Legalization almost certainly would bring with it additional substance abuse in the state, and the long-term public costs associated with that would vastly exceed the relatively modest amount of new revenue legal weed might bring in.
The economics of illegal goods weighs extremely heavily in favor of legalization and taxation rather than banning and enforcing, as Gary Becker, Kevin Murphy, and Michael Grossman outline in The Economic Theory of Illegal Goods: The Case of Drugs (NBER Working Paper, 1994).
Corrections would suggest that when marijuana is illegal, the government raises the cost of selling marijuana until production falls to the socially optimal quantity. This is a costly activity. Police must be hired, lawyers must be provided, and offenders must be jailed. The end result of all the government's expenditure is to raise price and reduce quantity consumed by the law of demand. However, legalizing marijuana and taxing its sale until the quantity demanded was reduced to the same socially optimal level as regulation would result in no change in marijuana consumption, by construction.
The graph below (click to enlarge) demonstrates the possible ways in which government can reduce marijuana consumption to its preferred point. Raising the cost of production and shifting the supply curve back is costly and loses consumer surplus. On the other hand, taxing marijuana use until consumption falls to Q_regulation results in government revenue that simply does not exist when marijuana is illegal. It is the difference between paying to destroy money and being paid to destroy it.
If the government seeks to regulate the quantity of marijuana consumed, it should choose to do so with a pricing mechanism such as taxation (from which it can earn revenue) rather than a ban (which is costly to enforce). The result is otherwise the same.
Corrections does note our a priori belief that demand will not shift out, or not shift out much, if drugs are legalized. We do not see the "stigma" of marijuana use as a major factor in utility inputs of extra-marginal users. We might further note that we see addiction arguments as lacking firm foundation in human behavior, as discussed in a previous post.