There is one sort of information almost everyone leaves behind them, namely the memories of people who knew that person when she was alive. By definition, there is no information theoretic bound preventing reconstructing a person with such precision that everyone who knew her won't be able to tell the difference. Similarly, there is no information theoretic bound preventing reconstructing a person that writes novels which the original version could have written, as far as the readers can tell. So, if we launch FAI tomorrow we can get a version of Pratchett sufficiently close to the original to "fool" his wife, daughter and fans. On the other hand, there might be stronger complexity theoretic bounds: reconstructing a person to fit a given set of "output" is a problem which is a priori only known to be in NP.
One of the themes of current scientific progress is getting more and more information out of tiny amounts of data. Who'd have thought that we could learn so much of distant and recent biological history from DNA, and so much about distant planets, stars, galaxies, and the cosmos from tiny differences in very small amounts of light?
Pratchett's death puts an extra edge on the question-- to what extent can people be re-created from what they've left behind them, especially if they've written novels which include a lot of their personality?
Any thoughts about theoretical limits of how much can be figured out from small amounts of data?