I cant figure out how to mention Charles Bennett (of IBM Aladan),
who was one of the most important figures in this area (and
is also at IBM).  The problem is that most of his research
deals with Quantum Information.  For example he came up with
a very insightfull and amusing way of relating information to
energy that plays on the solution to the old paradox of
Maxwells demon:

    Imagine a train of boxes each containing a single molecule.
    Each box holds one qubit of information; if the molecule
    is on the right half of the box it represents a 0 if it's
    on the left half it represents a 1.

    If we know the value of the qubit, we can slide a divider
    down the middle of the box and let the pressure of the
    molecule push the divider like a piston over tward the side
    that's a vaccum.  Doing so extracts energy from the box
    and erases the qubit.

    In this way Bennett's engine took information as fuel
    to produce work.

Personally, I think this example is much easier to understand than
the classical information/entropy connection invented by Shannon
and Von Neuman.

------------------------------------------------------------------------
[alternate]

Imagine a race.  The goal is to simulate a particle accelerator:
n particles go in, what comes out?  To make things fair, we'll
allow the contestants a reasonable margine of error.  The winner
will be the contestant who's solution scales best.

As turns out, modern computers can only solve this problem by using
O(exp(n)) resources.  But Nature is able to perform this "simulation"
using only O(n) particles.

Untill Richard Feynman made this startling discovery, it was widely
accepted that the model for modern computation (the Turing machine),
was the most powerful problem solving device in theory.  But if
Nature was able to beat the computer O(n) to O(exp(n)), it was
proof positive that a fundementally more powerful type of computer
was posible.