Principles of metareasoning

Russell, Stuart

doi:10.1016/0004-3702(91)90015-c

Cited by 211 publications

(210 citation statements)

References 11 publications

Supporting

Mentioning

208

Contrasting

Order By: Relevance

“…Building on the theory of rational metareasoning developed in artificial intelligence research (Russell & Wefald, 1991), this can be quantified by the value of computation (VOC):…”

Section: Strategy Selection Learning As Metacognitive Reinforcement Lmentioning

confidence: 99%

Strategy selection as rational metareasoning.

Lieder¹,

Griffiths²

2017

Psychological Review

134

110

View full text Add to dashboard Cite

Many contemporary accounts of human reasoning assume that the mind is equipped with multiple heuristics that could be deployed to perform a given task. This raises the question of how the mind determines when to use which heuristic. To answer this question, we developed a rational model of strategy selection, based on the theory of rational metareasoning developed in the artificial intelligence literature. According to our model people learn to efficiently choose the strategy with the best cost-benefit tradeoff by learning a predictive model of each strategy's performance. We found that our model can provide a unifying explanation for classic findings from domains ranging from decisionmaking to arithmetic by capturing the variability of people's strategy choices, their dependence on task and context, and their development over time. Systematic model comparisons supported our theory, and four new experiments confirmed its distinctive predictions. Our findings suggest that people gradually learn to make increasingly more rational use of fallible heuristics. This perspective reconciles the two poles of the debate about human rationality by integrating heuristics and biases with learning and rationality.Keywords: bounded rationality; strategy selection; heuristics; meta-decisionmaking; metacognitive reinforcement learningTo succeed in life we have to solve a wide range of problems that place very different demands on us: sometimes we have to think fast and sometimes we have to think slow (cf. Kahneman, 2011). For instance, avoiding a car accident requires a split-second decision, whereas founding a successful start-up requires investing a lot of time into anticipating the future and weighting potential outcomes appropriately. No single decision mechanism works well across all situations. To meet the wide range of demands posed by different decision problems, it has been proposed that the human brain is equipped with multiple decision systems (Dolan & Dayan, 2013) and decision strategies (Payne, Bettman, & Johnson, 1988). Dual-process theories are a prominent example of this perspective (Evans & Stanovich, 2013;Evans, 2003;Kahneman, 2011). The coexistence of multiple alternative strategies is not specific to decision making. People also appear to possess multiple strategies for inference (Gigerenzer & Selten, 2002), memory (Bjorklund & Douglas, 1997), self-control (Braver, 2012), problem solving (Fum & Del Missier, 2001), and mental arithmetic (Siegler, 1999) to name just a few.The availability of multiple strategies that are applicable to the same problems raises the question how people decide when to use which strategy. The fact that so many different strategies have been observed under different circumstances shows that people's strategy choices are highly variable and contingent on the situation and the task (Beach & Mitchell, 1978;Fum & Del Missier, 2001;Payne, 1982;Payne et al., 1988). Overall, the contingency of people's strategy choices appears to be adaptive. Even though under certain circumstances p...

show abstract

“…Building on the theory of rational metareasoning developed in artificial intelligence research (Russell & Wefald, 1991), this can be quantified by the value of computation (VOC):…”

Section: Strategy Selection Learning As Metacognitive Reinforcement Lmentioning

confidence: 99%

Strategy selection as rational metareasoning.

Lieder¹,

Griffiths²

2017

Psychological Review

134

110

View full text Add to dashboard Cite

show abstract

“…Choosing a cognitive system can be construed as metareasoning: deciding how to deploy cognitive resources in an efficient manner, considering both the expected utility and expected costs rather than using the most accurate strategy regardless of resource usage (Russell and Wefald 1991). This framework has been shown to explain general human strategy selection (Lieder and Griffiths 2017;Payne et al 1988) and has been used to describe how people allocate cognitive resources within a strategy for physical prediction (Hamrick et al 2015).…”

