Effects of an artificial agent as a behavioral model on motivational and learning outcomes

Earlier research provides inconsistent evidence for effects of pedagogical agents on learning outcomes. We claim that the agent's instructional method helps to explain the inconsistency, with agents that offer behavioral modeling as an instructional method leading to better learning. We conducted two studies to investigate effects of the use of an artificial agent as a persuasive behavioral model on individuals' learning outcomes in the context of computer training. Specifically, in both studies, participants watched an instructional video on how to perform a web search with their eyes using a novel eye-tracking software. Study 1 examined the effects of agent-delivered modeling vs. two non-modeling instructional methods (agent-delivered instructional narration and no agent, text-only instruction) on participants' (N = 197) self-efficacy and system-specific perceptions of ease of use. Study 2 extends findings of Study 1 by examining effects of agent-delivered modeling vs. two non-modeling instructional methods (agent-delivered instructional narration and no agent, voice-only instructional narration) on participants' (N = 99) declarative knowledge and task performance. Previous work with human behavioral models showed an advantage of behavioral modeling over other non-modeling instructional methods in influencing learning outcomes. Therefore, agent-delivered modeling was predicted to be more effective in influencing motivational (i.e., self-efficacy) and learning outcomes (i.e., declarative knowledge, task performance). In accordance with our hypotheses, results revealed that participants who received instructions from an artificial agent as a behavioral model reported significantly stronger self-efficacy beliefs, tended to have higher system-specific ease of use, exhibited enhanced declarative knowledge, and better task performance skills.