Vicarious Learning

In the AI Story

The power of a vicarious model depends on perceived similarity. A twelve-year-old watching Simone Biles perform a vault learns almost nothing about her own gymnastic capability, because the dissimilarity is too great to support inference. A twelve-year-old watching a classmate perform the same vault learns a great deal, because the similarity supports the inference "if she can, perhaps I can." Bandura's research showed that similarity along dimensions the observer considers relevant — age, gender, background, prior ability — modulates the efficacy-building effect of observing a model.

The AI discourse has created vicarious learning conditions unlike anything Bandura studied. Every successful builder posts their triumphs; every failed attempt dies quietly. The visible models are overwhelmingly success stories, and the failure stories are systematically excluded from the feed. An observer trying to calibrate her own AI self-efficacy against the apparent performance of similar others is working from radically biased data. She sees triumphalists building revenue-generating products in weekends and infers that her own inability to do the same reflects personal inadequacy rather than statistical selection.

This asymmetry has a specific psychological cost. Observers develop what Bandura's later research called "efficacy-undermining social comparison": they conclude that their struggles are anomalous when they are in fact modal. The fishbowl here is the feed itself, which shapes what is visible and hides the rest. Effective vicarious learning in the AI age requires surfacing the full distribution of outcomes, including the ones that did not generate shareable content.

Organizations can build vicarious learning environments by deliberately curating models that match their employees' starting points and by making failure and iteration visible rather than only outcome success. The engineer who sees her colleague struggle for three days before producing a working result learns something that no polished demo can teach: that the struggle is the work, that it ended in success, and that she can do the same.

Origin

Vicarious learning was established empirically in Bandura's Bobo doll experiments in the early 1960s. Children who observed an adult beat an inflatable clown reproduced the aggressive behavior without any direct reinforcement, demonstrating that learning could occur through pure observation. The finding became the foundation of social learning theory and, eventually, of the multi-source model of self-efficacy.

Key Ideas

Second-strongest source. Vicarious models update efficacy beliefs more weakly than direct mastery but strongly enough to matter when direct experience is absent.

Similarity dependence. The efficacy-building effect is proportional to the perceived similarity between observer and model.

Selection bias in the AI era. Social media surfaces triumphs and hides failures, distorting the distribution of visible models.

Process vs. outcome visibility. Observers learn more from watching models struggle and persist than from watching only the final success.

Deliberate curation. Organizations can engineer vicarious learning by selecting and surfacing models whose trajectories employees can plausibly follow.

Debates & Critiques

Whether social media 's distorted model distribution does net harm or net good to AI self-efficacy remains open. The triumphalist feed may be motivating for some observers and devastating for others; the same post that inspires one builder undermines another whose starting position is less favorable.