next up previous
Next: The Turing Test Up: Computationalism is Dead; Now Previous: Introduction

What is Computationalism?

Fetzer assures us elsewheregif that people do advance the bald thesis that persons are Turing machines. They certainly do. In fact, though there may be a need to tinker with this proposition,gif it captures a large part of computationalism in charitable fashion. One reason this is so is because it doesn't follow from the fact that x is a TM that some process that x undergoes must be formalizable as (say) Turing computation of some effectively computable function. I would respectfully suggest that Cummins and Schwarz [8] -- who, as Fetzer notes in section 7 of his [11], propound the chain that computing is the execution of functions, which entails the execution of algorithms, which in turn involves disciplined step satisfaction -- take note of this fact. Specific examples are easy to come by: a composite TM could include a ``sub" TM that nondeterministically attempts to find composite numbersgif

In general, it seems to me that computability theory provides ``off the shelf" objects and concepts sufficient to formalize all the distinctions Fetzer finds computationalists making. For another example, consider ``asymmetrical decision procedures," to which the computationalist Johnson-Laird [16], as Fetzer notes elsewhere [12], appeals. Fetzer cites as an example of such a procedure the classic one for establishing that the set of all proofs in first-order logic is enumerable. It's of course well-known that there is a TM that performs the work in question; the problem here is known to be tex2html_wrap_inline332 in the Arithmetic Hierarchy ( tex2html_wrap_inline334 and tex2html_wrap_inline332 formalize that which can be solved by any of the techniques open to standard computationalism). In general, then, my advice to computationalists who come under Fetzer's scrutiny is to drop all reference to ersatz computation and embrace the rich but certified offerings of computability theory.

Fetzer himself believes that at least the spirit of my recommendation is (imprudently) affirmed by computationalists, for he says:

[T]he computational conception appears to have arisen from the almost irresistible temptation to appeal to a domain about which a great deal is known (formal systems and computability theory) as a repository of answers to questions in a domain about which very little is known (the nature of mentality and cognition) (p. 25, [12]).

But I offer analogous advice for some of those intent on razing computationalism. For example, Penrose's objection (which Fetzer reports and discusses in [12]) that thinking can't be computation because computation is deterministic is a weak one in light of the fact that TMs can be allowed to move from state to state in nondeterministic and probabilistic fashion. Such machines don't have state transition functions; transitions are simply n-tuples from an appropriate relation on states and symbols.

Fetzer does consider an approximation of the move of simply identifying persons with TMs, as can be seen by considering what he has to say about dreams.gif Fetzer says:

Many kinds of thought processes of human beings as thinking things fail to satisfy [the conditions that determined computation]. Dreams and daydreams, for example, fail to satisfy them because they have definite starting point and no definite stopping point: they begin and they end, but they have no given premises or conclusions" (sect. 8, para. 1).

But he also tells us that

A computationalist could insist that dreams and daydreams, properly understood, really are instances of disciplined step satisfaction (or of the execution of complete or partial functions) by observing that, even in the case of dreams or daydreams, a system simply shifts from state tex2html_wrap_inline340 at time tex2html_wrap_inline342 to state tex2html_wrap_inline344 at time tex2html_wrap_inline346 in some specific sequence. ([12], p. 24)

Immediately after considering this version of the heart of computationalism, Fetzer says that, with respect to the system in question,

One could insist that any system that passes through those states is computing some particular function -- the function f that takes input tex2html_wrap_inline340 at tex2html_wrap_inline342 and returns output tex2html_wrap_inline344 at tex2html_wrap_inline346 . This defense, however, trivializes computationalism by ignoring the difference between causal processes and computational procedures. ([12], p. 24; see also [11], sect. 9, para. 3)

It is possible that I could be persuaded here, but my suspicion is that a clever computationalist will insist that the only outright function that needs to figure in a description of an AI is one mapping percepts to behavior. So for example a Turing machine that starts with one tex2html_wrap_inline358 on its tape and simply copies that tex2html_wrap_inline358 over and over again ``forever" is doing enough, in principle, to count as an artificial agent. (For a marvelous introduction to AI based on this percept tex2html_wrap_inline362 behavior conception, see [18].) At any rate, it isn't Fetzer's intention, nor is it mine, to set out computationalism in glorious detail.gif All that is needed at present is a rough-and-ready characterization of computationalism, and here's what I suggest:

Computationalism consists of the following four propositions.

A function f is effectively computable if and only if f is Turing-computable.
Persons are Turing machines.
The Turing Test is valid.
Computationalists will succeed in building persons.
Computationalists will succeed in building Turing Test-passing artifacts. (This proposition is presumably entailed by its predecessor.)

next up previous
Next: The Turing Test Up: Computationalism is Dead; Now Previous: Introduction

Selmer Bringsjord
Tue May 21 00:31:50 EDT 1996