Physical Computation: A Mechanistic Account
This book is about the nature of concrete computation—the physical systems that perform computations and the computations they perform. I argue that concrete computing systems are a kind of functional mechanism. A functional mechanism is a system of component parts with causal powers that are organized to perform a function. Computing mechanisms are different from non-computing mechanisms because they have a special function: to manipulate vehicles based solely on differences between different portions of the vehicles in accordance with a rule that is deﬁned over the vehicles and, possibly, certain internal states of the mechanism. I call this the mechanistic account of computation.
When I began articulating and presenting the mechanistic account of computation to philosophical audiences over ten years ago, I often encountered one of two dismissive responses. Response one: your view is obvious, well known, and uncontroversial—utterly dull. Response two: your view is counterintuitive, implausible, and untenable—totally worthless. These were not the only responses. Plenty of people engaged the substance of the mechanistic account of computation and discussed its pros and cons. But these radical responses were sufﬁciently common that they deserve to be addressed upfront.
If the mechanistic account elicited either one of these responses but not the other, perhaps the mechanistic account would be at fault. But the presence of both responses is encouraging because they cancel each other out, as it were. Those who respond dismissively appear to be unaware that the opposite dismissive response is equally common. If they knew this, presumably they would tone it down. For although reasonable people may disagree about whether a view is true or false, it is unreasonable to disagree on whether something is obviously true or obviously false. If it’s so obvious, how can there be equally informed people who think the opposite is obvious?
The ﬁrst dismissive response—that the mechanistic account is so obvious that it’s dull—seems to be motivated by something like the following reasoning. For the sake of the argument, let’s assume along with many philosophers that computation is a kind of symbol manipulation. There is an important distinction between the syntax of symbols (and, more generally, their formal properties) and their semantics. To a ﬁrst approximation, syntactic (more generally, formal) properties are those that determine whether a symbolic structure is well formed—they make the difference between ‘the puzzle is solvable’ and ‘puzzle the is solvable’; semantic properties are those that determine what symbols mean—they make the difference between ‘i vitelli dei romani sono belli’ in most languages, where it means nothing; in Latin, where it means go, Vitellus, at the Roman god’s war cry; and in Italian, where it means the calves of the Romans are good-looking. Most people ﬁnd it intuitively compelling that computations operate on symbols based on their formal or syntactic properties alone and not at all based on their semantic properties. Furthermore, many philosophers assimilate computational explanation and functional analysis: computational states are often said to be individuated by their functional relations to other computational states, inputs, and outputs. Therefore, computational states and processes are individuated functionally, i.e., formally or syntactically. Saying that computation is mechanistic, as my account does, is just a relabeling of this standard view. Therefore, the mechanistic account of computation is nothing new. Something like this reasoning is behind the ﬁrst dismissive response. It is deceptively persuasive but, alas, it goes way too fast.
A ﬁrst problem is that physical systems don’t wear their syntactic (or formal) properties on their sleeves. If the mechanistic account were based on syntactic properties, it should begin with an account of syntactic properties that does not presuppose the notion of computation. I don’t know of any such account, and fortunately I don’t need one. For the mechanistic account of computation is painstakingly built by specifying which properties of which mechanisms are computational, without ever invoking the notion of syntax (or formal property). Thus, the mechanistic account may provide ingredients for an account of syntax—not vice versa (Chapter 3, Section 4).
A second problem is the implicit assimilation of functional analysis and computational explanation, which is pervasive in the literature. I reject such an assimilation and argue that functional analysis provides a partial sketch of a mechanism (Chapter 5), defend a teleological account of functional mechanisms (Chapter 6), and argue that computational explanation is a speciﬁc kind of mechanistic explanation (Chapter 7).
An additional issue is that computations are often individuated semantically—in terms of functions from what is denoted by their inputs to what is denoted by their outputs. And philosophers interested in computation are often interested in how computation can explain cognition, which is usually assumed to deal in representations. After all, cognitive states and processes are typically individuated at least in part by their semantic content. Thus, many philosophers interested in computation believe that computational states and processes are individuated by their content in such a way that at least part of their essence is semantic. I call this the semantic account of computation. Therein lies the motivation for the second dismissive response: since computation is essentially semantic and the mechanistic account of computation denies this, the mechanistic account is obviously and horribly wrong.
But the semantic account of computation has its own problems. For starters, the notion of semantic property is even more obscure and more in need of naturalistic explication than that of syntactic property. In addition, I argue that individuating computations semantically always presupposes their non-semantic individuation, and that some computations are individuated purely non-semantically. Therefore, contrary to the second dismissive response, computation does not presuppose representation (Chapter 3).
But if we reject the view that computation presupposes representation, we risk falling into the view that everything performs computations—pancomputationalism (Chapter 4). This is not only counterintuitive—it also risks undermining the foundations of computer science and cognitive science. It is also a surprisingly popular view. Yet, I argue that pancomputationalism is misguided and we can avoid it by a judicious use of mechanistic explanation (Chapter 4).
The mechanistic account begins by adapting a mechanistic framework from the philosophy of science. This gives us identity conditions for mechanisms in terms of their components, their functions, and their organization, without invoking the notion of computation. To this general framework, a mechanistic account of computation must add criteria for what counts as computationally relevant mechanistic properties. I do this by adapting the notion of a string of letters, taken from logic and computability theory, and generalizing it to the notion of a system of vehicles that are deﬁned solely based on differences between different portions of the vehicles. Any system whose function is to manipulate such vehicles in accordance with a rule, where the rule is deﬁned in terms of the vehicles themselves, is a computing system. I explain how a system of appropriate vehicles can be found in the natural (concrete) world, yielding a robust (nontrivial) notion of computation (Chapter 7).
|Download Ebook||Read Now||File Type||Upload Date|
|Download Now here||Read Now
|December 10, 2019|
Do you like this book? Please share with your friends, let's read it !! :)
How to Read and Open File Type for PC ?