on substack, a reader commented that interview summaries are often bland and uninsightful. i agree.

summarisation must assume that the underlying information is patterned (summaries focus on the pattern and omit the non-patterned data) and that the data that doesn’t conform to the pattern is sufficiently irrelevant to be discarded from the summary — summaries are a kind of compression. the summary problem is that the really emergent signals are rarely (never?) identifiable as a pattern yet. so they are discarded as noise instead of signal. 

i think even most humans, without training and guidance, would focus more on obvious signal than on emergent signal that appears to be noise. 

to pick up emergent signals, research needs to be designed better:

on substack, a reader commented that interview summaries are often bland and uninsightful. i agree.

summarisation must assume that the underlying information is patterned (summaries focus on the pattern and omit the non-patterned data) and that the data that doesn’t conform to the pattern is sufficiently irrelevant to be discarded from the summary — summaries are a kind of compression. the summary problem is that the really emergent signals are rarely (never?) identifiable as a pattern yet. so they are discarded as noise instead of signal. 

i think even most humans, without training and guidance, would focus more on obvious signal than on emergent signal that appears to be noise. 

to pick up emergent signals, research needs to be designed better:  https://vaughntan.org/betterinterviews