Realistic and complex visual chasing behaviors trigger the perception of intentionality

Abstract

We not only perceive the physical state of the environment, but also the causal structures underlying the physical state. Determining whether an object has intentionality is a key component of this process. Among all possible intentions, the intention that has arguably been studied the most is chasing-often via a reasonably simple and stereotyped computer algorithm ("heat-seeking"). The current study investigated the perception of multiple types of chasing approaches and thus whether it is the intention of chasing that triggers the perception of chasing, whether the chasing agent and the agent being chased play equally important roles, and whether the perception of chasing requires the presence of both agents. We implemented a well-studied wolf chasing a sheep paradigm where participants viewed recordings of a disc (the wolf) chasing another disc (the sheep) among other distracting discs. We manipulated the types of chasing algorithms, the density of the distractors, the target agent in the task, and the presence of the agent being chased. We found that the participants could successfully identify the chasing agent in all conditions where both agents were present, albeit with different levels of performance (e.g., participants were best at detecting the chasing agent when the chasing agent engaged in a direct chasing strategy and were worst at detecting a human-controlled chasing agent). This work therefore extends our understanding of the types of cues that are and are not utilized by the visual system to detect the chasing intention.

Fig 1

(A) Stimuli were generated in a separate stimulus generation phase where human participants controlled either the sheep-alone (gray disc) against a computer-controlled wolf (white disc marked with a “W”; note that for the human participant playing as the sheep, the wolf would have been without the letter mark, the same as the distractor discs) or else different humans controlled the sheep and wolf. Note that for clarity of illustration purposes only 10 distractors are shown, but the actual display contained 19 distractors. The “human sheep” view was then translated into movie format for use in Experiment 1. (B) Human-controlled sheep were tasked with avoiding a wolf that was controlled either by the computer, guided by a direct chasing algorithm (top panel), or else that was controlled by another human being (right panel), that could decide for herself/himself at each moment in time which direction to go in order to most effectively catch the sheep.

Fig 2. Identification accuracy of computer wolf and human wolf trials.

Error bars represent the standard error of the mean. The dashed line represents chance level performance. Participants’ wolf identification rate in both conditions was significantly better than chance. However, the wolf identification rate was significantly higher in the computer wolf than in the human wolf condition.

Fig 3. Illustration of different chasing algorithms for Experiment 2–4.

In the direct chasing condition, the wolf always moved toward where the sheep was located. In the sheep mimicking condition, the wolf mimicked the sheep’s behavior unless they were close to the sheep, then the wolf switched to the direct chasing algorithm. In the sheep interception condition, the wolf predicted the sheep’s movement trajectory and moved toward the predicted location unless they were close to the sheep, then the wolf switched to the direct chasing algorithm.

Fig 4. Identification accuracy across different chasing algorithms.

Each bar represents the identification accuracy of a chasing algorithm and two distractor conditions. Error bars represent the standard error of mean. Dashed lines represent chance level performance in 10 and 19 distractors conditions. Asterisks above bars indicate significance level (*: < .05; **: < .01; ***: < .001). Participants’ wolf identification rate was best in the direct chasing condition, followed by the sheep interception, and lastly the sheep mimicking. Participants were also better in the 10 distractors condition compared to the 19 distractors condition.

Fig 5. Identification accuracy in the wolf identification task and sheep identification task.

Each bar represents one chasing algorithm (from left to right: direct chasing, sheep mimicking, sheep interception). Error bars represent standard error of mean. Dashed line represents the average chance level performance across 10 and 19 distractors conditions. Participants showed a better identification rate in the sheep identification task compared to the wolf identification task across the board.

Fig 6. Identification accuracy separated by the number of distractor discs.

Horizontal dashed lines represent chance level identification rate. Participants’ wolf identification rate was at chance in both the 10 and 19 distractors conditions.