To paraphrase George Box, "All models are wrong, but some are useful (in making predictions)".
Check out Milton Friedman's Methodology on Positive Economics which tackles this question head on.
Friedman's key point is that the test for a science is not whether the assumptions are true (i.e. home economicus, firms are price takers, information is instantaneous diffuse and homogenous, etc.) but whether if we act as if the assumptions are true can we make testable predictions?
More complex economic models certainly exist where certain assumptions are relaxed (for example, Krugman's international trade theory, or industrial organization, or Stiglitz's research on asymmetric information). One could set about developing ever elaborate models of the world, but more complex models are usually not covered at introductory economics courses (also you get diminishing marginal explanatory power after a couple of factors anyway).
Another argument is that several informed marginal buyers and sellers may be sufficient to drive markets towards efficiency. For example, a lay person may not know how to value an option, but the interaction of sophisticated market markers, buyers, and hedgers act to produce a schedule of option prices that are efficient for someone looking to hedge their portfolio. A more prosaic example is that the competition for the discriminating marginal buyer of laundry detergent, cell phone, etc. products products that increase the consumer surplus of non-discriminating agents thru competition.
There's also a deeper philosophy of science argument (see Karl Popper's theory of falsifiability) that builds on Friedman's positivist view of economics. The test for a model (which are necessarily abstractions) is not in the truth or falsity of the assumptions, but in the accuracy of the prediction. A science is only a science if it has hypotheses that can be falsified.
Indeed in the realm of physics there are quantities that are not detected, so much as measured. At the limits of science, we cannot even ask "do quarks exist" we can only test the standard model by asking whether the measurements predicted by the assumption of quarks are borne out.
There's an apocryphal story about Richard Feynmann when asked what the Cophenhagen many-worlds intepretation means to him where he replies "Shut up and calculate!" i.e. the "reality" of quantum mechanics (whatever that refers to) should not be interpreted. What we have are models whose empirical predictions can be tested.
For example, you can get to the moon and back using Newtonian physics although the assumptions are false (time is not constant for all observers). You can use general relativity to make GPS satellites although the assumptions are false (there are more dimensions than space and time, and energy is discrete). The most powerful predictive theory of modern science is Quantum Electrodynamics. However, QM posits non-locality ("spooky action at a distance"), wave-particle duality, and other paradoxes that exist only if we make the fallacy of reification.