COUNTERFACTUALS AS EXPLANATIONS
People often use counterfactual “if only…” thoughts to explain past behavior, e.g., “If only I had had more time I would have been able to write a better essay”. What sorts of counterfactuals work best as explanations? The question is especially relevant to helping to design Artificial Intelligence (AI) systems that can provide explanations for their decisions, that humans find helpful. For example, an automated loan application system could explain its decision to turn down a loan for 10K by providing a counterfactual that changes a relevant variable, e.g., “if you had applied for a loan under 5K your application would have been approved” or “if you had earned more than 25K last year your application would have been approved”. Three postgraduate students are working on counterfactual explanations in Explainable AI (XAI). PhD student Greta Warren (co-supervised by Professor Mark Keane in UCD and funded by the Insight Research Centre) has carried out experiments comparing causal and counterfactual explanations of the decisions of an AI system; MSc student Lenart Celar from Google is examining the role of domain expertise in counterfactual explanations in XAI; and PhD student Xinyue Dai (co-supervised by Professor Mark Keane and funded by a Teagasc Walsh scholarship) is working on counterfactual explanations for AI decision support systems in precision agriculture.
COUNTERFACTUALS AND MORAL JUDGMENTS
How do people come to accept morally bad actions? People judge an immoral action, such as “a passenger in an airplane does not want to sit next to a Muslim passenger and asks the stewardess to be moved to another seat”, to be not at all morally acceptable. But they consider it less immoral after they imagine alternative circumstances in which it would have been moral. They do so even if they spend just 20 seconds thinking about how it could have been moral. With Dr Beyza Tepe, from Bahçeşehir University in Turkey, we have been examining the role counterfactual thoughts play in shifting moral judgments.
COUNTERFACTUALS IN EXPLAINABLE ARTIFICIAL INTELLIGENCE
Counterfactuals about what could have happened are used in an array of Artificial Intelligence (AI) applications, and especially in eXplainable AI (XAI). Counterfactuals about what other decisions could have been made can aid the provision of explanations by AI systems, to make their decisions intelligible to developers and users. But not all counterfactuals are equally helpful in assisting human users to construct an interpretable model of an AI system. Our new projects on counterfactuals in XAI aim to test discoveries about the nature of the counterfactuals that humans create, to examine the creation of counterfactual explanations and to maximise the effectiveness of counterfactual use in AI. Two new postgraduate students have begun to carry out research on this topic this September – Greta Warren, co-supervised with Dr. Mark Keane in UCD and funded by the Insight Centre in UCD, and Lenart Celar, of Google.
Ruth gave two talks on the topic at the International Joint Conference on Artificial Intelligence, IJCAI19, in Macao, China in August 2019: one was the keynote talk “Constraints on Counterfactuals” in the Explainable AI Workshop (see the workshop website here) and see the keynote slides here. The second talk was in the survey track of the main conference on “Counterfactuals in Explainable Artificial Intelligence (XAI): Evidence from Human Reasoning” (see the conference website here) and the paper is available here.
Mark gave a talk in the Explainable AI Workshop on “The twin system approach as one generic solution for XAI: An overview of ANN-CBR twins for explaining deep learning” and in the main conference his work with Eoin Kenny was presented by Eoin, “Twin-Systems to Explain Artificial Neural Networks using Case-Based Reasoning: Comparative Tests of Feature-Weighting Methods in ANN-CBR Twins for XAI”. Mark’s IJCAI paper is available here.
Dr. Orlando Espino, La Laguna University, Tenerife, is visiting the lab for the Michaelmas semester, to carry out new experiments in collaboration with Ruth Byrne and Phil Johnson-Laird (Princeton University) on how people understand future counterfactuals about possibilities and certainties. Mary Ann Ciosk, an MSc student from KU Leuven, is visiting the lab for the academic year to carry out experiments on counterfactuals in fiction. Dr. Isabel Orenes is continuing to carry out further eye-tracking studies of counterfactuals, this time examining negative counterfactuals, in collaboration with Ruth Byrne, following the publication of their experiments on counterfactuals:
Orenes, I., García Madruga, J.A.,Gómez-Veiga, I., Espino, O. & Byrne, R.M.J. (2019) The comprehension of counterfactual conditionals: evidence from eye-tracking in the visual world paradigm. Frontiers in Psychology. 10:1172. pdf
Our research program examines the sorts of mental representations and cognitive processes that people rely on to make inferences, especially from counterfactual conditionals. Two new series of experiments have been completed with Dr Orlando Espino, La Laguna University, Tenerife, one series tests the suppression of inferences by knowledge about background conditions in counterfactual conditionals, and the other series uses comprehension priming measures to examine how people keep track of the epistemic status of the conjecture and the presupposed facts for counterfactuals. Related experiments have been completed with Dr. Isabel Orenes, in collaboration with Juan Garcia-Madruga, UNED Madrid, to examine how people understand counterfactual conditionals using eye-tracking in the visual world paradigm. In collaboration with Phil Johnson-Laird from Princeton University and New York University, we are examining how people think about the probability of counterfactuals.
