If you look at the results of the last survey, that's exactly what happened, and the mean was far higher than the median (which was reported along with the standard deviation). I agree, it would have been a big improvement to specify which sense was meant.

Also, answering year such that P(<singularity by year X> | <singularity occurs>) would be the best way to get a distribution of answers on when it is expected. So that's what I did. If you interpret the question the other way, then anyone with a 30-49.9999% chance of no singularity, has to put a date that is quite far from where most of their probability mass for when it occurs lies.

Suppose I believe that there is a .03% probability of a singularity for each of the next 1000 years, and then decaying by 1/2 every thousand years after that. That puts my total singularlty probability in the 52% range, with about half of my probability mass concentrated in the next 1000 years. But to answer this question literally, the date I'd have to give would be around 7000AD, even though I would think it was about as likely to happen by 3011AD as after 3011AD.