Figure 4: Superior learning performance of Ratbots over unenhanced rats.

We collected data in 16, 16, 18, 16, 15 and 18 sessions for T1, T1 Control, T2, T2 Control, T2 with mask, T3 respectively (denoted as N in Fig. 4). The subjects learned one maze in a session with eight trials. One subject was unable to finish all the experiments because of an accidental destruction of the implanted electrodes, resulting in the inconsistent session numbers. (a) Learning curves of V1 ratbots in rule learning (T1 in red), control rats in a maze without landmarks (T1 Control in blue), and V2 ratbots with rule combination and rule transfer for new mazes (T3 in green). (b) Learning curves of V2 ratbots with rule application of landmark following (T2 in red), control rats in a maze with landmarks (T2 Control in blue), and V2 ratbots with a face mask applying the rule of landmark following (T2 with mask in green). The arrows on the x-axis indicate correct rat choices more than 80% of the time. Standard deviation is depicted by the shaded area. Triangles under the horizontal axis show the results of the ANOVA, that is, whether differences between each pairs of groups were significant in individual trials. The ANOVA results of the average performance across trials 1–4 and 5–8 are also depicted in the upper two rectangles in the figures. All experiments were carried out using the same six subjects as control rat, V1 ratbot and V2 ratbot (the order of experiments conducted can be referred to section Methods “Experiments”). ***denotes significant difference (p < 0.001), **denotes p < 0.05, *denotes p < 0.1 and • denotes no significant difference.