Before predictive deduction was formally defined, there were only three defensible decision-making tools: tradition/habit, experimentation, and intuition. Each has at least one major theoretical weakness.

    To follow tradition/habit is to pick choices that worked for you or someone else previously. Choices that clearly failed don't persevere as traditions or habits. Therefore, lasting traditions/habits are usually no worse than choices made previously. Furthermore, assuming that we recognize our mistakes, we'll eventually improve our traditions/habits, through trial and error. However, as argued above, this assumption is unreliable. The fact that societies progress so slowly suggests that people fail to realize when their traditions/habits are less than optimal. As a result, traditions/habits theoretically lead people to make the same old mistakes over and over.

    An alternative to tradition/habit is experimentation. The use of experimentation as a decision-making tool involves "inductively" inferring from a series of measurements what the future consequences of potential choices would be. The flaw in this procedure is called the "problem of Hume" (after David Hume and his 1739 A Treatise of Human Nature). Today, all logicians agree that the conclusions yielded by this kind of inference, called "ampliative" or "incomplete" induction, don't necessarily follow from the premises. In other words, even if experimentation is performed flawlessly, it may yield inaccurate predictions. Especially regarding clearly chaotic systems (such as social systems), the best experimenters never completely trust their conclusions.

    Instead of trying to argue that experimentation is reliable, its defenders offer the following lesser arguments to justify its use:

1. Experimentation has a good track record (in the natural sciences),
2. The reliability of experimentation improves as we gather more measurements,
3. Although predictions inferred from experiments might not be perfectly accurate, their probability of accuracy is better than chance, and
4. No other method can do better than experimentation.

    The second justification (suggested by C. S. Pierce) only holds true for the study of asymptotic and periodic systems. Continuing experiments on such systems typically suggest smaller and smaller corrections to our theories, so there is a sense in which series of experiments converge on truth. With chaotic systems, however, to experimentally confirm that a theory is even close to characterizing a system would require taking measurements over an infinite span of time, since radically diverse chaotic systems can appear indistinguishable for any finite length of time before diverging. We can't afford to spend infinite amounts of time making decisions, so it isn't practically feasible to converge on chaotic systems the way we would non-chaotic systems.

    If experimentation were the only alternative to random decision-making, then the third argument, the valid claim that experiments have better-than-chance probability of yielding accurate predictions, would justify their continued use. However, there are other alternatives. Therefore, mere comparison to random decision-making is insufficient. We must consider all other potential alternatives as well.

    That  leaves only the fourth justification, the argument that it's impossible to do better than experimentation does. It will, of course, take this entire document to completely defend my counterexample to that claim. For now, let me just punch some holes in the standard argument that predictive deduction is impossible. The argument goes as follows:

"Deduction requires beginning with general facts, and general facts can't be gleaned from limited experience. Since human knowledge shall always come from limited experience, humans will never be able to know the kinds of facts they need to perform deduction reliably."

    The third defensible decision-making tool (along with tradition/habit and experimentation) was intuition. Our lack of understanding about intuition makes it both difficult to trust and difficult to prove that it can't be perfected. The phenomena we call "common-sense", "sensibility", "sixth-sense", "inspiration", "gut-feeling" and "business instinct" (etc.), may all be the same thing. Likewise, what we call "intuition" may actually be more than one phenomena. Each instance of "intuition" could involve any of a number of thought processes, some reliable and others not. I'll try to use this fact to prove that all subconscious reasoning, including intuition, must always be somewhat unreliable.

    At the same time, it seems likely that at least one of the procedures we classify as "intuition" would be a subconscious application of predictive deduction. If so, and if predictive deduction works, then intuition may be better than chance, even regarding cases that would foil tradition and experimentation. Since that appears to be the case, the limited success of intuition (unreliable, but better than chance) provides an extra reason to trust predictive deduction; if we deny that predictive deduction works, then some of the successes of intuition would be unexplained.