Rationality in Cognitive Science Some Views of Rationality in Cognitive Science Anderson s (1991) Principle of Rationality: The cognitive system optimizes the adaptation of the behavior of the organism. (optimality necessary for rationality) Newell s (1982) Maximum Rationality Hypothesis: If an agent has knowledge that one of its actions will lead to one of its goals, then the agent will select that action. (actions, knowledge, and goals are linked; optimality not assumed) Simon s (1957) Principle of Bounded Rationality: limited resources (knowledge, time) may prevent agent from carrying out actions that are optimal with respect to its goals. 1/27/2005 ZOL 867 1
View from Behavioral Ecology The rational decision is that which would be in the best interest of the animal making the decision, i.e., maximize fitness. Rationality often equated with optimality. Rationality need not involve conscious thought, or even learning (i.e., the rational decision may be genetically programmed by natural selection) Concept of optimality (hence, rationality ) is meaningless without specifying constraints Bluegill sunfish: should fish feed in weeds by shore (low feeding rate, slow growth) or in open water (high feeding rate, rapid growth)? 1/27/2005 ZOL 867 2
Logic of Optimality Analysis Assumptions built into a model Decision variable (may be continuous or categorical) Optimization criterion ( currency related to fitness) Constraints (e.g., time, memory, information) Core of model: functional relationship between decision and performance measure (in appropriate currency), bounded by constraints Test of model: Compare observed decisions with the optimum predicted by model Failure of model not necessarily taken as evidence that animal is suboptimal; may mean we haven t discovered relevant constraints Ultimate goal: what factors have shaped functional design of behavior? Analysis says NOTHING about mechanisms, which might be mediated with simple rules of thumb rather than sophisticated calculations Example: bluegill sunfish 1/27/2005 ZOL 867 3
Tversky & Kahneman (1974) Heuristics: decision-making strategies, based on past experience, that generally give correct answers (but not always) Some examples: Gambler s fallacy: being tricked by runs Hindsight bias: knowledge of outcome affects apparent likelihood of outcome Risk sensitivity Framing effects Tversky & Kahneman interpreted use of heuristics as examples of human irrationality or suboptimal behavior 1/27/2005 ZOL 867 4
Example of Risk Sensitivity You are given a paycheck and are told you have to invest some of it in a retirement plan. The question is which plan you choose You have two plans to choose from: A: Guaranteed gain of $50 B: 0.5 probability of losing $50 and a 0.5 probability of gaining $150 On average, people will end up with a gain of $50 with either option, so (by some definitions of rationality) should be indifferent (on average). However, individual people aren t indifferent, and choice will vary with circumstance Many animals are also risk sensitive 1/27/2005 ZOL 867 5
Example of Framing Effects Imagine that the U.S. is preparing for the outbreak of a new disease, which is expected to kill 60,000 people. Choose between alternative plans that have been developed to combat the disease, considering the following issues: If Plan A is adopted, 20,000 people will be saved If Plan B is adopted, there is a 1/3 probability that 60,000 will be saved and a 2/3 probability that nobody will be saved Now think of it this way: If Plan C is adopted, 40,000 people will die If Plan D is adopted, there is a 1/3 probability that nobody will die and a 2/3 probability that 60,000 will die 1/27/2005 ZOL 867 6
Gigerenzer & Todd Visions of Rationality Demons Bounded Rationality Unbounded Rationality Optimization Under Constraints Satisficing Fast and frugal heuristics G&T: aversion to overzealous mental computation Demons vs. B.R.: what is the difference? Simon: bounds come from structure of mind and structure of environment Fast/frugal heuristics: employ a minimum of time, knowledge, and computation to make adaptive choices in real environments 1/27/2005 ZOL 867 7
Gigerenzer & Todd, cont d Alternative views: G & T vs. Kacelnik & Krebs: there are no optimal strategies in many real-world environments in the first place [so it is presumably misguided to invoke optimality assumption?] G & T vs. Cosmides & Tooby: C & T focus primarily on specially designed strategies rather than possibility that mental tools can be recombined and tested within each other G & T: (real world) vs. Kahneman & Tversky (domain of logic and probability): K & T do not analyze the fit between cognitive mechanisms and their environments. 1/27/2005 ZOL 867 8
Other Questions/Issues Differences in views of rationality/optimality Where do optimality criteria come from (for decision-maker)? What issues arise in deciding optimality criteria for design of artificial systems? T & G: relevance of performance criteria when there are no optimal strategies If organisms aren t optimal, then what use are optimality models? Does use of simple Heuristics/Rules of Thumb entail suboptimal decisionmaking (especially if they work better than more complex decision-rules)? Could one analyze the optimal design of heuristics? Is it meaningful to talk about simple vs. complex heuristics? Is degree of reliance on heuristics a valid way to rank cognitive sophistication of intelligent systems? Can a cognitive science of heuristics produce a general theory or only piecemeal description of decision-making according to context? 1/27/2005 ZOL 867 9