We all wonder what might have been. How would reality be different if a certain event had not occurred? If you had taken that job or married that person?
“People especially tend to do this when a negative event happens, a car accident, for example,” said Henrike Moll, assistant professor of psychology at USC Dornsife.
“You’ve probably caught yourself thinking, ‘Oh my gosh, if I had left the house two minutes later, then I would not have gotten into this accident. Or if only I had slammed the break a second earlier, I could have prevented this disaster.”
Engaging in this kind of thinking is called counterfactual reasoning, the subject of a study being conducted by Moll and Dehghani, assistant professor of psychology and computer science at USC Dornsife’s Brain and Creativity Institute.
The two have received a $750,000 grant over three years from the Office of Naval Research to investigate and model the emergence of counterfactual reasoning, which plays a critical role in the way people diagnose, predict, decide, and make moral and legal judgments.
Every time we consider the consequences of what would happen if we ate that piece of chocolate cake or smoked that cigarette, or lied, or even went to lunch with someone, we are using counterfactual reasoning.
“Counterfactual reasoning is something we do frequently, across many different types of tasks: in planning, diagnosis and decision-making, we always think about the alternatives,” said Dehghani, co-principal investigator on the study. “And the philosophical question is: How do we do that? How do we construct an alternative world in which it’s actually raining today even though in reality, it’s sunny?”
Understanding this cognitive process is an important question in computer science and artificial intelligence, the researchers noted.
“When we’re building intelligent robots that think like us, and have them in situations where they have to make decisions, or perform diagnosis, counterfactual reasoning is needed,” Dehghani said. “So when an agent or robot knows a set of facts, and wants to do counterfactual reasoning, a set of those facts would have to be retracted or changed.”
But the change must have consequences, and where would researchers stop this belief propagation?
“So, if a robot is thinking about what would have happened if it had left the house two minutes later, it has to let go of the belief that it left the house at time X, and try to imagine a world in which it left the house at time X+2mins,” Dehghani said. “Now, the question is when and where should this change in its belief system. Should it also affect its belief in the fact that is sunny? Should it change its belief in the sky being blue?”
This is where children enter the picture. In their research, Moll and Dehghani will observe and interact with children, focusing on the connection between pretend play and counterfactuals in early development.
Pretend play, typically showing by age two, is characterized by acting “as if” something were something it is not. For example, when a child pretends a banana is a phone or a long stick is a horse. Imaginative play, and especially play with this form of object substitution (using x as y) may be a precursor to counterfactual thought.
Despite the connections between pretend play and counterfactuals, the link had not been empirically investigated until now. Moll and Dehghani will be the first to look for concurrent and longitudinal relations between pretend play and counterfactual reasoning in young children.
“What’s interesting is that children spontaneously, when they’re about a year and a half years old, drift into these alternative worlds,” said Moll, principal investigator on the project. “We don’t view this as counterfactual reasoning, that’s a high-level term that requires more sophisticated cognitive abilities, but what we do see them doing is engage in pretend play.
“For example, if I were to take this recorder and pretend to bite into it, producing exaggerated chewing movements and sounds; that’s pretend play,” Moll said. “We think this may be a precursor to counterfactual reasoning because you act as if something were the case. So you already project yourself into an alternative world. You pretend to be in a world that is not actual, because this recorder is not a sandwich.”
Pretend play is unique to humans, Moll noted.
“We don’t see pretend play in apes; they don’t take a stick and bite it as if it were a banana,” Moll said. “This form of symbolic play, where you’re pretending something is the case when it’s not, is something that’s specifically human and it might be related to language, which also relies on symbols in order to represent things.”
In their study of children, the researchers will test for children’s understanding of regret, the prime example of counterfactual emotions. Regret affords the consideration of past counterfactuals.
“You can’t experience regret unless you are able to reason counterfactually,” Moll said. “So if we find that children experience emotions of regret, we can attribute counterfactual reasoning to them.”
To determine regret, one part of the study may involve an arcade-like container filled with plush teddy bears and toys. The child would have to choose between two mechanical claws. The chosen claw would grab a toy and drop it at a shoot that would open. But the game is rigged so that the chosen claw fails. Moll and Dehghani would observe any signs of regret in the child.
The longitudinal part of the study will assess how counterfactual reasoning develops over time. Moll and Dehghani plan to follow the same children for a few years.
“So the basis of this study is to see how children develop counterfactual reasoning,” Dehghani said, “and see whether you can use that process to model counterfactual reasoning in computers.”