tl;dr: I think the study's idea of 'baseline' behavior relies on a payout-maximizing model we know is incorrect.Thanks, I'm well past the point of believing PR office summaries of studies on these topics. And sure enough, it looks to me like this is being oversold and also has some concerning gaps. Anyone want to help me understand that?
Press release: “To the brain, information is its own reward, above and beyond whether it’s useful”
Article: We observed that subjects’ information acquisition behavior was indeed largely driven by instrumental benefit... While we found no evidence for simplistic constant utility bonus, the utility of anticipation improved behavioral modeling.
Which is to say, subjects specifically did not place an inherent value on information independent of its utility. Rather, they adjusted its value based on its utility in a way which did not perfectly maximize payout.
More broadly and importantly, I think this is a worthwhile study, but I'm worried it's propagating a known behavior up through extra layers and calling it a new result. It's well established that people playing games of chance don't behave as perfect payout-maximizers, and specifically that they appear to be loss-averse and ambiguity-averse (e.g. Ellsberg's paradox). They also appear to place nonlinear value on the utility of money, changing their choices when all game outcomes are scaled with the same expected utility (Allais paradox). Given all of that, we can say that a player's desired outcome in such a game is not simply maximizing expected payout. As such, "perfectly rational" behavior at the information-buying step is not the decision which maximizes payout, but the one which maximizes their performance under their chosen utility function.
But the study works from the assumption that "Under standard economic accounts, agents accept the lottery if its EU is higher than the utility of status quo u(0), and reject otherwise (Fig. 2a). Furthermore, they purchase the information if its cost is lower than VOI and forgo if higher". A host of studies show that this is not how people actually play lotteries. As a result, the study concludes that anticipation utility is acting at the information-buying step, while ignoring the possibility of strictly-rational purchasing under an alternate utility function.
This could have been easily tested with a control in which players were asked to play the variable-outcome lotteries with perfect information, to establish baseline play against which to compare information-buying behavior. I'm still interested in the game results and the MRI data, but the press release tells us this result "could only be explained by a model that captured both economic and psychological motives". I'm not ready to accept that when the control is purely theoretical and based on a model we already know doesn't work.