Here’s an interesting argument I have been thinking about, and I’d like to pursue a train of thought which comes from the consideration of it:
1) If Naturalism is true, then Determinism is true.
2) Naturalism is true.
3) Therefore, Determinism is true.
A follow up argument looks like this:
4) If Determinism is true, then no experiment is repeatable.
5) Determinism is true.
6) Therefore no experiment is repeatable.
7) If no experiment is repeatable, then inductive reasoning in the Natural Sciences is mistaken.
8) No experiment is repeatable.
9) Therefore, inductive reasoning in the Natural Sciences is mistaken.
What does this argument show? Well, it undercuts Empiricism if one is a Naturalist.
The first premise is perhaps controversial, since some Naturalists may want to appeal to some measure of indeterminism, thanks in part to quantum mechanics. However, the kind of randomness which quantum mechanics is purported to introduce doesn’t help one escape the argument, for in the case that things are random they are just as challenging for induction as if they were completely determined. Now, if things are completely determined, why would that be a problem for induction? Well, quite simply this: If things are determined, then every experiment performed has a determined result. If one runs some experiment x five or so times, and finds that four out of five times the experiment yields result P whereas one out of five times it yields ~P, then one generalizes from that experimental data to a theory. However, this process presumes that the experiment was performed under similar or relevantly similar conditions. This, it seems, is never true on determinism, since every single experiment yields its own result in a completely determined way. Even if 5/5 experiments had yield result P the experimental data just cannot be legitimately extrapolated to a generalized theory quantifying over all ‘relevantly similar’ states. Instead, each experiment is done in such a way that it’s particular result is only attributable to that particular experiment, but not to other experiments which seem similar, but are themselves completely physically determined to yield some result. It should be clear that what I mean by ‘repeatable’ in the above argument is simply that one can run the experiment under ‘relevantly similar’ conditions. However, there is no guarantee, nor justification for believing (at least on Naturalism – and at least on the face of it) that the conditions are relevantly similar.
Induction presumes normativity, but there is nothing normative about a succession of events which are completely determined. How is one to know, for instance, when one makes a machine flip a coin with a precise amount of force, at a particular angle, that just because it has always landed ‘heads’ on all previous occasions, it will do so again? One presumes that the experiment is regulated by something normative, and any discrepancy (if it landed on ‘tails’) would be accounted for by some unaccounted for feature of the environment. However, we do not ever have exhaustive knowledge of the environment in which an experiment is done. Suppose that what caused the coin to flip in such a way that it landed ‘heads’ was the proximity of the moon to the earth at the set of times t0 – t5 where the experiments were all run at or after t0 and at or before t5. One might suppose that given continued experiments after t5, one would eventually have enough experimental data to figure out that something in the environment affected the results from t0-t5. However, the problem is that all the experiments after t5 are effected by their environments as well, and there is no guarantee that experiments from t6-t15 are not affected by the explosion of several stars in a neighbouring galaxy in such a way that the experiments are all useless insofar as having accurate predictive power for a period t16-tn where the conditions change again.
The problem is precisely that no event is not causally determined by its entire environment, and not only can we not ensure that any two experiments share the same environment, we also know that (at least if we accept what science tells us about the universe) no two events ever share the same situation, since the universe is physical, and nothing physical is ever in a quiescent state, therefore the state of the universe is never ‘identical’ to a previous state (of course, somebody might want to argue that being in a non-quiescent state doesn’t logically preclude the universe being in an identical state to some previous state, but the objection isn’t germane if one accepts that the universe is in a state of expansion called ‘escape velocity’, precluding bounce-back).
One objection to this line of reasoning might go something like this: perhaps this is true, but it is also true that any states across t0-tn are also relevantly similar insofar as the experiments under question are concerned, so that extrapolating the results to a generalized theory will provide accurate predictive projections. The problem is that if one accepts that events are causally determined by their entire environment, one has to show that the environment is not only relevantly similar from t0-tn, but also from tn+1 to tn+x (where tn+1 to tn+x covers the period for which the predictive power of the theory is intended to hold). It seems, however, that this cannot be done. Perhaps one could argue that if it were not, then any theory would not have predictive power, therefore any theory demonstrated to have predictive power is evidentially corroborated. In other words, we know some theory is reliable not only because of inductive reasoning applied to some set of experiments, but also because it projects predictions which we have observed to be accurate. The problem is, of course, that even this ‘predictive constraint’ on theories is as tenuous as inductive reasoning was in the first place, and for the same reason: we just are not at an epistemic vantage point from which we can have sufficient information about the environment of the universe to place overwhelming confidence in any scientific theory.
However, once a Naturalist see’s all this, they must decide what other/better option (other than empiricism) there is. Surely a Naturalist will not want to be a Rationalist. Indeed, there is reason to believe that a Naturalist cannot be a Rationalist (as it is no coincidence that all Rationalists in philosophy have been Theists). Since there is no other option, a Naturalist will have to adopt Empiricism with tenuousness. However, I wonder then if a Naturalist can ever claim scientific knowledge about anything, since knowledge requires the triad of ‘truth, belief, & justification’ and epistemic justification seems a difficult constraint to satisfy.
Interestingly, I think the Naturalist in this position may want to argue for an epistemological position called “Contextualism” which goes something like this:
David Lewis’s contextualism takes off from the intuitive infallibilist claim that one knows that p just in case one’s evidence eliminates all possibilities in which not-p. This does not lead to skepticism (in all contexts), however, because the scope of “all” is taken to be contextually restricted. Thus, Lewis adds to this biconditional, soto voce, ” -Psst! – except for those possibilities that we are properly ignoring.”
~ Ernest Sosa, Jaegwon Kim, Jeremy Fantl, and Matthew McGrath “Epistemology: An Anthology” (second edition). Introduction to Part VIII, p.664
If nothing else, this argument at least stifles any naive scientistic claim to knowledge in which we can place all our confidence – it demonstrates that the results of the empirical method are always extremely tenuous. Moreover, it becomes evident that the Theist may be able to do better with empiricism. The Theist, even if she believes in strict determinism, is able to say that human beings are generally placed by God in an epistemic vantage point from which induction will reliably lead to knowledge concerning the created order. Notice that the connection between Theism and induction is stronger than people often recognize, precisely because Theism entails inductive-reliabilism and justifies the inductive method (something Hume and Humeans cannot do except pragmatically). It does so because all language about ‘God’ is analogous, and is therefore based on univocal/equivocal language about the world, much of which is tempered by inductive reasoning, thus even to say “God exists”, if it is a true proposition, entails that the language used is legitimate, and this language is predicated analogously, ergo etc.