Not a bad idea, especially for things like airport/public surveillance auto-police dispatch ‘suspicious behavior’ algorithms, or justice system ‘reoffender risk’ algorithms.
One problem you’ll run into is that most ‘algorithmic decisionmaking’ isn’t performed using a discrete algorithm, but is a ‘trained’ model based off various ‘machine learning’ processes like bayesian filtering applied to huge datasets, essentially a ‘black box’ that is indecipherable and incomprehensible for humans.
Most people think of algorithms like an algebraic formula where various factors can be included or excluded, or made more or less important. The actual ‘formula’ would look like a multilayered graph with trillions of connected lines, each of which is individually weighted based on trillions of trial and error simulations that are not made from any comprehension or reasoned logic.
It would be like asking a dog handler to show how the dog smells the drugs. No one can tell you. They can describe the training, or some science about smell, but they can’t explain it, they can only demonstrate that it delivers measurable outcomes.
Then you get into Heisenberg uncertainty principle territory, where you are forced to ask how is the outcome measurement defining the very meaning of ‘fact’ being measured, and left with no possible way to get an answer.
Not a bad idea, especially for things like airport/public surveillance auto-police dispatch ‘suspicious behavior’ algorithms, or justice system ‘reoffender risk’ algorithms.
One problem you’ll run into is that most ‘algorithmic decisionmaking’ isn’t performed using a discrete algorithm, but is a ‘trained’ model based off various ‘machine learning’ processes like bayesian filtering applied to huge datasets, essentially a ‘black box’ that is indecipherable and incomprehensible for humans.
Most people think of algorithms like an algebraic formula where various factors can be included or excluded, or made more or less important. The actual ‘formula’ would look like a multilayered graph with trillions of connected lines, each of which is individually weighted based on trillions of trial and error simulations that are not made from any comprehension or reasoned logic.
It would be like asking a dog handler to show how the dog smells the drugs. No one can tell you. They can describe the training, or some science about smell, but they can’t explain it, they can only demonstrate that it delivers measurable outcomes.
Then you get into Heisenberg uncertainty principle territory, where you are forced to ask how is the outcome measurement defining the very meaning of ‘fact’ being measured, and left with no possible way to get an answer.