1: \begin{abstract}

2:   A long-standing debate is whether social influence improves the

3:   collective wisdom of a crowd or undermines it. This paper addresses this

4:   question based on a na\"{\i}ve learning setting in influence systems

5:   theory: in our models individuals evolve their estimates of an unknown

6:   truth according to the weighted-average opinion dynamics. A formal

7:   mathematization is provided with rigorous theoretical analysis. We obtain

8:   various conditions for improving, optimizing and undermining the crowd

9:   accuracy, respectively. We prove that if the wisdom of finite-size group is improved, then the

10:   collective estimate converges to the truth as group size increases, provided individuals' variances are finite. We

11:   show that whether social influence improves or undermines the wisdom is

12:   determined by the social power allocation of the influence system: if the

13:   influence system allocates relatively larger social power to relatively

14:   more accurate individuals, it improves the wisdom; on the contrary, if

15:   the influence system assigns less social power to more accurate

16:   individuals, it undermines the wisdom. At a population level,

17:   individuals' susceptibilities to interpersonal influence and network

18:   centralities are both crucial. To improve the wisdom, more accurate

19:   individuals should be less susceptible and have larger network

20:   centralities. Particularly, in democratic influence networks, if

21:   relatively more accurate individuals are relatively less susceptible, the

22:   wisdom is improved; if more accurate individuals are more susceptible,

23:   the wisdom is undermined, which is consistent with the reported empirical

24:   evidence. Our investigation provides a theoretical framework for

25:   understanding the role social influence plays in the emergence of

26:   collective wisdom.

27: \end{abstract}

28: