Key Research Question
- How do we integrate information from others' choices into our own valuation process (e.g. social conformity)? What is the neural basis of it? And how does this integration changes the neural connectivities?
- How do we develop a mental model of another player during cooperative and competitive interactions? What is the neurl implementation of such a mental model? How does this Theory-of-Mind capacity change intra-brain and inter-brain connectivity recorded with EEG hyperscanning?
- What is the depth of social reasoning that can we engage in (e.g. "thinking about you, thinking about me, thinking about you, ...")? Do more levels of social reasoning require increasing prefrontal computations?
- What is the utility model of altruistic decision-making that clear defies the self-centered, egoistic utility model of "rational" economic decision making? What personality variables and genetic dispositions facilitate altruistic decision-making?
Most of our decisions are made in a social context and are therefore influenced by other people. Picture yourself at a lunch counter and the 4 people in front of you all order the same meal. Does that change your own choice what to have for lunch? This is an example for an observational social influence on decision-making. You observe what others choose and that triggers a re-valuation process for the available options. However, you are not competing with the other people for a limited resource (all menu items are still available to you).
This changes in a competitive or cooperative decision context. If you are at an auction and bidding on a popular item, you will try to guess how long the others will keep in the bidding race. Likewise, when you and another person have to coordinate your decision to win a card game like "sheepshead" or "Doppelkopf:, you will also try to think about what the other person know and what s/he will do, so you can adjust your choice. In other words, you are trying to come up with a model of the person and his decision-making process. This capability is commonly referred as Theory of Mind and it involves the capability to think about the thoughts and intentions of others, to put yourself in another person's shoes (perspective taking), and to think recursively about your and another person's over several steps (like in a game of chess). Especially, recursive social reasoning is one of the hallmarks of social decision-making. It involves thoughts like "I am thinking about you, who is thinking about me, who is thinking about you ..." Although this figure of thought could potentially go on ad infinitum human cognitive capacity is limited. When tested experimentally, human participants often accomplish two levels of recursivity (although recent this limit has been called into question).
Whereas observational, competitive and cooperative decision-making can be all seen as trying to maximize the own reward or profit, altruistic decision-making blatantly violates this axiomatic principle of behavioral economics. Yet, altruistic behavior is quite common throughout human society and also throughout the animal kingdom. While some explanations construe the rewards of altruistic behavior in the psychological rather than in the economic domain ("I give money to the beggar to feel good about myself."), this explanation falls short in instances, when personal suffering is consciously accepted to help another person.
Social Influence in Human Decision-Making
The decisions of other can have profound influence on our own choices. This has been demonstrate very early in the social psychological experiments by Solomon Asch (1951): a participant on the study was seated with up to 6 confederates of the experimenter (a fact unknown to the participant) and all were asked to perform simple line judgement tasks. When all confederates chose the (obviously) wrong line, participants were more inclined to choose this (wrong) option as well - a powerful demonstration of social conformity, even though it was not authoritatively enforced. However, when debriefed afterwards, some participants reported that they only made the wrong choice in order please the experimenter, to avoid conflict with the other participants, or to conform with the (supposed) goals of the study. However, they maintained their private veridical beliefs. Other participants, however, reported that they actually changed their private beliefs toward the group opinion. Although social conformity has been an active research subject for many decades, relatively little is known about the computational (How do we integrate the information about others' choices into our own decision?) and neural mechanisms.
In this project we are investigating this form of "observational" social influence in a novel decision-making task, in which subject first make a choice, then observe the decisions of 4 other players, and then decide whether to change or maintain their original choice. All of this is realized through real-time internet-based communication between the computers of the 5 players. One of the participants undergoes fMRI scanning while performing the task, whereas the others are situated in an experimental test room nearby.
