Ask most simulation and scenario designers where the learner belongs, and they'll say the same thing: put them in the driver's seat, first person, taking the fire directly. It feels obviously right, and it's often quietly wrong.
Here's the problem with the hot seat. The moment a learner is personally on the spot, the experience starts to feel like assessment. Their attention divides between learning the material and defending their performance. And there's a subtler cost: you can't safely show bad approaches. If the learner personally commits the mistake, the feedback lands as judgment. If you steer them into the mistake to make a teaching point, it lands as entrapment. Either way, the scenario designer loses one of the most valuable moves in the craft: letting a tempting wrong approach play out where everyone can learn from it.
There's a better seat, and decades of learning research point straight at it.
What does the research say about learning by observing?
In the early 2000s, researchers studying tutoring made an inconvenient discovery: learners who merely observed a tutoring dialogue often learned a striking amount. In one study, students who overheard a virtual tutee asking a virtual tutor a lively series of questions wrote significantly more in free recall, and asked significantly more and deeper questions of their own afterward, than students who heard the same content delivered as a monologue (Craig, Gholson, Ventura & Graesser, 2000). Later work went further: pairs of students who observed a recording of another student being tutored, while working the same physics problems, learned to solve those problems as effectively as the tutees who were actually being tutored (Chi, Roy & Hausmann, 2008). The effect has a name: vicarious learning through observed dialogue.
Why would watching a conversation rival receiving instruction? Because a dialogue exposes thinking, not just conclusions. When a tutee hesitates, asks the question the observer was too embarrassed to ask, tries the plausible-but-wrong approach, and gets corrected, the observer sees the full reasoning path: the misstep, the correction, and the why behind the correction. The research identifies the active ingredient with unusual precision. In the Craig study, the virtual tutee modeled certain question categories heavily, and the observers picked up the habit: 68% of their own follow-up questions came from those modeled categories, against 48% for observers of the monologue, and on the strict deep-reasoning categories the split was 50% against 36%. Follow-up experiments then isolated the mechanism directly. When a deep-level-reasoning question preceded each piece of content in an observed dialogue, vicarious learners outperformed every other condition tested, including learners who interacted directly with the tutoring system themselves (Craig, Sullins, Witherspoon & Gholson, 2006).
None of this should surprise anyone familiar with Bandura's social learning theory (1977): humans are built to learn from observed models, especially models who resemble them.
Why is pure observation not enough?
Because observation done passively leaves most of the value on the table. Chi's ICAP framework ranks learning engagement in four modes: interactive beats constructive, constructive beats active, and active beats passive (Chi & Wylie, 2014). A learner who only receives information stores it in isolation, where it can be recalled but not readily applied. A learner who generates ideas, decisions, and explanations builds knowledge that connects to what they already know and transfers to new situations. The framework's authors are direct about where a silent viewer sits: watching without doing anything else is the passive mode, the lowest rung on the ladder.
The observation research itself draws the same boundary. In the physics study, it was pairs of observers, talking through the problems together, who matched the tutees. Students who observed alone, with no one to engage, gained less; their own before-and-after improvement was not statistically significant (Chi, Roy & Hausmann, 2008). Observing carried the content; interaction supplied the engagement that turned it into learning.
So the design question for instructional designers is not "hot seat or spectator seat?" Both give something up. The real question is: how do you keep the psychological safety of observation and the cognitive engagement of participation at the same time?
What changes when the learner takes the advisor seat?
Put a peer character in the difficult situation: someone at the learner's level, facing the learner's real challenges. Then make the learner that peer's advisor. The peer turns to the learner at the pivotal moments and asks what they should do. The learner decides; the peer acts on it; the consequences and the coaching play out in the conversation.
The hot seat
The learner is on the spot
The situation confronts the learner directly. Every choice feels graded.
Attention splits: learning the material vs. defending their performance
The advisor seat
A peer faces the situation. The learner advises.
The peer asks: “What do you think we should do?” The learner decides; the peer acts; the consequences play out.
All attention: on the decision. Same choices, none of the threat.
This one move collects an improbable stack of research-backed benefits at once:
- The generation effect. Information you generate is remembered better than information you merely receive, a finding so robust it held across recognition, recall, and every encoding variation the original experiments threw at it (Slamecka & Graf, 1978). Deciding for the peer forces the learner to generate the answer: the same cognition as deciding for themselves, with none of the defensiveness.
- The protégé effect. Being responsible for someone else's success changes how hard learners work. In the studies that named the effect, students who believed they were helping a computer character learn spent more time on learning activities and learned more than students working for themselves, with the biggest gains among lower achievers, and the researchers suggest the character absorbs the sting of failure, protecting the learner's own confidence (Chase, Chin, Oppezzo & Schwartz, 2009). Helping is more motivating than being tested.
- Safe error exposure. The peer can voice the tempting-but-suboptimal approach out loud, and research on worked examples shows that studying incorrect solutions alongside correct ones deepens learning once learners bring some prior knowledge to the comparison; for true novices, correct examples alone worked better (Große & Renkl, 2007). In the advisor seat, the peer absorbs the mistake; the learner absorbs the lesson.
- Real decision data without the assessment feel. The learner's choices still record exactly what they would do in each situation. But the emotional frame is "helping a colleague," not "being graded," so the data reflects genuine judgment rather than test-taking behavior.
Notice what happened on the ICAP ladder. The advisor is not a passive observer; they are generating a decision at every pivotal moment and getting a response to it, which moves them well up the framework's ladder from watching.
