|
The Logic of Heuristic Decision Making, Herbert A. Simon, p.1-35
p.1 The task of a comprehensive theory of action is to describe or prescribe the occasions for action, the alternative courses of action (or means of discovering them), and the choice among action alternatives. The task of a comprehensive logic of action is to describe or prescribe the rules that govern reasoning about the occasions for action, the discovery of action alternatives, and the choice of action.
[JLJ - Can anyone claim that the Mona Lisa was constructed from a logic of action? I think not.]
p.2 In a previous paper ["The Logic of Rational Decision"]... The central conclusion reached... is that there is no need for a special "logic of imperatives" or "logic of action"; the basis for the conclusion is that the practitioners in the fields examined clearly get along very well without one... The method, again, will be to point to what sophisticated practitioners actually do: to show that they reason rigorously about action without needing a special logic of action.
p.4-5 It follows that one requirement for a comprehensive logic of decision making is that it should handle the attention-directing decisions that determine when particular kinds of actions should be initiated. In this case, as in the case of choice models, we find that there already exist formal schemes for making attention-directing decisions, phrased in terms of ordinary mathematical concepts, and making use only of the declarative logic that is used in all mathematical reasoning.
p.6 Attention-directing imperatives play a large and important role, however, in all human behavior... It must be possible to make decisions about what to attend to before making decisions about what to do about that which is being attended to.
[JLJ - Yes, our scheme to 'go on' should tell us not only what to attend to, but exactly what we should do after we categorize what we observe.]
p.6 a central fact of the decision-making process... it must be carried out by an information processing system whose computational powers are puny in comparison with the complexity of the environment with which they must cope.
[JLJ - Yes, but a swimmer in the ocean does not need to compute the distance to the planet Mars, the precise salinity of the water or the distance to the sea floor, or any other irrelevant, complex parameter - he must simply perform survival actions (such as treading water and calling for help) to stay alive. In a similar manner, in our daily lives we critically need to act only to 'go on' within our present predicament, and of course the next, and to perform reasonable steps to anticipate what the driving forces of the environment will send our way. In acting to 'go on,' we need only to simplify the complexity of the environment using schemes which are not difficult to construct, or even to execute, even with a 'puny' brain. The goal will be to reduce the complexity of the environment to a simplified model we can 'play' with - we seek an equivalence with a child playing with blocks. We then 'play' with the 'blocks' in our mental model in such a practical way as to understand the driving forces in the environment including latent potentials, and in this way determine effectively how to 'go on' in our current predicament. Such schemes of simplification must occasionally (or even more often) produce errors of interpretation, but with back-up plans, resiliency, and an array of coping skills learned and developed in life, we emerge perhaps bruised, but rarely broken.]
p.7 The second stage in decision making is to devise or discover possible courses of action. This is the activity that in fields like engineering and architecture is called "design"... Design is concerned with devising possible means for ends.
p.8 A problem will be difficult if there are no procedures for generating possible solutions that are guaranteed (or at least likely) to generate the actual solution rather early in the game. But for such a procedure to exist, there must be some kind of structural relation, at least approximate, between the possible solutions as named by the solution-generating process and these same solutions as named in the language of the problem statement.
p.9 In sum, whenever there is a relatively simple relation mapping the names of actions on the names of solutions that can be used to guide the solution generator, finding problem solutions is simple. When there is no such relation, finding problem solutions is difficult.
p.9 An adaptive organism is connected with its environment by two kinds of channels. Afferent channels give it information about the state of the environment; efferent channels cause action on the environment... the mapping of efferents on afferents... the relations can only be discovered by experiment, by acting and observing the consequences of action.
[JLJ - Yes, but this oversimplification ignores the concept of strategic moves such as investments, traps and pressure on weak points that - while unproductive now or in the near term - can lead to payoffs later, often in an undetermined manner.]
p.10 To summarize the argument to this point: a problem of design exists when (1) there is a language for naming actions and a language for naming states of the world, (2) there is a need to find an action that will produce a specified state of the world or a specified change in the state of the world, and (3) there is no non-trivial process for translating changes in the state of the world into their corresponding actions.
p.11 It may be objected that our proposed meaning for "design" excludes almost everything called that in engineering and architecture.
p.16 The conclusion reached in the previous section might be restated thus: In these cases (and they will be the rule rather than the exception) where the achievement of a goal calls for a non-trivial planning action (discovery of alternatives) as well as a performance action (execution of alternatives) determined by the former, the model of the state space must include planning actions as well as performance actions as values of command variables.
