Reading this interesting and timely anecdote about perfect algorithms and hard-boiled eggs. It serves as a cautionary tale about striving for perfection where the domain within which such perfection is meant to operate is both imperfect and unconfined. Welfare algorithms have made news recently for disproportionately hurting the very people it is meant to serve being optimized to prevent fraud. This is a lot like the hard-boiled egg situation. There is a certain baseline expectation the algorithm operates on and variances are flagged as problematic. A real person's life in a difficult time does not follow any pre-defined pattern.
Every hour and day presents new and novel challenges that are unique to them - not even something that can be extrapolated to other in similar financial dire straits. So they need to respond and react to what is coming at them while somehow keeping their head above the water. Chances are a lot of those behaviors will look flaggable to the algorithm because it deviates from what it expects. Would it help atleast a bit to have the people such algorithm is meant to serve be the one to help define it - maybe they are the closest to the range of possible outcomes in any given scenario. Might now people expecting their three-minute eggs are in receipt of hard-boiled ones. Not much has changed.
Comments