I wonder if the double blind model can be applied right into the publishing stage? Or perhaps journals could have some requirement, 50/50 positive/negative results.
I think that the problem is we consider a result as either "positive" or "negative". All correctly preformed experiments add a positive ammount of knowledge. Humans tend to classify unexpected results as negative, but that is not necessary.
The problem then is that humans don't know how to use definitions. In a specific study, a result can be negative, but in the general view, it would be positive because it contributes to the body of science.
That's not what negative means here. For these studies the conclusion is a statistical test -- accept or reject the null hypothesis. If a drug doesn't give results significantly different from a placebo, that's "negative". More to the point, when a researcher collects a pile of data, does some analysis, and can't see a damn thing either way, then either the experiment was flawed to begin with, or that absence of meaning is a real, useful result. Maybe the sample size just wasn't large enough to show a small but significant trend. It takes a fair amount of confidence in one's skills to rule out the first possibility and announce loudly that there's nothing to be seen further down the path you took.