One counterfactual does not make an explanation

Counterfactual explanations gained popularity in artificialintelligence over the last years. It is well-known that it is possible togenerate counterfactuals from causal Bayesian networks, but there is noindication which of them are useful for explanatory purposes. In thispaper, we examine what type of counterfactuals are perceived as moreuseful explanations for the end user. For this purpose we have conducteda questionnaire to test whether counterfactuals that change an actionable cause are considered more useful than counterfactuals that changea direct cause. The results of the questionnaire showed that actionablecounterfactuals are preferred regardless of being the direct cause or having a longer causal chain