Does Every Physical System Compute?

In my previous post, I introduced pancomputationalism–the idea that every physical system performs computations. There are three main versions of pancomputationalism.

Unlimited pancomputationalism says that every physical system performs just about any computation you like. For example, a piece of the Berlin wall sitting outside a museum, like the one in the picture that accompanies this post, implements MS Word.

Unlimited pancomputationalism may sound crazy, but it follows from the account of physical computation that says that all it takes for a physical system to perform a computation is for the computation to map onto the physical system. The problem is that it’s quite easy to map a fairly arbitrary computation onto a fairly arbitrary physical system.

Virtually everyone who writes in the philosophy of computation agrees that unlimited pancomputationalism is very bad because it trivializes the claim that physical systems perform computations, and in order to avoid it we need a more constrained account of physical computation. But many of the accounts that have been proposed still have the consequence that every physical system performs (a limited number of) computations. This is limited pancomputationalism.

There are also some physicists who argue that the physical world is, at bottom, computational. So every physical system performs at least one fundamental computation–furthermore, that’s all there is to being a physical system. This is a special version of limited pancomputationalism, which I call ontic pancomputationalism.

Ontic pancomputationalism has its own special problems–first and  foremost, that no one has explained in any clear way what it means for a concrete physical to be made up of computations.

More generally, limited pancomputationalism has the unappealing consequence that rocks, chairs, and livers compute in the same sense in which digital computers and iphones do. That just sounds preposterous. How come no one has ever demonstrated how a rock or any other arbitrary physical system performs a nontrivial computation in any useful way? Physical computation doesn’t come that cheap.

If you don’t like (limited) pancomputationalism, you could try your luck with the semantic view of computation, which says that computation requires representation. This might be a way to avoid pancomputationalism, but it runs into problems of its own–simply put, computation does not, in fact, require representation.

Alternatively, if you don’t like pancomputationalism you should consider endorsing the mechanistic account of physical computation, according to which only special sorts of systems perform computations–or at least, only such special systems perform computations in the interesting sense in which digital computers and perhaps brains do.



  1. Nicholas Denyer

    Why not say that a physical system performs a computation if we can use it to perform that computation? I.e. if the mapping between the physical system and the computation is straightforward? Or would this proposal make computation require representation in the way you do not allow?

  2. Gualtiero Piccinini

    Excellent point, Nicholas! I couldn’t agree more. The requirement that a computation be usable is built into my mechanistic account. The mechanistic account requires that computing systems perform (computational) teleological functions, which in turn are contributions to a goal of organisms. In order to contribute to the (computational) goals of organisms, a physical process must be usable. I discuss usability more explicitly, including their relevant to the physical limits of computation, in Chapters 15 and 16.

  3. What impact would this notion of pancomputationalism have on J.J. Gibson’s notion of ‘affordance(s)’? If I were told that the chair has been left outside for some time; is badly weathered; is rickety; unstable- you’d have a heads up about what not to do (sit on the chair). Saying that the chair instantiates some kind of computational system, makes what difference to the chair affordances?

    • Gualtiero Piccinini

      Thanks, James. I suppose it depends on what kind of pancomputationalism you are talking about. If you are an unlimited pancomputationalism, it makes no difference because your chair performs just about any computation. If you are a limited pancomputationalism, perhaps the chair’s computational states implement its affordances. If you are against pancomputationalism, like I am, you think chairs don’t perform computations in any interesting sense, so computation is just irrelevant to chair affordances.


Comments are closed.