[JLJ - Here is where I disagree with Simon (!). The concept of evolutionary experimentation allows us to progress towards a goal without a planning stage - we simply ask ourselves, 'How might I proceed?' and then, 'How much should I care about that?' in sequence, using heuristics to determine not only what catches our attention, but how long that attention should practically be held or sustained. Design then emerges from the process of evolutionary experimentation, not 'planning,' per se. This is the concept of design for evolution, which stumbles upon great solutions, using as a starting point good solutions teeming with latent yet typically hidden adaptive capacity. Here the goal is simply to 'go on', to best future effect, from where we are, with what we have, with whatever we can construct, to whatever it is that we become, and to whatever fate that awaits.]
p.20 a theory of decision making is concerned with processes for selecting aspects of the environment for attention, processes for generating alternatives and processes for choosing among alternatives. In the design of these processes, the conservation of attention and of computing effort is a vital consideration.
[JLJ - Yes, but there ought to be a high-level concept sitting atop all of this. Why do we decide? We 'decide' because we have to 'go on,' and we often plan - not to reach a specific end state (the world is too complex and dynamic to foresee that) - but instead to develop a position with the capacity to change to confront whatever it is we will encounter down the road, known or unknown at the planning stage. Technically, do we truly 'decide,' or do we instead follow a scheme where we have 'pre-decided' how we are going to act - perhaps we merely determine whether we have case a or b or c or d in front of us, then execute our pre-determined action plan based on what we determine to be the case. A computer program represents such a scheme where the programmer has cleverly determined heuristics for branching actions which can be executed without thought, in a way that practically and effectively mimics actual thought. Figure that one out. The concept compares to airline pilots using a pre-flight checklist before takeoff - the list itself (if expertly written and accurately followed) can be blindly executed, and when completed, the results ought to be the same as those obtained with proper thought - the plane is ready for flight.]
p.20 Design uses a wide range of heuristic devices... Much remains to be learned about the nature and effectiveness of these devices.
[JLJ - Yes, I agree, but much of 'design' lies only in an experienced experimenting with the possibilities. Herbert Simon only failed in his computer chess research because he (and his team) were not able to effectively construct the heuristics of expertise - the drivers of action, the semi-intelligent information-cue-rich mechanisms useful for capturing and sustaining the attention of the machine as it performs the equivalent of 'thinking'. This work has been successfully accomplished today. Perhaps if he had spent some time in a chess club, pushing wood as a patzer, he could have completed his work by arriving at the heuristics required for expert action in chess or other complex games of strategy. To resolve the complexity of the position of the pieces on the chessboard requires an evolutionary exploration or experimentation with the possibilities for action and reaction - in such a way (and with such a thoroughness) that allows us to diagnostically (and therefore practically) estimate an adaptive capacity to mobilize the pieces to coerce, that is likely effective in the future positions we will encounter from beyond our planning horizon.]
p.25 A planning action is an action that is supposed to discover a means to an end, programs of action having some likelihood of realizing the requirements of the end to some degree. Sometimes, as Simon points out, the planning action is trivial... In still other cases... there is always the possibility of thinking further about the problem... we must realize that we have a set of alternatives to choose between only as the result of a prior planning action... The "first" planning act is simply done. The question we face at the time of decision is whether it has been done well.
[JLJ - Is not all planning, in effect, an improvised revision of the plans that we already have with us at the moment, in our current predicament, encumbered as we are, and kind of in the 'middle of things', that we are presently executing, in order to 'go on?' We already have plans that were discussed and debated, and we also have the plan B and plan C ideas that we chose not to execute. We also have the new ideas that entered our minds today, this week and this month. Plans 'work' only when the applied leverage we direct against the driving forces of our time ought to work. Change the leverage, or the fulcrum, or the driving forces, and the plans perhaps must now change.]
p.32 most schemes of practical action are workable only because they seek the satisfactory instead of the best.
[JLJ - Perhaps, but one would expect the scheme of practical action to be operable so that in the limit - as time, money, effort, interest increases, that we can approach the best. A practical scheme, by its very nature, arrives at workable answers immediately, good answers quickly, and ought to be operable over time to arrive at better answers, even in the limit approaching the best, but can be abandoned at any time if pressures in another direction require us to fight another fire, so to speak.]
p.40 We shall call the type of goodness at issue here differential goodness. it is assessed by taking into comparative account - not just the situation under consideration, but - the possible alternatives to the situation under consideration.
p.121 To act is to intentionally ("at will") to bring about or to prevent a change in the world (in nature).
[JLJ - Too vague. In a complex world which is changing at every moment, better to say that to act is to adopt a strategic posture towards the present, the foreseeable, and even the unforeseeable.]
p.127 Our logic of action, so far, considers only the presence of one agent in the world. We can, however, generalize the theory to cases when there are two, three, or more agents. Then our theory of action becomes a theory also of the interaction of agents.
p.136 I do not think that the logic of action, as such, provides a useful instrument for dealing with problems in game or decision theory. It is rather the case that methods which have been developed within the theories of games and decisions may be fruitfully applied to the theory of action.
p.140 Von Wright says, correctly I think, that a fully developed logic of action requires a logic of change, where change is construed as a transformation of states.
[JLJ - Well yes, all action takes place in a larger sea of action with driving forces, which in general is a state of change.]
p.
|
