Block’s “Psychologism and Behaviorism” has two aims: 1) hone in on the best version of behaviorism about intelligence, 2) show that even this very best behaviorist account of intelligence fails (for broadly ‘functionalist’ reasons).

A terminological disclaimer...

Block’s preferred philosophical terminology is idiosyncratic...

• Block’s “psychologism” (about cognition/intelligence) = our “functionalism” (about cognition/intelligence)
  • “Let psychologism be the doctrine that whether behavior is intelligent behavior depends on the character of the internal information processing that produces it.”
• Block’s “functionalism” = our “conceptual functionalism” (about all mental states)

Block’s Neo-Turing Test

Block’s description of the original Turing Test:

• “The Turing Test involves a machine in one room, and a person in another, each responding by teletype to remarks made by a human judge in a third room for some fixed period of time, e.g., an hour. The machine passes the test just in case the judge cannot tell which are the machine's answers and which are those of the person.”

And for what is the Turing Test a test? Intelligence!

• Caveat: “Note that the sense of ’intelligent’ deployed here--and generally in discussion of the Turing Test--is not the sense in which we speak of one person being more intelligent than another. ‘Intelligence’ in the sense deployed here means something like the possession of thought or reason.”

So to a first approximation, the conception of intelligence we find embedded in the Turing Test...

i. Intelligence just is the ability to pass the Turing Test (if it is given). (‘crude operationalism’)

Basic problem: Measuring instruments are fallible, so we shouldn’t confuse measurements for the thing being measured.

• No Turing Test judge will be infallible. Or, rather, we shouldn’t want our test to vary, depending on who happens to be the judge.

Initial solution: Put it in terms of behavioral dispositions instead...

• You can fail the test in some weird corner case while still having the general disposition to pass the test in most situations/scenarios. A single failure should not be conclusive evidence against a system having real intelligence.

ii. Intelligence just is the behavioral disposition to pass the Turing Test (if it is given). (Familiar Rylean Behaviorism)

• Basic problem: “In sum, human judges may be unfairly chauvinist in rejecting genuinely intelligent machines, and they may be overly liberal in accepting cleverly-engineered, mindless machines.”

• Initial solution: Replace the imitation game with a simpler game of ‘simply produce sensible verbal responses’

  • Ex: Compare two responses to the question, “So what have you been thinking about?”: 1) “I guess you could say that I’ve been thinking about everything...or maybe nothing? it’s just so gosh darn boring to be stuck inside not talking to anyone.”, 2) “A contagious symphony waltzed past my window. The third of february frowned while I stung a dagger.”
  • Note: this move away from the imitation game a a simpler “are these verbal responses sensible?” test drastically lowers the bar. But (according to Block) we’ll see that behaviorists can’t even clear this lowered bar.

iii. “Intelligence (or more accurately, conversational intelligence) is the disposition to produce a sensible sequence of verbal responses to a sequence of verbal stimuli, whatever they may be.”

Basic problem: The standard functionalist objections to behaviorism.

• In particular, the ‘perfect actor’ objection (think Putnam’s super-super spartans) shows that a determined deceiver could fool a test.

But! And here’s the crucial point:

• “As mentioned earlier, there are all sorts of reasons why an intelligent system may fail to be disposed to act intelligently: believing that acting intelligently is not in its interest, paralysis, etc. But intelligent systems that do not want to act intelligently or are paralyzed still have the capacity to act intelligently, even if they do not or cannot exercise this capacity.”

Upshot: The behavioral tests are maybe not necessary for intelligence, but they’re perhaps sufficient!

iv. “Intelligence (or, more accurately, conversational intelligence) is the capacity to produce a sensible sequence of verbal responses to a sequence of verbal stimuli, whatever they may be.” (Block’s Neo-Turing Test for Intelligence)

Perhaps behaviorism gives the right account of intelligence, even if not the right account for other kinds of mental states/properties...

Block against Block’s Neo-Turing Test

Block’s objection to the Neo-Turing Test’s conception of intelligence comes down to a fairly simple claim: we can conceive of a system that has the capacity to produce perfectly sensible outputs, without itself producing those outputs intelligently

On-the-fly reasoning is more sophisticated than a simple lookup table--And it’s that more sophisticated thing that we’re trying to characterize.

The Conceivability Test

• “The set of sensible strings so defined is a finite set that could in principle be listed by a very large and clever team working for a long time, with a very large grant and a lot of mechanical help, exercising imagination and judgment about what is to count as a sensible string.”
• “Imagine the set of sensible strings recorded on tape and deployed by a very simple machine as follows. The interrogator types in sentence A. The machine searches its list of sensible strings, picking out those that begin with A. It then picks one of these A-initial strings at random, and types out its second sentence, call it ``B''. The interrogator types in sentence C. The machine searches its list, isolating the strings that start with A followed by B followed by C. It picks one of these ABC-initial strings and types out its fourth sentence, and so on.”
• “...such a machine will have the capacity to emit a sensible sequence of verbal outputs, whatever the verbal inputs, and hence it is intelligent according to the neo-Turing Test conception of intelligence. But actually, the machine has the intelligence of a toaster. All the intelligence it exhibits is that of its programmers.”
• “I conclude that the capacity to emit sensible responses is not sufficient for intelligence, and so the neo-Turing Test conception of intelligence is refuted (along with the older and cruder Turing Test conceptions).”

The Upshot: The Cognitive/Rational/Intelligent Mind as a Computer

Our concept of intelligence involves something more than just the ‘external’ pattern of sensory inputs and behavioral outputs. The internal pattern of information processing that produces those inputs/outputs also matters. Those patterns need to be produced in the right way--in a way that looks something like ‘abstract rational thought’ rather than some dumb lookup table.

TL;DR: functional organization matters to intelligence!

Objections to Block

Objection 1: “Your argument is too strong in that it could be said of any intelligent machine that the intelligence it exhibits is that of its programmers.“

Block’s Reply: “The trouble with the neo-Turing Test conception of intelligence (and its predecessors) is precisely that it does not allow us to distinguish between behavior that reflects a machine's own intelligence, and behavior that, reflects only the intelligence of the machine's programmers.”

Objection 3(ish): This is a merely verbal dispute. You insist that the concept of intelligence ‘includes something more’ than mere input/output patterns. But unless you can specify what this extra special ingredient is, you’re just helping yourself to a magical (potentially-spooky?) conception of ‘intelligence’!

Block’s Reply: “...my point is based on the sort of information processing difference that exists.” All you need to grant is that there is an interesting and worthwhile difference between lookup-table algorithms and ‘on-the-fly’ reasoning algorithms. Call it whatever you want. That seems to be something we care about when we talk about ‘intelligence’.

Objection 6(ish): Are you sure what you’ve described is actually conceivable? Don’t you run into a problem where the physical size of your lookalike intelligence would have to be larger than the size of the physical universe?

Block’s Reply: “My argument requires only that the machine be logically possible, not that it be feasible or even nomologically possible. Behaviorist analyses were generally presented as conceptual analyses, and it is difficult to see how conceptions such as the neo-Turing Test conception could be seen in a very different light.”

My Objection (?): Doesn’t “actual and potential bahavior”--the sort of thing that a Rylean logical behaviorist likes--ultimately stem from the pattern of ‘internal’ infromation processing? That is, the two always in fact go together. And so for any postulated “lookalike” intelligence, so long as it’s finite (which it has to be), we can dream up scenarios in which it would malfunction in a telling way, or otherwise fail to demonstrate the full range of cognitive flexibility that a ‘real’ intelligence can exhibit.

A reply (?): Remember that Block’s point is conceptual, rather than empirical. Those two things (input/output patterns vs. information-processing patterns) may in fact be tightly connected. But that tight connection would be explained via the information-processing conception of intelligence.

(Ultimately, I suspect this comes down to realism vs. anti-realism about intelligence/cognition/rationality as a kind of real ‘achievement’, separable from less-philosophically-interesting brute causal laws. Wait for Dennett...)