I think his point is not that random citizens can't be corrupt, it's that a random assortment of citizens are less likely to be corrupt (and probably more likely to call out corruption if they see it).

A random assortment of citizens seems just as likely to be more corrupt than less, if it's even possible to measure that objectively.

However, it does seem likely to me that random citizens would be less capable of governing effectively in areas where actual knowledge of law, politics or some degree of domain knowledge is required.