A new model theory of how people make inferences about facts and possibilities was developed by Sunny Khemlani and Phil Johnson-Laird, published in Khemlani, S., Byrne, R.M.J. & Johnson-Laird, P.N., (2018). Facts and possibilities: A model- based theory of sentential reasoning. Cognitive Science. In press. Dr Celia Rasga from ISPA Lisbon published her study of counterfactual reasoning in children with autistic spectrum disorder in Rasga, C., Quelhas, A. C., & Byrne, R. M. (2017). How children with autism reason about other’s intentions: false-belief and counterfactual inferences. Journal of Autism and Developmental Disorders. 47, 6, 1806 – 1817. Sergio Moreno-Rios, from the University of Granada, completed his study of inferences by mock jurors, published in Moreno-Rios, S. & Byrne, R.M.J. (2018). Inferences from disclosures about the truth and falsity of expert testimony, Thinking and Reasoning, 24, 1, 41 – 78.
Ruth talked about some of these topics at the London Reasoning Workshop in July 2017, the University of Trento CiMEC 10th Anniversary conference in October 2017, the Workshop on the Philosophy of Cognitive Science, Buenos Aires, Argentina in November 2017, and the Philosophy of Imagination Conference, Ruhr-University Bochum Germany in March 2018.
COUNTERFACTUALS AND MORAL JUDGMENTS
With Stefania Pighin and Kayta Tentori from the University of Trento, we have carried out experiments to examine how readers and actors make judgments about whether to cooperate with a partner or not, to test the effects of counterfactual thoughts about how things could have turned out better or worse in social dilemmas such as the ‘prisoners’ dilemma’. Hannah Gallivan has completed an experiment on counterfactual thoughts and moral judgments in individuals with acquired brain injury and individuals with spinal cord injury, for her thesis for a Clinical Doctorate at Trinity.
Several other projects on how people make moral judgments have recently been completed in the lab. Shane Timmons was awarded his PhD, which was funded by a TCD postgraduate scholarship and a grant from the John Templeton Foundation awarded to Ruth Byrne. He has published a paper on the effects of fatigue on moral judgments, Timmons, S. & Byrne, R.M.J. (2018). Moral Fatigue: The Effects of Cognitive Fatigue on Moral Reasoning. Quarterly Journal of Experimental Psychology, in press, and one on moral hindsight in judgments about morally good actions, Byrne, R.M.J. & Timmons, S. (2018). Moral hindsight for good actions and the effects of imagined alternatives to reality. Cognition. 178, 82-91. Mary Parkinson, who completed her PhD the previous year, funded by the Irish Research Council, has published a paper on moral reasoning about risky choices, Parkinson, M. & Byrne, R.M.J. (2017). Moral judgments of risky choices: a moral echoing effect. Judgment & Decision Making, 12, 3, 236-252, and one on the effects on moral judgments of imagining how things could have been different, Parkinson, M. & Byrne, R.M.J. (2017). Counterfactual and semifactual thoughts in moral judgments about failed attempts to harm. Thinking and Reasoning. 23, 4, 409-448.
Shane talked about some of these experiments at the London Reasoning Workshop in July 2017 and Ruth gave a talk about some of them at the Experimental Psychology Conference, Shoal Bay, Australia in April 2017, and at the Experimental Psychology Society’s symposium on moral judgment, Queen’s University Belfast, in March 2017.
CURRENT PROJECTS 2017
One long-standing research question pursued in the lab is, what sorts of mental representations and cognitive processes do people rely on to make deductive inferences? A current set of experiments with Dr. Isabel Orenes, UNED Madrid, who visited TCIN for three months in Michaelmas term 2016, uses eye-tracking in the visual world paradigm to examine whether people envisage a single possibility when they understand counterfactual conditionals or whether they consider alternative possibilities. Isabel gave a talk on this project at the international meeting on reasoning in London in July. Another set of experiments with Dr Orlando Espino, La Laguna University, Tenerife, who visited TCIN for several weeks last semester, examines whether people construct embodied representations or iconic ones that can contain symbols when they reason with conditionals, by measuring the inferences people construct. Ruth gave a talk on these projects at the Interdisciplinary workshop on counterfactual thinking in Toronto in November, and at the annual conference on logic and cognition in Poznan in September.A third project, with Dr Sabrina Haimovici, postdoctoral fellow from the philosophy department at the University of Buenos Aires, Argentina who has been a visiting researcher at TCIN since 2015, examines dual process accounts of reasoning. Sabrina gave a talk on this project at the International Thinking Conference in Brown University, Rhode Island in August. A fourth project with Marta Couto, who was recently awarded her PhD from ISPA Lisbon and who visited TCIN for several weeks each year during her PhD, examined reasoning with advice conditionals and has recently been published: Couto, M., Quelhas, A. C., & Byrne, R. M. J. (2017). Advice conditionals about tips and warnings: interpretations and inferences. Journal of Cognitive Psychology, in press. Some of this research is funded by the Spanish Ministry of Science and Innovation and the Portuguese Foundation for Science and Technology.
A recent research question examined in the lab is, how do people make moral judgments?
One project carried out by Mary Parkinson who was awarded her PhD last year, examines whether moral reasoning relies on domain specific or domain general mechanisms, by measuring the judgments of wrongness and moral responsibility that people make when they read about transgressions of various sorts. Mary has recently published some of these experiments: Parkinson, M., & Byrne, R. M. J.(2017). Judgments of moral responsibility and wrongness for intentional and accidental harm and purity violations. Quarterly Journal of Experimental Psychology, in press. Shane Timmons who is currently carrying out his PhD, has examined the effects of cognitive fatigue on moral judgment, by testing the moral decisions people make after they have carried out laboratory tasks designed to deplete cognitive resources, or after they have carried out everyday tiring tasks such as attending an evening statistics lecture. Shane gave a talk on these experiments at the International Thinking Conference in Brown University, Rhode Island in August 2016. A third project examines how people reason about morally inspiring actions, by testing people’s moral judgments and pro-social helping behavior after they have read uplifting newspaper stories about noble self-sacrificial acts. Ruth talked about some of these experiments at the summer school in cognitive sciences at the University of Quebec at Montreal in June and at the EASP meeting on counterfactual thinking in Aix-en-Provence in June 2016. This research is funded by the Irish Research Council and the John Templeton Foundation US. Ruth has recently written an invited article in Current Directions in Psychological Science on some of this work: Byrne, R.M.J. (2017). Counterfactual thinking: from logic to morality. Current Directions in Psychological Science. In press.
How do people imagine alternatives to reality and create thoughts about how the past might have turned out differently ‘if only…’ or thoughts about how it might have turned out the same ‘even if…’? In one project, with PhD student Raluca Briazu from the University of Plymouth, who visited TCIN for three months in Hilary term 2016, the effects of ‘if only’ thoughts on deception have been examined. Raluca gave a talk on counterfactuals and deception at the International Thinking Conference in Brown University, Rhode Island in August. Another project with Celia Rasga, who was recently awarded her PhD from ISPA Lisbon and who visited TCIN for several months each year during her PhD, examined children’s counterfactual reasoning and false belief reasoning about intentions and has recently been published: Rasga, C., Quelhas, A.C., & Byrne, R.M.J. (2016). Children’s reasoning about other’s intentions: False-belief and counterfactual conditional inferences, Cognitive Development, 40, 46 – 59. A similar study has recently been completed with children with autistic spectrum disorder.
Ruth talked about some of these ideas at the Center for Language, Logic and Cognition in Turin in April and at the Workshop on the Scientific Imagination in LSE London in October 2016. Some of the research was funded by the Portuguese Foundation for Science and Technology. A third project, with Dr Marta Straga from the University of Ferrara, who visited TCIN for one month in Michaelmas term 2016, compares the counterfactual alternatives that people construct about the past (‘things would have been better for me if…’) to the prefactual alternatives that they construct about the future (‘things will be better for me if…’) when they attempt to solve puzzles. Ruth published an invited review on counterfactual thinking in the Annual Review of Psychology in 2016: Byrne, R.M.J. (2016). Counterfactual Thought. Annual Review of Psychology. 67, 135–157