3 ms·
One problem is that g is not very well defined. It's just some vague factor. I think it would make sense to define it as something like: the extent someone has
by BenoitEssiambre 5y ago
One problem is that g is not very well defined. It's just some vague factor.
I think it would make sense to define it as something like: the extent someone has well calibrated Bayesian reasoning that exploits information to its maximum to do accurate predictions.
It's hard to test though because there is a dearth of synthetic benchmarks or algorithms to compare to. Humans seem to still be better at performing these types of Bayesian "common sense" tasks than any synthetic procedure, so we can only compare using tests that we subjectively agree other "smart" humans seem to be good at.
- pessimizer 5y ago> the extent someone has well calibrated Bayesian reasoning that exploits information to its maximum to do accurate predictions. The concept of g actually completely rules this interpretation out based on nothing. If g was a matter of learning reasoning skills and applying them, you'd be able to teach it. It wouldn't be some magical race/sex/class number.
- BenoitEssiambre 5y agoUsing calibrated Bayesian reasoning can certainly be an innate trait. Some people seem good at making good predictions and assessing uncertainty of the data and knowledge they have whether they've been thought or not and it seems like an unusually difficult thing to teach. Will you make that much better predictions even if you read all of the lesswrong.com cannon?
- jhanschoo 5y agoI dont think your definition is accurate though. IQ tests like progressive matrices dont test only if you have good probabilistic models but how well you can identify the pattern in the matrices. If the patterns follow a minimum entropy distribution, there is only so much that falls to technique, and the technique is intuitive (prefer simple over complex); there is still the problem of raw computation that is guessing and rejecting.