How is AliveSim built around the advisor seat?
The approach comes first, and it stands on its own: give the learner a peer to advise, surround the decision with the people it affects, and deliver coaching inside the conversation. AliveSim was built around exactly this pattern. Its Guided Scenarios play out among 3D avatar characters: a peer character who faces the situation and genuinely needs the learner's help, subject characters the situation is about, such as the patient, the customer, or the direct report, and a mentor character available for coaching. The characters converse with each other and with the learner, and the learner participates through decisions, so every line the learner encounters is authored, reviewed, and on-message.
The multi-avatar design is what lets each research finding do its work. The decisions belong to the peer's situation, so the learner is steering a colleague through a real moment rather than answering quiz items about one. Because the peer can safely believe the wrong thing, the scenario can surface the exact misconception a learner population actually holds and let the peer voice it, which is the safe error exposure the worked-example research describes: the tempting approach gets examined in the open instead of hidden behind a wrong-answer penalty. And when a choice needs correcting, the coaching arrives as conversation from a character rather than as a verdict from the system, so the learner stays inside the situation while they reconsider. What the learner experiences is the vicarious-learning finding from the research, rebuilt as a design: reasoning made visible in dialogue, with the learner's own judgment driving every turn.
For how the coaching itself works at each choice, see our guide to decision-point mentoring; for how to construct the choices, see designing decision points.
The takeaway
Where the hot seat tests a learner, the advisor seat develops their judgment. If your scenarios put learners in first person by default, you're paying an assessment tax on every learning moment, and giving up the safest, most human way to show what not to do. The learner who advises a peer makes every decision the hot seat would have asked of them, generates every answer, and carries responsibility for someone else's success, all without spending a single moment defending themselves. The research has been pointing at the other chair for more than twenty years. Content developers who take it seriously don't abandon challenge; they relocate it to the seat where challenge produces learning instead of defense.
References
- Bandura, A. (1977). Social Learning Theory. Englewood Cliffs, NJ: Prentice-Hall.
- Chase, C. C., Chin, D. B., Oppezzo, M. A., & Schwartz, D. L. (2009). Teachable agents and the protégé effect: Increasing the effort towards learning. Journal of Science Education and Technology, 18(4), 334–352.
- Chi, M. T. H., Roy, M., & Hausmann, R. G. M. (2008). Observing tutorial dialogues collaboratively: Insights about human tutoring effectiveness from vicarious learning. Cognitive Science, 32(2), 301–341.
- Chi, M. T. H., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243.
- Craig, S. D., Gholson, B., Ventura, M., & Graesser, A. C. (2000). Overhearing dialogues and monologues in virtual tutoring sessions: Effects on questioning and vicarious learning. International Journal of Artificial Intelligence in Education, 11, 242–253.
- Craig, S. D., Sullins, J., Witherspoon, A., & Gholson, B. (2006). The deep-level-reasoning-question effect: The role of dialogue and deep-level-reasoning questions during vicarious learning. Cognition and Instruction, 24(4), 565–591.
- Große, C. S., & Renkl, A. (2007). Finding and fixing errors in worked examples: Can this foster learning outcomes? Learning and Instruction, 17(6), 612–634.
- Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592–604.
Related questions
What is vicarious learning?
Vicarious learning is learning by observing someone else learn, rather than by being the person instructed. The term has been used alongside observational learning and social learning since Bandura's early work. The modern research adds an important refinement: what the observer watches matters enormously. Learners who overheard a tutoring dialogue, in which a tutee asked questions and worked through misunderstandings with a tutor, recalled more and went on to ask deeper questions themselves than learners who heard the same content delivered as a monologue (Craig, Gholson, Ventura & Graesser, 2000). A dialogue exposes reasoning, hesitation, and correction, which is exactly the material an observer can learn from.
What is the protégé effect?
The protégé effect is the finding that learners put more effort into learning when they are responsible for someone else's success than when they are learning only for themselves. In studies with a computer character students believed they were instructing, students who worked on behalf of the character spent more time on learning activities and learned more than students doing the same tasks for themselves, with the strongest benefits among lower-achieving students (Chase, Chin, Oppezzo & Schwartz, 2009). The researchers suggest that this sense of responsibility also protects the learner's own ego: the character carries the failures, so learners engage with mistakes instead of being discouraged by them. Advisor-seat scenario design applies the same principle: the learner is working for the peer character's success, not defending their own record.
Does observing a conversation really work as well as participating in one?
Under the right conditions, it gets remarkably close. Pairs of students who observed a recorded tutoring dialogue while working the same physics problems learned to solve them as effectively as the tutees who were actually being tutored (Chi, Roy & Hausmann, 2008). The same study shows the boundary, though: students who observed alone, with no one to think out loud with, gained less. Observation carries the content; some form of active participation supplies the engagement. That is precisely why the advisor seat pairs the two rather than choosing between them: the learner observes the peer's situation but generates every decision in it.
How do you keep an observer engaged instead of passive?
Give the observer a job that requires generating something. The ICAP framework predicts that interactive engagement beats constructive, which beats active, which beats passive (Chi & Wylie, 2014), and a viewer with no role sits in the passive mode. The advisor seat is one reliable way to move an observer up the ladder inside a scenario: the peer character turns to the learner at each pivotal moment and asks what to do, so the learner must commit to a decision before the story continues. Other techniques point the same direction, such as characters addressing the learner directly to draw them into the conversation, but the constant is that the learner produces decisions rather than merely watching outcomes.
Published July 17, 2026 · 8 min read