Fig. 2.

Rapid learning in observers. a Experimental design: Observers (OBS) were separated from experimenters (EXP) by a screen that restricted visual interactions to the perch equipped with the air-puff delivery mechanism. During a training phase (left panel), OBS watched EXP perform the task. Thereafter, the knowledge learned by OBS was tested in the same paradigm with a new (naive) OBS (right panel). b On the first training day (left), this example EXP showed roughly equal densities of escapes (rasters) for unpuffed (light blue) and puffed trials (black). The air-puff sounds are visible in spectrograms of microphone recordings (red arrows). On the last day of training (right), the cumulative escape probabilities (black and blue lines, bottom) discriminate puffed from unpuffed trials (z-test of individual proportions, p < 0.01). c Difference of escape probabilities (dPesc) during training of an EXP (blue) and during testing of its OBS (green). Solid curves are smoothing spline fits (parameter: 5*10⁻⁵). d Bar plot of dPesc, showing that OBS (green diamonds, n = 9) discriminate significantly better in the first 3 blocks of their testing than EXP (blue circles, n = 10) in the first 3 blocks of their training (EXP ≠ OBS, p = 0.013, Wilcoxon ranksum). e Upon reaching the learning criterion, the average dPesc (3-block average) in OBS is significantly larger than in EXP (EXP ≠ OBS, p = 0.005, Wilcoxon ranksum). f OBS reach the learning criterion in fewer trials than EXP (EXP ≠ OBS, p = 0.015, Wilcoxon rank sum). The bars indicate the median across birds in each group, the black line indicates the lower bound (800 trials)