Objections to Computationalism: A Survey

ROCZNIKI FILOZOFICZNE Tom LXVI, numer 3 – 2018 DOI: http://dx.doi.org/10.18290/rf.2018.66.3-3 MARCIN MIŁKOWSKI * OBJECTIONS TO COMPUTATIONALISM: A SURVEY* 1. INTRODUCTION Computationalism has been a fruitful and broad research tradition in cognitive research. Its main commitment is to claim that the brain is a kind of information-processing mechanism, and that information-processing is necessary for cognition. Notably, the radical view that computation is both sufficient and necessary for cognition is rarely accepted1 because it implies that any computer is already a cognitive system, and this view is counterintuitive, particularly since such a machine may be engaged only in utterly trivial operations. In this paper, however, I do not wish to enter the debate on the issue of whether there is a single correct view on the nature of cognition for all cognitive science disciplines (but see: MIŁKOWSKI 2013; AKAGI 2017). The positive view will not be developed here. In particular, the full account of physical computation will be set aside because it has already been elucidated in book-length accounts (FRESCO 2014; MIŁKOWSKI 2013; PICCINI 2015). For the purposes of this paper, it suffices to say that the current consensus among realists about physical computation is, roughly, that it occurs always and only in physical mechanisms whose function is to compute, or to process vehicles of information (understood as degrees of freedom of a physical medium); in the case of cognitive systems, these vehicles may also Dr hab. MARCIN MIŁKOWSKI — Zakład Logiki i Kognitywistyki w Instytucie Filozofii i Socjologii Polskiej Akademii Nauk; adres do korespondencji: ul. Nowy Świat 72, 00-330 Warszawa; e-mail: * The work on this paper was funded by a National Science Centre (Poland) research grant under the decision DEC-2014/14/E/HS1/00803. 1 Even Alan Newell required that the computation has to be of sufficient complexity. (See NEWELL 1980, 135–83). 58 MARCIN MIŁKOWSKI bear semantic information. Instead of elaborating on this account further, a review of objections is offered below, as no other comprehensive survey is available. However, the review presented below may not encompass all kinds of objections to computationalism. Hopefully, the most popular kinds of objections are included. The survey suggests that the majority of objections fail because they make computationalism a straw man. A number of them are, at best, red herrings. Some of them, however, have shown that stronger versions of computationalism are untenable, as well. Historically, they have helped to shape the theory and methodology of computational modeling. In particular, a number of objections show that cognitive systems are not only computers, or that computation is not the sole condition of cognition. This objection is already incorporated in the formulation of the main commitment above; no objection, however, establishes that there might be cognition without computation. The objections are ordered in the following way: (1) as related to the (supposed) nature of computers; (2) as related to semantics; (3) against computational functionalism; (4) as related to physical computation. In the conclusion, I briefly summarize the state of the debate. 2. OBJECTIONS RELATED TO THE (SUPPOSED) NATURE OF COMPUTERS C OMPUTER METAPHOR IS JUST A METAPHOR A computational description of a cognitive system is sometimes described as a computer metaphor. The use of the term suggests that the proposed description is rough and highly idealized; thus, it cannot be treated literally. For example, Karl Friston writes about the mathematical formulation of the free-energy principle in the following way: “Crucially, under some simplifying assumptions, these variational schemes can be implemented in a biologically plausible way, making them an important metaphor for neuronal processing in the brain.” (FRISTON 2012, 2101; see also, e.g., EKMAN 2003). As such, this is not an objection to computationalism. Obviously, all kinds of scientific models use idealizations. However, by using the term, others suggest that no computational model may be treated seriously; all are mere metaphors (DAUGMAN 1990, 9–18). OBJECTIONS TO COMPUTATIONALISM: A SURVEY 59 A defender of computationalism might concede this and weaken his position to say that idealizations inherent in computational modeling are misleading. But the position is also tenable in the stronger version (NEWELL & SIMON 1972; PYLYSHYN 1984, xiv–xvi). This is because computer metaphors cannot really be tested and rejected, whereas computational models can. The objection, in other words, fails. It is also confused: all scientific models are idealized, which does not make them metaphorical. S OFTWARE IS NOT IN THE HEAD This objection is that there is no simple way to understand the notions of software and hardware as applied to biological brains. But the software/ hardware distinction, popular in the slogan “the mind is to the brain like software is to hardware” (BLOCK 1995; PICCININI 2010), need not be applicable to brains at all for computationalism to be true. This objection is probably based on a conflation of stored-program computers with all possible kinds of computers. There are non-program-controllable computers: they do not load programs from external memory to internal memory in order to execute them. A mundane example of such a computer is a logical AND gate. In other words, while it may be interesting to inquire whether there is software in the brain, even if there were none, computationalism could still be true. This objection, therefore, fails. C OMPUTERS ARE JUST FOR NUMBER - CRUNCHING Another intuitive objection, already stated (and defeated) in the 1950s, is that brains are not engaged in number-crunching, while computers compute over numbers. But if this is all computers do, then they don’t control missiles or send documents to printers. After all, printing is not just number crunching. The objection rests, therefore, on a mistaken assumption that computers can only compute numerical functions. Computer functions can be defined not only by integer numbers but also through arbitrary symbols (NEWELL 1980), and, as physical mechanisms, computers can also control other physical processes. C OMPUTERS ARE ABSTRACT ENTITIES Some claim that, because symbols in computers are, in some sense, abstract and formal, computers — or at least computer programs — are abstract 60 MARCIN MIŁKOWSKI as well (LAKOFF 1987; BARRETT 2016; BARRETT, P OLLET, & S TULP 2014). In other words, the opponents of computationalism claim that it implies ontological dualism (SEARLE 1990). However, computers are physical mechanisms, and they can be broken, set on fire etc. These things may be difficult to accomplish with a collection of abstract entities. Computers are not just symbol-manipulators. They do things, and some of the things they do are not computational. In this minimal sense, computers are phy (...truncated)