Prerequisite / Read this first: Scientific Evidence, Legal Evidence, Rational Evidence
Consider the statement "It is physically possible to construct diamondoid nanomachines which repair biological cells." Some people will tell you that molecular nanotechnology is "pseudoscience" because it has not been verified by experiment - no one has ever seen a nanofactory, so how can believing in their possibility be scientific?
Drexler, I think, would reply that his extrapolations of diamondoid nanomachines are based on standard physics, which is to say, scientific generalizations. Therefore, if you say that nanomachines cannot work, you must be inventing new physics. Or to put it more sharply: If you say that a simulation of a molecular gear is inaccurate, if you claim that atoms thus configured would behave differently from depicted, then either you know a flaw in the simulation algorithm or you're inventing your own laws of physics.
My own sympathies, I confess, are with Drexler. And not just because you could apply the same argument of "I've never seen it, therefore it can't happen" to my own field of Artificial Intelligence.
What about the Wright Brothers' attempt to build a non-biological heavier-than-air powered flying machine? Was that "pseudoscience"? No one had ever seen one before. Wasn't "all flying machines crash" a generalization true over all previous observations? Wouldn't it be scientific to extend this generalization to predict future experiments?
"Flying machines crash" is a qualitative, imprecise, verbal, surface-level generalization. If you have a quantitative theory of aerodynamics which can calculate precisely how previous flying machines crashed, that same theory of aerodynamics would predict the Wright Flyer will fly (and how high, at what speed). Deep quantitative generalizations take strict precedence over verbal surface generalizations. Only deep laws possess the absolute universality and stability of physics. Perhaps there are no new quarks under the Sun, but on higher levels of organization, new things happen all the time.
"No one has ever seen a non-biological nanomachine" is a verbalish surface-level generalization, which can hardly overrule the precise physical models used to simulate a molecular gear.
And yet... I still would not say that "It's possible to construct a nanofactory" is a scientific belief. This belief will not become scientific until someone actually constructs a nanofactory. Just because something is the best extrapolation from present generalizations, doesn't make it true. We have not done an atom-by-atom calculation for the synthesis and behavior of an entire nanofactory; the argument for nanofactories is based on qualitative, abstract reasoning. Such reasoning, even from the best available current theories, sometimes goes wrong. Not always, but sometimes.
The argument for "it's possible to construct a nanofactory" is based on the protected belief pool of science. But it does not, itself, meet the special strong standards required to ceremonially add a belief to the protected belief pool.
Yet if, on a whim, you decide to make a strong positive assertion that nanomachines are impossible, you are being irrational. You are even being "unscientific". An ungrounded whimsical assertion that tomorrow the Sun will not rise is "unscientific", because you have needlessly contradicted the best extrapolation from current scientific knowledge.
In the nanotechnology debate, we see once again the severe folly of thinking that everything which is not science is pseudoscience - as if Nature is prohibited from containing any truths except those already verified by surface observations of scientific experiments. It is a fallacy of the excluded middle.
Of course you could try to criticize the feasibility of diamondoid nanotechnology from within the known laws of physics. That could be argued. It wouldn't have the just plain silly quality of "Nanotech is pseudoscience because no one's ever seen a nanotech." Drexler used qualitative, abstract reasoning from known science; perhaps his argument has a hidden flaw according to known science.
For now, "diamondoid nanosystems are possible" is merely a best guess. It is merely based on qualitative, abstract, approximate, potentially fallible reasoning from beliefs already in the protected belief pool of science. Such a guess is not reliable enough itself to be added to the protected belief pool. It is merely rational.