## Estimating human priors on causal strength

### BibTeX

@MISC{Yeung_estimatinghuman,

author = {Saiwing Yeung and Thomas L. Griffiths (tom},

title = {Estimating human priors on causal strength},

year = {}

}

### OpenURL

### Abstract

Bayesian models of human causal induction rely on assumptions about people’s priors that have not been extensively tested. We empirically estimated human priors on the strength of causal relationships using iterated learning, an experimental method where people make inferences from data generated based on their own responses in previous trials. This method produced a prior on causal strength that was quite different from priors previously proposed in the literature on causal induction. The predictions of Bayesian models using different priors were then compared against human judgments of strength of causal relationships. The empirical priors estimated via iterated learning resulted in the best predictions.

### Citations

1120 |
Causality: Models, Reasoning, and Inference
- Pearl
- 2000
(Show Context)
Citation Context ...ly showed how this approach could be extended to infer causal strength. Causal graphical models are probabilistic models in which a graph is used to denote the causal relationships between variables (=-=Pearl, 2000-=-; Spirtes, Glymour, & Scheines, 2001). In the graph, nodes represent variables and edges represent the causal connection between those variables. Following Griffiths and Tenenbaum (2005) and Lu et al.... |

493 |
Causation, prediction and search
- Spirtes, Glymour, et al.
- 1993
(Show Context)
Citation Context ... this approach could be extended to infer causal strength. Causal graphical models are probabilistic models in which a graph is used to denote the causal relationships between variables (Pearl, 2000; =-=Spirtes, Glymour, & Scheines, 2001-=-). In the graph, nodes represent variables and edges represent the causal connection between those variables. Following Griffiths and Tenenbaum (2005) and Lu et al. (2008), we focus on causal systems ... |

239 |
An Enquiry Concerning Human Understanding
- Hume
- 1999
(Show Context)
Citation Context ... induction The British philosopher David Hume pointed out that people are not “able to comprehend any force or power by which the cause operates, or any connexion between it and its supposed effect” (=-=Hume, 1739-=-/2004, p. 47), suggesting that causal relationships need to be inferred from the observed contingencies of cause and effect. A number of models have been proposed to account for how this inference mig... |

227 |
Modern Applied Statistics with S
- Venables, Ripley
- 2002
(Show Context)
Citation Context ...ing this data. Figure 2 shows the density of the empirical priors based on the responses from the final iteration of all chains, smoothed via kernel density estimation with a bivariate normal kernel (=-=Venables & Ripley, 2002-=-). The generative SS priors gives high probability to regions where only one of w0 or w1 is high, while the empirical priors generally prefers w1 to be high, with the distribution on w0 being more uni... |

224 |
From covariation to causation: A causal power theory
- Cheng
- 1997
(Show Context)
Citation Context ...psychological models of human causal induction have focused on various schemes for comparing the probability of an effect occurring in the presence and absence of a cause (e.g., Ward & Jenkins, 1965; =-=Cheng, 1997-=-). However, recent work has explored connections between ideas from Bayesian statistics and human cognition, using causal graphical models to precisely define the problem of causal induction (Griffith... |

104 | Spontaneous evolution of linguistic structure: An iterated learning model of the emergence of regularity and irregularity
- Kirby
- 2001
(Show Context)
Citation Context ...terated learning (Kalish, Griffiths, & Lewandowsky, 2007; Griffiths, Christian, & Kalish, 2008). Iterated learning was originally proposed as a simple model of the cultural transmission of languages (=-=Kirby, 2001-=-). In this case, we imagine a chain of agents, where each agent observes data generated by the previous agent (such as a set of utterances), forms a hypothesis about the process that generated those d... |

90 | Structure and strength in causal induction
- Griffiths, Tenenbaum
(Show Context)
Citation Context ...ng, 1997). However, recent work has explored connections between ideas from Bayesian statistics and human cognition, using causal graphical models to precisely define the problem of causal induction (=-=Griffiths & Tenenbaum, 2005-=-; Lu, Yuille, Liljeholm, Cheng, & Holyoak, 2008) and to formalize the effects of prior knowledge (Griffiths & Tenenbaum, 2009). A key part of these Bayesian models is to precisely specify the prior kn... |

35 |
God does not play dice: Causal determinism and preschoolers’ causal inferences
- Schulz, Sommerville
- 2006
(Show Context)
Citation Context ...s characteristic of the empirical prior points to some degree of preference for deterministic causal systems, consistent with previous research on human causal induction (Griffiths & Tenenbaum, 2009; =-=Schulz & Sommerville, 2006-=-). Comparing models to human judgments We can now compare Bayesian models based on the three different priors – uniform (Griffiths & Tenenbaum, 2005), sparse and strong (Lu et al., 2008), and the empi... |

