Probabilistic reasoning in intelligent systems: networks of plausible inference
Probabilistic reasoning in intelligent systems: networks of plausible inference
An algorithm for deciding if a set of observed independencies has a causal explanation
UAI '92 Proceedings of the eighth conference on Uncertainty in Artificial Intelligence
The causal Markov condition, fact or artifact?
ACM SIGART Bulletin
Causality: models, reasoning, and inference
Causality: models, reasoning, and inference
Uncertainty in Artificial Intelligence
Uncertainty in Artificial Intelligence
A contribution to the theory and practice of cognitive prostheses
SG'03 Proceedings of the 3rd international conference on Smart graphics
Hi-index | 0.00 |
The Markov condition describes the conditional independence relations present in a causal model that are consequent to its graphical structure, whereas the faithfulness assumption presumes that there are no other independencies in the model. Cartwright argues that causal inference methods have limited applicability because the Markov condition cannot always be applied to domains, and gives an example of its incorrect application. Cartwright also argues that both humans and Nature, fairly commonly, design objects that violate the faithfulness assumption. Because both arguments suggest that data is not likely to be ideal, we suggest that problems of the theory be separated from problems of the data. As regards the Markov condition, conflicted intuitions about conditional independence relationships in certain complex domains can be explained in terms of measurement and of proxy selection. As regards faithfulness, we show that violations of this assumption do not affect the predictive powers of causal models. More generally, the criticisms of causal models, taken constructively, reveal the subtlety of the ideal, while clarifying the source of problems in data.