Reminder! Although I haven’t yet written abuot the general principle, the original Drake’s Equation was bullshit. Things like this are even more bullshit since they exploit the human bias of assigning significant probabilities to everything elicited creating an unpacking bias where unpacked items are assigned much larger summed probabilities than the corresponding packed categories, meaning that the apparent probability of a conjunction goes down as you helpfully break it into more and more parts. By these means I could equally make the Moon landing appear impossible, just as I could make cryonics appear more and more likely by considering more and more disjunctive pathways to success. It also fails as probability theory because conditional dependency.
Again, general reminder: Across all cases not backed up by actual sampling, someone who offers to helpfully “elicit” a set of “conjunctive” probabilities and multiplies them together to get some low number, without considering any disjunctions, assuming conditional independence, and with no warnings about unpacking bias, is using a Fully General Counterargument that will underestimate the probability of anything. I have yet to see a good Breaking X Down for any X, unless X is a whole population (not a significant subsector of it) and the breakdown is just the actual data about X.
By these means I could equally make the Moon landing appear impossible
Viewed from which historical time?
just as I could make cryonics appear more and more likely by considering more and more disjunctive pathways to success.
This is not the first time you claim that, but AFAIK you never did. I’m skeptical that this is possible.
It also fails as probability theory because conditional dependency.
Unless you propose a plausible mechanism for two variables to be correlated, it is reasonable to assume that they are approximately independent, (Occam’s razor, principle of maximum entropy, etc.). Also, correlations can be positive or negative.
I understand the concern about unpacking bias, and read about a related experiment also by Kahneman (I think) who elicited a higher probability when he asked experts to estimate the likelihood a specific scenario (deflation of the rouble leads to a Soviet invasion of Germany and nuclear war) than a general scenario (nuclear war). So I would be cautious of handling an equation with multiple, obviously overlapping terms. I’ll update the original post when I’m back at a computer to include a health warning in the first paragraph.
I don’t think I fully understand the criticism of this piece though; are you saying the modelling approach is incoherent or simply cautioning people not to just plug it into the cryo-Drake equation without considering the unpacking bias?
I feel like the unpacking/packing biases ought to be something that should be easier to get around than some other biases. Fermi estimates do work (to some extent). I somewhat wonder if perhaps giving log probabilities would help more.
Reminder! Although I haven’t yet written abuot the general principle, the original Drake’s Equation was bullshit. Things like this are even more bullshit since they exploit the human bias of assigning significant probabilities to everything elicited creating an unpacking bias where unpacked items are assigned much larger summed probabilities than the corresponding packed categories, meaning that the apparent probability of a conjunction goes down as you helpfully break it into more and more parts. By these means I could equally make the Moon landing appear impossible, just as I could make cryonics appear more and more likely by considering more and more disjunctive pathways to success. It also fails as probability theory because conditional dependency.
Again, general reminder: Across all cases not backed up by actual sampling, someone who offers to helpfully “elicit” a set of “conjunctive” probabilities and multiplies them together to get some low number, without considering any disjunctions, assuming conditional independence, and with no warnings about unpacking bias, is using a Fully General Counterargument that will underestimate the probability of anything. I have yet to see a good Breaking X Down for any X, unless X is a whole population (not a significant subsector of it) and the breakdown is just the actual data about X.
Viewed from which historical time?
This is not the first time you claim that, but AFAIK you never did. I’m skeptical that this is possible.
Unless you propose a plausible mechanism for two variables to be correlated, it is reasonable to assume that they are approximately independent, (Occam’s razor, principle of maximum entropy, etc.). Also, correlations can be positive or negative.
I understand the concern about unpacking bias, and read about a related experiment also by Kahneman (I think) who elicited a higher probability when he asked experts to estimate the likelihood a specific scenario (deflation of the rouble leads to a Soviet invasion of Germany and nuclear war) than a general scenario (nuclear war). So I would be cautious of handling an equation with multiple, obviously overlapping terms. I’ll update the original post when I’m back at a computer to include a health warning in the first paragraph.
I don’t think I fully understand the criticism of this piece though; are you saying the modelling approach is incoherent or simply cautioning people not to just plug it into the cryo-Drake equation without considering the unpacking bias?
You might be thinking of an earlier discussion of this issue involving car failure diagnostics: http://lesswrong.com/lw/fz9/more_cryonics_probability_estimates/82oh?context=1#82oh
I feel like the unpacking/packing biases ought to be something that should be easier to get around than some other biases. Fermi estimates do work (to some extent). I somewhat wonder if perhaps giving log probabilities would help more.