Section: Metareasoning Over Cognitive Systemsmentioning

confidence: 99%

Different Physical Intuitions Exist Between Tasks, Not Domains

2018

View full text Add to dashboard Cite

Does human behavior exploit deep and accurate knowledge about how the world works, or does it rely on shallow and often inaccurate heuristics? This fundamental question is rooted in a classic dichotomy in psychology: human intuitions about even simple scenarios can be poor, yet their behaviors can exceed the capabilities of even the most advanced machines. One domain where such a dichotomy has classically been demonstrated is intuitive physics. Here we demonstrate that this dichotomy is rooted in how physical knowledge is measured: extrapolation of ballistic motion is idiosyncratic and erroneous when people draw the trajectories but consistent with accurate physical inferences under uncertainty when people use the same trajectories to catch a ball or release it to hit a target. Our results suggest that the contrast between rich and calibrated versus poor and inaccurate patterns of physical reasoning exists as a result of using different systems of knowledge across tasks, rather than being driven solely by a universal system of knowledge that is inconsistent across physical principles.

show abstract

“…Some of the agent control approaches involve development of utility models or deliberation policies that determine actions taken by agents in an open environment (Russell & Wefald, 1991;Schut & Wooldridge, 2001). We expect that similar utility models can be developed based on several criteria such 165 as confidence on the knowledge being built and cost of agent tasks, and be used in combination with our existing learning control strategies.…”

Section: Related Workmentioning

confidence: 99%

Goal-Directed Metacontrol for Integrated Procedure Learning

Kim¹,

Myers²,

Gervasio³

et al. 2011

Metareasoning

View full text Add to dashboard Cite

Developing systems that learn how to perform complex tasks presents a significant challenge to the artificial intelligence community. As the knowledge to be learned becomes complex, with diverse procedural constructs and uncertainties to be validated, the system needs to integrate a wide range of learning and reasoning methods with different focuses and strengths. For example, one learning method may be used to generalize from user demonstrations, another to learn by practice and exploration, and another to test hypotheses with experiments. The POIROT system pursues such a multistrategy learning methodology that employs multiple integrated learners and knowledge validation modules to acquire complex process knowledge for a medical logistics domain (Burstein et al., 2008).For a learning system of such complexity, activities of participating agents must be coordinated to ensure that their collective activities produce the desired procedural knowledge. This kind of control is inherently metalevel (Anderson & Oates, 2007; Cox & Raja, Chapter 1) in that it requires the system to reflect on what it is doing and why, to monitor its progress, and to make adjustments to its 135 behaviour when performance falls short of expectations. Without such introspection, effective coordination and prioritization of the base-level learning and reasoning components would not be possible. This type of introspection corresponds to a form of metareasoning centered on "stepping back" from the system to analyze its behavior, as discussed by Perlis (Chapter 2). As such, it contrasts with the majority of work to date on metareasoning, which has focused on the problem of bounded rationality, as described by Zilberstein (Chapter 3).Developing a metalevel reasoner for such a complex, integrated learning system poses several challenges, including• Assessing the progress of learning over time;• Systematically addressing conflicts and failures that arise during learning;• Addressing gaps and shortcomings of the individual and aggregate learning results;• Supporting flexible interactions among agents that pursue different learning strategies.We describe a metalevel framework for coordinating the activities of a community of learners to create an integrated learning system. The metalevel framework is organized around learning goals, which are formulated through introspective reasoning to identify problems and requirements for the ongoing 136 learning process. These learning goals are posted to a shared blackboard to direct the other components in the system. Goals can be either process or knowledge oriented.Process goals define specific tasks to be performed as part of the learning process and are used to coordinate the activities of the various learning and reasoning components. Examples of process goals for task learning include hypothesis creation, hypothesis merging, explanation of observations, and hypothesis validation through experimentation.Knowledge goals provide the means for a component to convey the need for additional information to f...

show abstract

Principles of metareasoning

Cited by 211 publications

References 11 publications

Strategy selection as rational metareasoning.

Strategy selection as rational metareasoning.

Different Physical Intuitions Exist Between Tasks, Not Domains

Goal-Directed Metacontrol for Integrated Procedure Learning

Contact Info

Product

Resources

About