33 | Theory-based causal induction
- Griffiths, Tenenbaum
- 2009
(Show Context)
Citation Context ...usal graphical models to precisely define the problem of causal induction (Griffiths & Tenenbaum, 2005; Lu, Yuille, Liljeholm, Cheng, & Holyoak, 2008) and to formalize the effects of prior knowledge (=-=Griffiths & Tenenbaum, 2009-=-). A key part of these Bayesian models is to precisely specify the prior knowledge that people have about the strength of causal relationships. In previous models of human causal induction, priors on ... |

31 |
Is causal induction based on causal power? Critique of Cheng
- Lober, Shanks
- 2000
(Show Context)
Citation Context ...1−P(e +|c− . The causal power model provided bet) ter fit than ∆P for some human data. However, there have been debates about the lack of fit to human data for both models (see Buehner & Cheng, 1997; =-=Lober & Shanks, 2000-=-). 1709Models based on Bayesian statistics Griffiths and Tenenbaum (2005) proposed a Bayesian framework for studying causal induction. This framework uses causal graphical models to distinguish betwe... |

25 |
Causal Induction: The power PC theory versus the Rescorla-Wagner theory
- Buehner, Cheng
- 1997
(Show Context)
Citation Context ...xpressed ∆P as power = 1−P(e +|c− . The causal power model provided bet) ter fit than ∆P for some human data. However, there have been debates about the lack of fit to human data for both models (see =-=Buehner & Cheng, 1997-=-; Lober & Shanks, 2000). 1709Models based on Bayesian statistics Griffiths and Tenenbaum (2005) proposed a Bayesian framework for studying causal induction. This framework uses causal graphical model... |

25 | Language evolution by iterated learning with Bayesian agents
- Griffiths, Kalish
- 2007
(Show Context)
Citation Context ...e agents select hypotheses using Bayesian inference, then as the chain gets longer the probability that an agent selects a particular hypothesis converges to the prior probability of that hypothesis (=-=Griffiths & Kalish, 2007-=-). Simulating this process of iterated learning in the laboratory thus provides a way to estimate people’s priors (Kalish et al., 2007). In fact, there is no need for data to be passed between people ... |

20 | Bayesian generic priors for causal learning
- Lu, Yuille, et al.
- 2008
(Show Context)
Citation Context ...rk has explored connections between ideas from Bayesian statistics and human cognition, using causal graphical models to precisely define the problem of causal induction (Griffiths & Tenenbaum, 2005; =-=Lu, Yuille, Liljeholm, Cheng, & Holyoak, 2008-=-) and to formalize the effects of prior knowledge (Griffiths & Tenenbaum, 2009). A key part of these Bayesian models is to precisely specify the prior knowledge that people have about the strength of ... |

16 | Iterated learning: Intergenerational knowledge transmission reveals inductive biases
- Kalish, Griffiths, et al.
- 2007
(Show Context)
Citation Context ...ns about the abstract properties of the causal system (Lu et al., 2008). In this paper, we present a new approach to estimating human priors on causal strength, using the method of iterated learning (=-=Kalish, Griffiths, & Lewandowsky, 2007-=-; Griffiths, Christian, & Kalish, 2008). Iterated learning was originally proposed as a simple model of the cultural transmission of languages (Kirby, 2001). In this case, we imagine a chain of agents... |

14 |
Simplicity and probability in causal explanation
- Lombrozo
- 2007
(Show Context)
Citation Context ...generic priors — a theoretically driven prior that makes systematic assumptions about the abstract properties of a system. They argued that people have preference for causal models with fewer causes (=-=Lombrozo, 2007-=-), and for causes that minimize complex interactions (Novick & Cheng, 2004). Based on these arguments, they specified the sparse and strong (SS) prior as P(w0,w1) ∝ e −α(1+w0−w1) + e −α(1−w0+w1) in th... |

11 |
Using category structures to test iterated learning as a method for identifying inductive biases
- Griffiths, Christian, et al.
- 2008
(Show Context)
Citation Context ...causal system (Lu et al., 2008). In this paper, we present a new approach to estimating human priors on causal strength, using the method of iterated learning (Kalish, Griffiths, & Lewandowsky, 2007; =-=Griffiths, Christian, & Kalish, 2008-=-). Iterated learning was originally proposed as a simple model of the cultural transmission of languages (Kirby, 2001). In this case, we imagine a chain of agents, where each agent observes data gener... |