Objection 5

``Yes, but what your computists do is not decomposable into smaller, purely mechanical steps, which is the hallmark of an algorithm. They are supposed to read a story (and, if I understand you, perhaps read it again some finite number of times), and then, just like that, render a judgment. This is more like magic than mechanism."

This is objection is a complete non-starter. In order to see this, let's prove, in a thoroughly traditional manner, that a certain well-defined problem is effectively solvable. Recall that all Turing Machines can be recast as flow diagrams (cf. [6]). Next, note that any TM represented by a flow diagram having as part the fragment shown in Figure 2 would be a non-halting TM (because if started in state 1 with its read/write head scanning the leftmost 1 in a block of 1s -- and we can assume the alphabet in question to be a binary one consisting of {0,1} -- it will loop forever in this fragment. Let M be a fixed TM specified for computist Smith in flow diagram form, and let this diagram contain the fragment of Figure 2. Suppose that Brown looks for a minute at the diagram, sees the relevant fragment, and declares: ``Nonhalter!" In doing this, Brown assuredly decides M, and his performance is effective. And yet what's the difference between what Brown does and what my ``Eco-ish" agents do? The activity involved is decomposable in both cases. There are innumerable ``subterranean" cognitive processes going on beneath Brown's activity, but they are beside the point: that we don't (or perhaps can't) put them on display does not tell against the effectiveness in question. The fact is that Brown simply looks at the diagram, finds the relevant fragment, assimilates, and returns a verdict.²³ The same is true of my agents in the case of stories.

 

Figure 2: A Flow-Diagram Fragment That Entails Non-Halting

Before turning to the next objection, let me point out that the predicates I and $I_{\mbox{{\tiny UE}}}$ really aren't exotic, despite appearances to the contrary. All those who try to harness the concepts of theoretical computer science (concepts forming a superset of the formal ones canvassed in this paper) in order to get things done end up working with predicates at least as murky as these two. A good example is to be found in the seminal work of John Pollock, which is based on the harnessing of theoretical computer science (including AH) so as to explicate and implement concepts like warrant, defeasibility, prima facie plausibility, and so on.²⁴