Semantic Engines and Syntactic Engines

Semantic Engines and Syntactic Engines

How can meaning make a difference? It doesn’t seem to be the kind of physical property, like temperature or mass or chemical composition, that could cause anything to happen. What brains are for is extracting meaning from the flux of energy impinging on their sense organs, in order to improve the prospects of the bodies that house them and provide their energy. The job of a brain is to “produce future” in the form of anticipations about the things in the world that matter to guide the body in appropriate ways. Brains are energetically very expensive organs, and if they can’t do this important job well, they aren’t earning their keep. Brains, in other words, are supposed to be semantic engines. What brains are made of is kazillions of molecular pieces that interact according to the strict laws of chemistry and physics, responding to shapes and forces; brains, in other words, are in fact only syntactic engines.

Imagine going to the engineers and asking them to build you a genuine-dollar-bill-discriminator, or, what amounts to the same thing, a counterfeit-detector: its specs are that it should put all the genuine dollars in one pile and all the counterfeits in another. Not possible, say the engineers; whatever we build can respond only to “syntactic” properties: physical details—the thickness and chemical composition of the paper, the shapes and colors of the ink patterns, the presence or absence of other hard-to-fake physical properties. What they can build, they say, is a pretty good but not foolproof counterfeit-detector based on such “syntactic” properties. It will be expensive, but indirectly and imperfectly it will test for counterfeithood well enough to earn its keep.

Any configuration of brain parts is subject to the same limitations. It will be caused by physicochemical forces to do whatever it does regardless of what the input means (or only sorta means). Don’t make the mistake of imagining that brains, being alive, or made of proteins instead of silicon and metal, can detect meanings directly, thanks to the wonder tissue in them. Physics will always trump meaning. A genuine semantic engine, responding directly to meanings, is like a perpetual motion machine—physically impossible. So how can brains accomplish their appointed task? By being syntactic engines that track or mimic the competence of the impossible semantic engine.[^1] But is this even possible? Some philosophers have argued that if the micro-causal story of how a brain works is complete (without any mysterious gaps), there is simply no room for meaning to make a difference. Dennett’s response, developed in chapter 33, is that semantic properties—such as truth, meaning, and reference—can play a role in causal explanations without introducing a gap in the physical account. That demonstration is not included in this excerpt.

[^1]: As philosopher John Haugeland (1981, p. 23) put it, the first principle of AI is “if you take care of the syntax, the semantics will take care of itself.” There are different ways of thinking of this slogan; in its first, overly hopeful version, it motivated the search for an enormous database, an axiomatized formalization of world knowledge that could be maintained and exploited by a (purely syntactical) inference engine. (CYC is the best example.) Dennett regards that approach as unfeasible, but the slogan survives as a good expression of the idea that the brain is a kind of computer (and hence a syntactic engine) that, thanks to its design, approximately does the work of a semantic engine.