Gamblers tend to derive utility from gambling, i.e. the act of gambling gives them a rush. The extreme case of this is video games. I'll never win any money by purchasing a Wii (like the lottery, I lose with probability close/equal to 1). But I will derive more happiness from playing Wii sports than I will from the money I spent on it. Similarly, gamblers enjoy playing the game, and this gives gambling positive utility (for them).
However, the argument Karelis seems to be making is that the marginal utility of increased income is zero (for some income range which constitutes poverty). People don't tend to do actions with zero marginal utility, so they don't try to increase their income. Basically, earning $15k/year doesn't make you any happier than $10k/year, so why bother to increase your income?
(Note: he actually claims that "economics doesn't apply to the poor", which is either stupidity or hyperbole. )
One interesting policy implication of his model (if it is true, which I doubt): taxing the poor is good policy. If a poor person is no happier at 15k/year than at 10k/year, then we should tax all poor people making $15k/year at 33%. It won't make them any more unhappy, whereas taxing a richer person will. It also suggests that any wealth transfers to the poor that dont solve all their problems are wasted (welfare is "all or nothing").