Behavioral (model-free) findings indicate a strong effect of coherence of the group decisions (how many other players are choosing the same option) on the propensity to change or maintain the original choice: the more other group players decide against my first choice, the more likely I am to switch my first choice. Conversely, the more other group players are deciding with my first choice, the more likely I am to stick with my original choice. Preliminary fMRI findings show that this "conformity effect" correlates with activity in regions known from process social information and conflict (among them temporo-parietal junction (TPJ), anterior cingulate cortex (ACC), and insula). We also developed a new computational model that is aimed at uncovering how this social information is influence the valuation process. We can show that the number of people that decide with or against my first choice is significantly biasing my second choice to stay or switch. In addition, the difference in expected values and a general bias to switch or stay is also affecting the second choice. Furthermore, continuous learning of expected value is not only influence by my experienced rewards, but also by how much the other group members have won in the last 3-4 trials.
Cooperative Decision-Making and Theory of Mind
Engaging in either cooperative or competitive exchanges requires the capacity to think about another person's thoughts, intentions, and strategies. This capacity of other referred to as Theory of Mind, and it involves building a mental model of another person, which tries to predict - based on the supposed intentions and strategies - how the other person will behave and which choices s/he will make. A classic (neuropsychologiecal) test for this capacity is the Sally-Anne task, which is presented either live or as a cartoon. "Two girls, Sally and Anne, are in a room containing a box and basket. Sally puts a ball in the box and then leaves the room. Meanwhile, Anne takes the ball out of the box and puts it into the basket. Sally, then returns to the room. Where will she look for the ball?"
People with Theory-of-Mind (ToM) capacities will answer that Sally will look in the box, because that is where she thinks the ball (still) is, because she has left it there. In other words, a person with ToM capacities can abstract from his/her own knowledge base, and adopt the knowledge base of another person (Sally), which may be less comprehensive as his own (The person knows that the ball is in the basket.) The person has built a mental model of another person (Sally) and uses this model to query it for predicted actions.
In this project we have developed a novel decision-making task, which mimics the situation in the Sally-Anne-Task. Two players make choices for probabilistic rewards, and they both have to figure out (given the probabilistic uncertainty of the task), whether each of them is making choices to maximize personal gains. Then, one of the players, the Teacher, receives a signal that the reward contingencies in the other players (the Learner) have changed, whereas his own remain the same. He then has to "communicate" this change of contingencies to the Learner by using only his choices, which then have to deviate from optimality. The Learner has to detect this change of choice pattern in the Teacher and then adapt his own choices to regain choice optimality. This novel task thus requires both players to develop mental models of the other player to predict his choice pattern and then incorporate the output of this mental model into one's own choices.
Computationally, we are using mutli-agent reinforcement learning models (e.g. iPOMDPs) to model joint choice behavior of both players. The general framework (Gmyntrasiewicz & Doshi, 2005) provides methods for sophisticated mental models including several layers of reasoning. During this novel decision-making task, which involves real-time interactions through internet communication, we are also recording concurrent EEG measurements from both players (EEG hyperscanning). We are currently analyzing the "trials of social reasoning" following switches in reward contingencies using in time-frequency analyses of the trial period, in which the two players have to think about another and develop a model of the other person's strategies. Furthermore, we are also investigating neural synchronisations during ToM processes both within and between brains.
Recursive Social Reasoning
Competitive decision-making requires me to think strategically about about my competitors in order to figure out what they are going to do. However, if taken serious, then this means that my competitors also reason about my strategy. Therefore, my model of my competitors strategy has to take their thinking about my strategy into account. It is obvious that this line of reasoning can go ad infinitum. Plenty of evidence from behavior economics point to a limited capacity of humans to carry out this recursive from of social reasoning. Most studies suggest that Level-2 reasoning ("I think about, what you think about me.") is a natural limit to the human cognitive capacity, although there are a few reports that find even higher levels of reasoning.
In this project we are aiming at identifying the neural signature of different levels of social reasoning. Rather then estimating the current level of reasoning of a player using computational modeling, we are instructing our participants on the level of reasoning by requiring them to answer specific question tied to each level (although we cannot be sure about it, which is why we are also using computational modeling to assess a players level of reasoning). We employ a modified matching pennies task in a real-time interaction, with one of the participants in the fMRI scanner. We will use several Level-k models of social reasoning, as well as iPOMDPs in a model-based fMRI analysis to identify the neural architecture of social reasoning, which we expect to unravel in prefrontal activations.