I find this stuff really fascinating, too, and spent part of high school trying to get a teacher to support me in an attempt to write a program that could identify an author by their style (no luck, sadly).
This problem is reminds me a lot of determining whether or not a sequence of number is random. Please pardon the obligatory Dilbert comic:
Any attempt to read a sequence of actions and determine if it was generated by a specific person is going to have to be probabilistic, just like a test for randomness. It seems to me like the trick is accurately calculating that probability. Take identifying someone by their typing style, for instance. We can ask someone to type some passage of text and measure the accuracy and time between keystrokes to try to identify a user. But users will vary, and it is almost certain that in a large enough pool of people there will be two whose variations overlap somewhat. The users won't be identical, but there will exist certain output sequences that will be plausible for either user. Then the trick is determining which user is more likely.
I feel that this is a problem that humans may be a lot better than computers at.
Another point of interest: people change over time, so the authentication will have to change as well. Skills improve or deteriorate. If you ask a user to type a specific passage to identify themselves a lot, they will get better at typing that passage, and maybe at typing in general. When I was researching identifying authors by their writing styles, I found out that authors change style a lot over the course of a lifetime, to the point that an author's early work and later work may be less similar than some different authors are.
On a practical level, how would you maintain the authentication scheme in the face of changing skills? On a philosophical level, if a person's skill changes so much so that they no longer authenticate, is the authentication test right? Are they a different person?
This would make a great discussion over a bottle of wine some day
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
branchandroot
no subject
Date: 2008-12-09 06:01 am (UTC)This problem is reminds me a lot of determining whether or not a sequence of number is random. Please pardon the obligatory Dilbert comic:
Any attempt to read a sequence of actions and determine if it was generated by a specific person is going to have to be probabilistic, just like a test for randomness. It seems to me like the trick is accurately calculating that probability. Take identifying someone by their typing style, for instance. We can ask someone to type some passage of text and measure the accuracy and time between keystrokes to try to identify a user. But users will vary, and it is almost certain that in a large enough pool of people there will be two whose variations overlap somewhat. The users won't be identical, but there will exist certain output sequences that will be plausible for either user. Then the trick is determining which user is more likely.
I feel that this is a problem that humans may be a lot better than computers at.
Another point of interest: people change over time, so the authentication will have to change as well. Skills improve or deteriorate. If you ask a user to type a specific passage to identify themselves a lot, they will get better at typing that passage, and maybe at typing in general. When I was researching identifying authors by their writing styles, I found out that authors change style a lot over the course of a lifetime, to the point that an author's early work and later work may be less similar than some different authors are.
On a practical level, how would you maintain the authentication scheme in the face of changing skills? On a philosophical level, if a person's skill changes so much so that they no longer authenticate, is the authentication test right? Are they a different person?
This would make a great discussion over a bottle of wine some day