Behe's entire argument of Intelligent Design rests on the argument of irreducible complexity (IC). Behe describes a bio-chemical 'system' to be IC if it is "a single system composed of several well-matched, interacting parts that contribute to the basic function, wherein the removal of any one of the parts causes the system to effectively cease functioning".
Further, and to his argument for ID, he states that these interacting parts have no function on their own, and thus would not be 'selected' via evolution. Consequently, the IC system would have to evolve in toto, and the probability of such a complete system coming into existence via natural selection would be so small as to be effectively zero. Ergo, the only way such a system can exist is via the intervention of a designer.
This hypothesis is indeed both testable and predictive. For testability, one has to observe none of the component pieces of an IC system anywhere else. For all the given examples of IC systems (eyes, bacteria flagellum, bombadier beetles, blood clotting protein cascade, kreb's cycle, etc.), the test has failed. 'Sub-systems' for all the above are observed elsewhere in nature, though their function is completely different from the IC system in question. For example, in the bacteria flagellum, a good portion of the structure is found, intact, in the Type III Secretory System. The TTSS is used by some pathogenic bacteria to inject protein toxins into host cells. Quite a different function from the molecular motor base of the flagellum, but there it is. Additional proteins involved in the flagellum are also found elsewhere, and thus the hypothesis of IC is proved false:
Component pieces of an IC system do indeed arise independently in organisms via evolution. Which means the examples put forward are not irreducibly complex, by definition. And therefore natural selection 'explains' the examples and there is no need for a designer.
Conversely, IC predicts that since components parts of an IC system are only involved with the IC system, they should not be able to be found outside the IC system. The above paragraph shows this to be false, and thus IC fails as a hypothesis via predictability.
Similar situations have been presented for every system put forward by Behe as an IC system. The fatal flaw in the hypothesis is the assumption that the components of the IC system can have no other function than that performed within the IC system. The crystallin protein in the lens of the human eye, used to help focus light, starts out embryonically as a liver protein!
Additionally, scientists have also shown there are straightforward paths of evolution for these IC systems, and thus ID is not the only way these systems could arise, as stated by Behe.
Additionally, scientists have pointed out other similar systems to Behe's examples which perform similar functions, but are comprised of simpler components, thus negating the functional irreducibly complex aspects of his examples.