Collaboration Variability in Autism Spectrum Disorder

Abstract

This paper addresses how impairments in prediction in young adults with autism spectrum disorder (ASD) relate to their behavior during collaboration. To assess it, we developed a task where participants play in collaboration with a synthetic agent to maximize their score. The agent's behavior changes during the different phases of the game, requiring participants to model the agent's sensorimotor contingencies to play collaboratively. Our results (n = 30, 15 per group) show differences between autistic and neurotypical individuals in their behavioral adaptation to the other partner. Contrarily, there are no differences in the self-reports of that collaboration.

Keywords: autism; collaboration; neurodiversity; prediction; sensorimotor contingencies.

FIGURE 1

Example of a trial during the task. Targets fall from the top of the screen and players need to intercept them before they hit the ground. The player on the left (blue) is controlled by a human and the one on the right (green) is the synthetic agent. This example represents interaction with the “Middle” agent.

FIGURE 2

The left image represents the probability to go for the target depending on the player’s relative distance to the target. The right image depicts the Point of Social Subjective Equality (PSSE). Here, the green line represents the AI agent (in this case, the “Middle” one) and the black line, a simulated perfectly matched participant. The red dashed line represents the moment when both agents have the same probability of going for the target. The x-axis represents the relative distance of the agent from the target; the y-axis, the probability it has to go for the target.

FIGURE 3

Representation of the curve of the probability to intercept the target based on the target’s location for each of the three proposed agents. From left to right: the curve of the “Middle” agent lies in the middle as both the synthetic agent and the human player have the same probability of intercepting the target. In contrast, the curve of the “Wider” synthetic agent is slightly skewed toward the left, as this agent will enter the space of the human participant. In contrast, the “Narrower” agent’s curve is skewed toward the right; this agent has a higher probability of staying toward its half of the space and allowing the human participant to intercept the target.

FIGURE 4

Representation of the setup used in the task. Participants sit in front of a computer screen where the game was displayed. Participants controlled the motion of their avatar using a touchpad.

FIGURE 5

Representation of the experimental protocol. First, participants are introduced to the task and fill in a short questionnaire. The task comprises three phases, which contain one, three, and one blocks (of 150 trials per block), respectively. In “Phase 1,” participants play alone. In “Phase 2,” participants randomly play with each of the AI agents for one block (three blocks in total). In “Phase 3,” participants are presented with a random agent in each trial (50 trials per agent). Self-reports on perceived collaboration, engagement, and agents’ and target predictability are presented between phases/blocks. Finally, participants fill in a short questionnaire and undergo an interview and debriefing.

FIGURE 6

PSSE intersections from the TD and ASD groups (top and orange and bottom and blue, respectively). From left to right: the PSSE curve of the AI agent (depicted in green) and that of the participants, depicted in orange (TD) and blue (ASD), when playing with a “Middle” (N), “Wider” (A), and “Narrower” (S) agent respectively. The bright colors represent standard deviation, while the darker and thicker line represents the mean.

FIGURE 7

Differences between groups in the difference in shift per agent (compared to the “Middle” one). The upper plots represent early (0–50) trials and the lower plots, late (100–150) trials. Blue represents the ASD group; orange, the TD; and the red line represents the median.

FIGURE 8

Matrix representing the relationship between Phase 2 (actual) and Phase 3 (predicted) behavior. The blue matrix represents the group in the autistic spectrum, while the orange one represents data from the neurotypical group.

FIGURE 9

Differences between agents and groups in perceived agent predictability, target predictability, engagement, and collaboration. Blue represents the results for the “Narrower” agent; orange, for the “Middle” agent; and green, from the “Wider” one. Red lines represent means.

FIGURE 10

Differences in perceived collaboration between the participants that perceived the AI agent as synthetic or as a human player. The white dot represents the median.

FIGURE 11

Frequency of characteristics commented about during the interviews. Blue represents collaboration-related characteristics; purple, color-related ones; green represents characteristics related to field use; and orange, characteristics related to movement. The sum of the frequencies inside each group surpasses 100% because some subjects highlighted more than one characteristic.