Note that the last survey made it explicitly clear that the question was “what is the year such that P(Singularity before year|Singularity ever) = P(Singularity after year|Singularity ever) = 0.5”, whereas in the previous surveys it was ambiguous between that and “P(Singularity before year) = P(Singularity after year) + P(no Singularity ever) = 0.5”.
Note that the last survey made it explicitly clear that the question was “what is the year such that P(Singularity before year|Singularity ever) = P(Singularity after year|Singularity ever) = 0.5”, whereas in the previous surveys it was ambiguous between that and “P(Singularity before year) = P(Singularity after year) + P(no Singularity ever) = 0.5”.