Figure 4

Representative results for transmission chains with human participants in which people learn functions. (a–d) Each row shows a single chain. (i) The (x, y) pairs were presented to the first participant in the chain, being represented as the width and height of horizontal and vertical rectangles, respectively. Participants then made predictions of the value of y for new x values ((ii) n=1, (iii) n=2, (iv) n=3, (v) n=4, (vi) n=5, (vii) n=6, (viii) n=7, (ix) n=8, (x) n=9). These predictions formed the (x, y) pairs given to the next person in the chain, whose data appear in (ii)–(x) and so forth. Consistent with the previous research exploring human inductive biases for function learning, chains produced linear functions with mostly positive slopes, regardless of whether they were initialized with (a) a positive linear function, (b) a negative linear function, (c) a nonlinear function or (d) a random collection of points.