In a
previous post, I suggested that there is something wrong with the standard (“cogitative”) conception of decision-making in psychology. In this post, I would like to outline an alternative conception, what we might call the “distributed conception”.
A close look at robotics suggests that decision-making should not be construed as a deliberative process.
Deliberative control (Mataric, 1997) or sense-model-plan-act (SMPA) architectures have been unsuccessful in controlling autonomous robots (Brooks, 1999; Pfeifer & Scheier, 1999). In these architectures, (e.g. Nilsson, 1984), “what to do?” was represented as a logical problem. Sensors or cameras represented the perceptible environment while internal processors converted sensory inputs in first-order predicate calculus. From this explicit model of its environment, the robot’s central planner transformed a symbolic description of the world into a sequence of actions (see Hu & Brady, 1996, for a survey). Decision-making was taken in charge by an expert system or a similar device. Thus the flow of information is one-way only: sensors → model → planner → effectors.
SMPA architectures could be effective, but only in environment carefully designed for the robot. The colors, lightning and objects disposition were optimally configured for simplifying perception and movement. Brooks describes how the rooms where autonomous robots evolve were optimally configured:
The walls were of a uniform color and carefully lighted, with dark rubber baseboards, making clear boundaries with the lighter colored floor. (…) The blocks and wedges were painted different colors on different planar surfaces. (….) Blocks and wedges were relatively rare in the environment, eliminating problems due to partial obscurations (Brooks, 1999, p. 62)
Thus the cogitative conception of decision-making, and its SMPA implementations, had to be abandoned. If it did not work for mobile robots, it is justified to argue that for cognitive agents in general the cogitative conception also has to be abandoned. Agents do not make decisions simply by central planning and explicit models manipulations, but by coordinating multiple sensorimotor mechanisms. In order to design robots able to imitate people, for instance, roboticists build systems that control their behavior through multiple partial models. Mataric (2002) robots, for instance, learn to imitate by coordinating the following modules:
- a selective attentional mechanisms that extract salient visual information (other agent's face, for instance)
- a sensorimotor mapping system that transforms visual input in motor program
- a repertoire of motor primitives
- a classification-based learning mechanism that learns from visuo-motor mappings
Neuroeconomics also suggests another--similar--avenue: there is no brain area, circuit or mechanisms specialized in decision-making, but rather a collection of neural modules. Certain area specializes in visual-saccadic decision-making (Platt & Glimcher, 1999). Social neuroeconomics indicates that decision in experimental games are mainly affective computations: choice behavior in these games is reliabely correlated to neural activations of social emotions such as the ‘warm glow’ of cooperation (Rilling et al., 2002), the ‘sweet taste’ of revenge (de Quervain et al., 2004) or the ‘moral disgust’ of unfairness (Sanfey et al., 2003). Subjects without affective experiences or affective anticipations are unable to make rational decision, as Damasio and his colleagues discovered. Damasio found that subjects with lesions in the ventromedial prefrontal cortex (vmPFC, a brain area above the eye sockets) had huge problems in coping with everyday tasks (Damasio, 1994). They were unable to plan meetings; they lose their money, family or social status. They were, however, completely functional in reasoning or problem-solving task. Moreover, Damasio and its collaborators found that these subjects had lower affective reactions. They did not felt sad for their situation, even if they perfectly understood what “sad” means, and seemed unable to learn from bad experiences. The researchers concluded that these subjects were unable to use emotions to aid in decision-making, a hypothesis that also implies that in normal subjects, emotions do aid in decision-making.
Consequently, the “Distributed Conception of Decision-making” suggest that making is:
Sensorimotor: the mechanisms for decision-making are not only and not necessarily intellectual, high-level and explicit. Decision-making is the whole organism’s sensorimotor control.
Situated: a decision is not a step-by-step internal computation, but also a continuous and dynamic adjustment between the agent and its environment that develop in the whole lifespan. Decision-making is always physically and (most of the time) socially situated: ecological situatedness is both a constraint on, and a set of informational resources that helps agent to cope with, decision-making.
Psychology should do more than documenting our inability to follow Bayesian reasoning in paper-and-pen experiment, but study our sensorimotor situated control capacities. Decision-making should not be a secondary topics for psychology but, following Gintis “the central organizing principle of psychology” (Gintis, 2007, p. 1). Decision-making is more than an activity we consciously engage in occasionally : it is rather the very condition of existence (as Herrnstein, said “all behaviour is choice” (Herrnstein, 1961).
Therefore, deciding should not be studied like a separate topic (e.g. perception), or an occasional activity (e.g. chess-playing) or a high-level competence (e.g. logical inference), but with robotic control. A complete, explicit model of the environment, manipulated by a central planner, is not useful for robots. New Robotics (Brooks, 1999) revealed that effective and efficient decision-making is achieved through multiple partial models updated in real-time. There is no need to integrate models in a unified representations or a common code: distributed architectures, were many processes runs in parallel, achieve better results. As Barsalou
et al. (2007) argue,
Cognition is coordinated non-cognition; similarly, decision-making is coordinated non-decision-making.
If decision-making is the central organizing principle of psychology, all the branches of psychology could be understood as research fields that investigate different aspects of decision-making. Abnormal psychology explains how deficient mechanisms impair decision-making. Behavioral psychology focuses on choice behavior and behavioral regularities. Cognitive psychology describes the mechanisms of valuation, goal representation, preferences and how they contribute to decision-making. Comparative psychology analyzes the variations in neural, behavioral and cognitive processes among different clades. Developmental psychology establishes the evolution of decision-making mechanisms in the lifespan. Neuropsychology identify the neural substrates of these mechanisms. Personality psychology explains interindividual variations in decision-making, our various decision-making “profiles”. Social psychology can shed lights on social decision-making, that is, either collective decision-making (when groups or institutions make decisions) or individual decision-making in social context. Finally, we could also add environmental psychology (how agents use their environment to simplify their decisions) and evolutionary psychology (how decision-making mechanisms are – or are not – adaptations).
Related posts:References
