Rational utility not working correctly

[tklebel]

The process for how agents assess their decisions and adapt effort levels is currently not working as intended. Although there clearly is a cost to sharing, agents initially share, until all share, and then they stop sharing altogether again. There is no mechanism on the funding side driving this (not yet implemented), this is purely the agents assessing their own success (resources) and investments (effort).

The bug is likely in the flow of events regarding recording the current and past success and effort, and when to compare these.

[tklebel]

The initial problem was fixed by strongly simplifying: if resources are lower, reduce effort, if higher, increase effort. However, with the introduction of the redistribution of resources after sharing data in 6bbbe962bc953caf347ce6270ad72fccbc6fb985, it seems that this is too simple: agents increase effort since they are getting some resources from others.

Maybe the solution is to tie their resource evaluation to the action from the last round (increased or decreased effort)

[tklebel]

The current issue lies in the interaction of the decision rule and the redistribution role. Only considering redistribution to agents themselves (only citation benefit, no benefit to other teams): When agents share data, they lose resources, but gain in subsequent ticks, since they get dividends from sharing data. This stabilises at effort = .5, since there the probability to share data is 50% -> on average, in one try they share data, lose resources, and lower effort. However, in the next round, they don't share data, gain resources (through dividends), and raise effort.

Unclear atm how to amend this. One way would be for the agents only to increase effort if the resource gain is more substantial. Another one would be making them more sensitive to resource losses, leading to a stronger decrease in effort than increases, thus making an asymmetric update function. I prefer the first, but both don't seem principled by a bit hacky. The third option would be to disregard this redistribution mechanism altogether, but then the model is very simple and assumes away issues of reuse and competition.

Another way would be to make them more dispositioned towards free-riding: whenever they receive resources from somewhere else they lower their own effort. But not sure how that makes sense and how to motivate it.

[tklebel]

@federico-bianchi it would be great if you could provide feedback on this. Also happy to provide more detail. The screenshot below shows the settings: basically everything turned on. Note that there is no incentive mechanism for data sharing right now. So given that there are costs, groups should stop sharing data, but as explained above, they settle at .5 effort. If the redistribution mechanism is turned off, they quickly stop sharing, as expected.

[tklebel]

@federico-bianchi I actually was able to fix the issue, by changing the utility updating function.

The current model has interesting behaviours: for relatively high settings of initial resources and initial effort, combined with a large originator-benefit (which is a little unrealistic, i.e., many people citing the data, but no one using it), some teams end up sharing data continuously while most don't (separated equilibrium). In other cases with lower initial settings, all groups simply stop sharing data.

So it will be crucial to find some reasonable parameters there. I believe, once we add interventions on the funding side, this will deliver interesting results.