As a strict determinist (based on a strict identity-theory of causation), I
agree with Laplace. Information can be rearranged, but not lost from
existence. Each particle, wave, or whatever, can be regarded as a "packet"
of information. To truly *lose* information, the law of conservation of
matter/energy would have to be violated. Obviously, if any indeterminist
theory of Quantum Mechanics is true, then information is lost all the time,
and it could not in principle be recovered. Put another way, if time were
reversed, there'd be zillions of points where time wouldn't "know" what to
do next, because the information that the current state of affairs
represented would be different from the information that the immediately
prior state of affairs had originally contained/represented.
Obviously, the Laplace view of prediction and "postdiction" depends on
absolutely perfect information about the whatever state of affairs one is
going to use for one's starting point, and, of course, it assumes that the
attempt to gather such information would not *change* the information that
was in fact present and thus invalidate the predictions based on it.
As to Laplace's dream, yes, it's a pipe dream even though whatever exists is
strictly deterministic. Perfect information about the information of
Existence is not possible, and, as complexity theory points out, even small
errors can rapidly yield grossly incorrect predictions. Even errors of one
part in trillions would quickly produce measurably false predictions in many
cases (such as the prediction of the position and speed of a molecule in a
hot and fairly dense gas). I think even Laplace knew that it would not be
possible in fact to carry out such a project; it was purely a
thought-experiment, a way of illustrating or exemplifying his views.
Further, the information about the system to be predicted must be kept away
from the system itself. If it "leaks" into that system from the system doing
the predicting, it will ruin the prediction by introducing factors not taken
into account in the original process of generating the prediction. This is
just *one* reason why one cannot generate strictly accurate and detailed
predictions about oneself.
If you can get a copy of Frederick W. Kantor's *Information Mechanics,"
you'll likely find it quite interesting. I don't know if it's still in
print, but it might be available in some libraries, in print or not.