Thanks, this was surprisingly interesting. I agree stylometrics is a relevant field, and this problem might require more awareness than it has got.
On the other hand… My epistemic status here is very low, but I guess that, in the long run, the asymmetry offense v. defense here would likely play against surveillance. It doesn’t seem hard to use writing assistance software to change your “stylometric signature”—even something very amateurish, like translating a text back and forth, renders very different results. I see an analogy here with cheating-detection software… which has been reportedly used by essay mills to ensure their products won’t be detected. The paper you cite (from 2006) suggest this is not particularly technically hard—if I understood it correctly.
What I think might be problematic is that we have conflicting desiderata: if you’re an anonymous blogger, you actually want to have a recognizable and definite style; you could use obfuscating techniques in other interactions (in your academic essays, or personal blog), but now the catch: if you’re the only one doing this, you’re basically signalling to everyone else you have a “secret identity”. But I guess this applies to other matters regarding surveillance as well.
So my point is: though I reckon your concern, I am still way more worried about other forms of surveillance in general, and I think the “crux” here is a mainly social issue, instead of a tech one.
Somewhat related : a while ago I wrote a benefit-risk analysis of research in improving device detection. I concluded the risks likely outweight the benefits.
Thanks, this was surprisingly interesting. I agree stylometrics is a relevant field, and this problem might require more awareness than it has got.
On the other hand… My epistemic status here is very low, but I guess that, in the long run, the asymmetry offense v. defense here would likely play against surveillance. It doesn’t seem hard to use writing assistance software to change your “stylometric signature”—even something very amateurish, like translating a text back and forth, renders very different results.
I see an analogy here with cheating-detection software… which has been reportedly used by essay mills to ensure their products won’t be detected. The paper you cite (from 2006) suggest this is not particularly technically hard—if I understood it correctly.
What I think might be problematic is that we have conflicting desiderata: if you’re an anonymous blogger, you actually want to have a recognizable and definite style; you could use obfuscating techniques in other interactions (in your academic essays, or personal blog), but now the catch: if you’re the only one doing this, you’re basically signalling to everyone else you have a “secret identity”. But I guess this applies to other matters regarding surveillance as well.
So my point is: though I reckon your concern, I am still way more worried about other forms of surveillance in general, and I think the “crux” here is a mainly social issue, instead of a tech one.