What he means is that we use the scientific facts to build theories,
models, and the science is in the models, not in the single facts.
This distinction between facts and theories, models, has become even more
distinct today, in the age of computers.
We can use computers to gather measurements, for example temperatures for
a thousand locations every ten seconds over a time period of three days.
Those measurements are the facts, the reality. But for them to become
meaningful to us, to make it possible to make predictions about the
future, we need to find a model, a theory which describes why the
temperatures change in those locations like they do over time.
Earlier we had to use human minds to see patterns and tendencies in such
a collection of measurements, and to create formulas which could be
satisfied by these measurements.
Today we have computers which can create formulas from measurements, so
the computers formulate the theories and create new scientific knowledge.
A very simple example of this process is a curve-fitting program, which
can be given a number of coordinates and find a curve, a formula, which
these coordinates fit fairly well in.
A much more complicated example is a weather forecast computer cluster
where predictions are made, and checked against what really happens, and
the theory is refined step by step until the computers can predict the
weather several days ahead with fairly good precision.