Fig. 4

All humans versus algorithms of all orders. For each of the 12 subjects there is a cluster of four bars depicting, from left to right, results for the human (in pink), the zeroth-order doppelgänger (in light gray), the first-order (in light brown), and the second-order (in light blue). The thick, red horizontal line at the top of each pink bar show that human’s mean early error across the 24 test blocks. The thick horizontal line at the top of each doppelgänger bar shows its mean early error across one million repetitions of the whole 24-block experiment. The thin vertical line shows the range of results achieved by the doppelgänger over those million repetitions. The graph shows that humans outperformed zeroth-order doppelgängers, achieving smaller early errors; i.e., the leftmost, pink bar in each cluster is shorter than the adjacent, light gray bar. And humans learned worse than second-order doppelgängers, while contests with first-order doppelgängers yielded mixed results