"both gambling and insurance are slightly-negative-expectation plays with occasional large payoffs"
When you take in account that personal utility functions aren't linear, insurance and gambling are no longer slightly-negative-expectation, but usually positive.
In other words, if U() is your utility function, U($1M) != 1MU($1). For most people, U($1M) > 1MU($1) and U(-$1M) < 1M*U(-$1).
I don't think that's true in the general case. For example, one penny has virtually no utility to me on its own, but there's plenty I can do with one pound, getting more than a hundred times the value from it.
I don't think that's quite what people usually mean when they talk about diminishing marginal utility. Granted, "marginal" was missing from the parent comment, but I gather that was the phenomenon being discussed.
Diminishing marginal utility implies that you gain more utility by acquiring your first penny than you do acquiring your hundredth. Now, at such small levels of money, you could certainly argue that almost nothing is for sale at 1 penny, but once you get above the level where the disutility of carrying around a coin is dwarfed by the utility of the money itself, diminishing marginal utility applies pretty well.
Things may get weird at the scale of pennies or billions of dollars, but at scales relevant for buying insurance or gambling diminishing marginal utility certainly holds.
Losing $10k when you have $20k hurts less than losing $10k when you have $10k.
No, I'm suggesting that it isn't linear, and different people have different functions, even at the same level of wealth.
There are even some techniques to discover and plot your own utility curve, which is quite useful when you're handling things like investing and insurance.
For example:
* Would you give $1 for a 10% chance of receiving $10?
* Would you give $1 for a 9% chance of receiving $10?
* Would you give $10,000 for a 1% chance of receiving $1M?
* Would you give $10,000 for a 0.9% chance of receiving $1M?
* Would you receive $10 for a 1% chance of losing $1000?
* Would you receive $10,000 for a 1% chance of losing $1M?
* Would you rather do nothing or have a 50%/50% chance of winning $1000 and losing $1000?
* Would you rather do nothing or have a 50%/50% chance of winning $1M and losing $1M?
It's the other way around, "any risk is bad" and "marginal utility is decreasing" are conclusions that you reach from your own function, not that drive your function.
* Would you give $1 for a 1% chance of winning $100?
* Would you give $1 for a 1.1% chance of winning $100?
* Would you give $1 for a 1.2% chance of winning $100?
.
.
.
* Would you give $1 for a 10% chance of winning $100?
Yes, gambling for small wins can be negative, if you don't include the "excitement" or "entertainment" in the utility function.
But then you're not talking about a pure monetary transaction. More like a trade. Which goes back to your point: nobody makes a voluntary transaction where they get less value than they provide.
Take charity donations, for example: people value the warm feeling from helping others and a clear consciousness more than the money they are giving.
You have to get through a few steps to agree though! You could disagree by saying that people's "revealed preferences" are their "actual preferences", or by saying that people's utility function after accounting for hyperbolic discounting is their "actual utility function."
Of course in the case of heroin addiction it's easy to poke fun at the notion of time-discounted utility functions, but you can't really shrug off the idea, since it's vital to explaining why people do all sorts of immediately neutral or unpleasant things like brushing their teeth, saving money, or exercising.
When you take in account that personal utility functions aren't linear, insurance and gambling are no longer slightly-negative-expectation, but usually positive.
In other words, if U() is your utility function, U($1M) != 1MU($1). For most people, U($1M) > 1MU($1) and U(-$1M) < 1M*U(-$1).