[cutemonster]

Auto generate IQ tests, so cannot be learned by trying many times / cannot learn by heart. And have people in the next generation do the same auto generated tests.

This won't be perfect, but, using mathematics, Id think it would be possible to know how (im)precise the comparisons would be (confidence intervals of the differences).

Also, might not work for really bright people (they'd learn how the auto generated tests get generated? They might sort of "disassemble" them and find the answers quickly?)

[darawk]

This doesn't work because of the Flynn effect. Raw IQ scores have been consistently increasing over time. So much so that the average person in the 1920s would be considered mentally challenged today, if you used their raw IQ scores. However, we know that obviously the average person in the 1920s was not mentally challenged in the sense that someone with that same raw score would be today.

This is all a pretty big mystery and suggests we really don't understand this "IQ" thing we are measuring. However, one big consequence is that we definitely cannot meaningfully compare scores across time.