Preference Falsification is likely the most important, under used theory in economics. I suspect these models will fail miserably because they will always have bad data until we stop silencing opinions we do not agree on.
Let's have constructive conversations please, please, please.
Huge amounts of economic data come from inherently boring, unbiased sources (employment records, tax records, exports, manufacturing output, etc...). What data do you think is being falsified? You seem to be referring to survey-type data.
Claiming any data as unbiased seems illogical to me, note how they are constantly revised.
Yes, people's response to surveys are one. Their reaction to policy is another. If this "ai" is modelling policy decisions and outcomes, what data is being feed in?
Further, how does one explain said "ai" recommendations and rational to the public?
(edit)
After looking over the code, I find this model incredibly simplistic and not capable of nearing the complexity IRL.
People's reactions to policy decisions/outcomes don't matter nearly as much as the actual outcomes. I think most people will agree that simultaneously raising GDP per capita while reducing inequality and increasing innovation would be a good outcome. Those are all things that can be quantified and measured with very little bias. Survey data isn't as important comparatively.
As for the model explaining itself: the model can show relationships and patterns that humans hadn't previously detected, because there were too many variables. We can deduce new economic strategies by observing model output, such that we can even carry out those strategies without further use of the model. There are also ways to make ML models explain themselves.
6 comments
[ 7.1 ms ] story [ 23.6 ms ] threadPreference Falsification is likely the most important, under used theory in economics. I suspect these models will fail miserably because they will always have bad data until we stop silencing opinions we do not agree on.
Let's have constructive conversations please, please, please.
Yes, people's response to surveys are one. Their reaction to policy is another. If this "ai" is modelling policy decisions and outcomes, what data is being feed in?
Further, how does one explain said "ai" recommendations and rational to the public?
(edit)
After looking over the code, I find this model incredibly simplistic and not capable of nearing the complexity IRL.
As for the model explaining itself: the model can show relationships and patterns that humans hadn't previously detected, because there were too many variables. We can deduce new economic strategies by observing model output, such that we can even carry out those strategies without further use of the model. There are also ways to make ML models explain themselves.