiftly that it seems as though those answers were already there. That’s what Frame-Arrays provide us with: once any frame fills its terminals, the terminals of the other frames in its array are also filled. When every change of view engages frames whose terminals are already filled, albeit only by default, then sight seems instantaneous.
  • A word-string seems “grammatical” if all of its words fit quickly and easily into frames that connect suitably to one another.
  • A word can only serve to indicate that someone else may have a valuable idea – that is, some useful structure to be built inside the mind. Each new word only plants a seed: to make it grow, a listener’s mind must find a way to build inside itself some structure that appears to work like the one in the mind from it was learned.
  • Suppessor-agents wait until you get a certain “bad idea”. Then they prevent your taking the corresponding action, and make you wait until you think of some alternative. Censor-agents need not wait; instead they intercept the states of mind that usually precede that thought.
  • When we learn in a serious context, the result is to change connections among ordinary agents. But when we learn in a humorous context, the principal result is to change the connections that involve our censors and suppressors.
  • Positive memory-agents must learn which mental states are desirable. Negative memory-agents must learn which mental states are undesirable. Suppressors merely need to learn which mental states are desirable. Censors must remember and learn which mentals states were undesirable.
  • The function of laughing is to disrupt another person’s reasoning. Laughter focuses attention on the present state of mind. By preventing you from taking seriously your present thought, and thus proceeding to develop it, laughter gives you time to build a censor against that state of mind.
  • Machines and brains require ordinary energy to do their jobs – and need no other, mental forms of energy. Causality is quite enough to keep them working toward their goal.
  • We turn to using quantities when we can’t compare the qualities of things.
  • Whenever we turn to measurements, we forfeit some uses of intellect. Currencies and magnitudes help us make comparisons only by connecting the differences among what they purport to represent.
  • The so-called problem of body and mind does not hold any mystery: minds are simply what brains do. Whenever we speak about a mind, we’re speaking of the processes that carry our brains from state to state.
  • There is not the slightest reason to doubt that brains are anything other than machines with enormous numbers of parts that work in perfect accord with physical laws.
  • Modifying or replacing the physical parts of a brain will not affect the mind it embodies, unless this alters the successions of states in that brain.

More about the Society of Minds can be found at Wikipedia



© 2020 Cedric Joyce