Tuesday, March 24, 2009

Points of View are Irreducible

I’m an admirer of the relational interpretation of quantum mechanics (RQM), due originally to Carlo Rovelli. The revised SEP entry on RQM by Federico Laudisa and Rovelli made reference to two new papers about RQM by philosophers of science Bas C. Van Fraasen and Michel Bitbol. Unfortunately, Van Fraasen’s paper is not yet published, and Bitbol’s paper (entitled “Physical Relations or Functional Relations?”) is in French. However, in the course of looking for this, I found another interesting paper by Bitbol in English, which was published in a journal called NeuroQuantology: “Consciousness, Situations, and the Measurement Problem of Quantum Mechanics.”

In this paper, Bitbol looks at the history of discussions about the putative role of consciousness in the measurement process. He concludes that it is a mistake to think that human minds need have anything special to do with the measurement process; but a careful analysis reveals that QM does necessitate at a minimum taking into account particular viewpoints. His analysis places Bitbol in the same general camp as RQM and also the “Perspectivist” interpretation I described in these posts which referenced Paul Merriam’s papers.

The bulk of Bitbol’s paper is his careful presentation of a thought experiment which shows the difference between classical and quantum physics as it relates to the role of the observer in a measurement. (This exposition is similar to the thought experiment known as “Wigner’s friend” -- see Henry Stapp's discussion in this doc file). Wave functions characterize the observables of a system relative to interaction with another particular system –- in fact we can characterize a whole chain of interactions via a wave function -- until the chain comes to the end with our observation. But, Bitbol says the following:

"I am not saying that WE are unique or privileged beings in nature (this would be collective solipsism of an absurd sort), but only that we are privileged beings for US! As soon as we establish a relation with an element of the measurement chain, this element acquires a determination relative to US. Nothing has thus to be changed in the physical description, since determinations of the measurement chain are still relative to something. But everything is different for US, since the determinations of the measurement chain are now relative to US. And a relation of which WE are one term is something quite peculiar, even if it is only peculiar…from OUR point of view."

As he says, in classical physics, the fact that we are situated subjects with a point of view can be “bracketed”, and we can view ourselves as just another object. In quantum mechanics, this “situatedness” is irreducible. This doesn’t mean we can’t naturalize ourselves in picturing the world - we don’t need to think human consciousness is something distinct from the rest of nature. But QM teaches us that this process of naturalization has a boundary – points of view cannot be reduced or eliminated from the picture.

Tuesday, March 10, 2009

Quantum Computing and Mind Simulation

At the end of a recent discussion (in the comments to this post), Allen and I started to discuss whether quantum computing theory had something to say about the potential for simulating the human mind on a computer. I then read a couple of review articles on quantum computing he referenced. Before getting to what I learned (below), I wanted to explain my prior philosophical view on the simulation question.

The Russellian Stance, Functionalism, and Simulation

I endorse a form of the Russellian approach to solving the hard problem of consciousness. Russell described the world as a causal network of events, and he noted that physical theory only describes the extrinsic or dispositional nature of these events. These events also have an intrinsic nature which ultimately grounds the qualitative and experiential character of consciousness. I think quantum mechanics provides support for this view: quantum measurements seem to fit perfectly into this Russellian picture as the base-level events which ultimately underpin both the physical and experiential facts.

Functionalism is the thesis that the mind can be described as an abstract causal system. As a practical matter, the functionalist’s description is taken at a coarse-grained level – i.e. there is some minimal scale below which the actual physical details of the brain/body system are assumed to be irrelevant to its function. It follows that such a functional model could be realized in any number of physical ways, including via computer. Computationalism is the variety of functionalism which pursues the computer modeling approach.

Now, I think the Russellian stance on functionalism and the potential for simulation is nuanced. On the one hand, functionalism is seen as misguided because it only considers extrinsic causal structure. On the other hand, unlike an old fashioned dualist, the Russellian shouldn’t rule out the possibility that the mind could be simulated. The mind, after all, is a product of a natural system – we don’t need extra immaterial stuff to explain it. Perhaps a simulation can get the functional structure right and the correct intrinsic experiential character will come along “for free.”

The problems come with the coarse-graining. In every functionalist account I’ve seen, this takes place at a scale where quantum mechanics is assumed to be safely irrelevant. But every process in the body ultimately is grounded in molecular, atomic and sub-atomic activity which must be described quantum mechanically. So, a coarse-grained, approximated simulation of the brain/body’s causal structure on some physical device would likely miss crucial details which lie at the quantum level (details I think simultaneously crucial to both extrinsic function and conscious experience)

How would a functionalist respond to the quantum question? First, many believe distinctive quantum phenomena effectively “wash out” in a macroscopic system like the human brain/body. This belief is often based on the presumed impact of environmental decoherence. I’m not going to pursue this issue in this post (I discuss this in some of my posts on quantum biology). Another response to the quantum question is an appeal to a commonly believed thesis that any physical system (including a quantum system) can be modeled by a classical computer -- so the traditional functionalist/computationalist approach wouldn’t be missing anything distinctive anyway. This is the view that I wanted to explore by reading up on quantum computing theory.

[Please note the discussion that follows may suffer even more than usual from by my ignorance of the subject matter.]

Simulation and the Church-Turing thesis

So, is it true that any physical system, including a quantum system, can be simulated by a classical computer? Well, this idea has been defended by appeals to versions of the Church-Turing thesis. The original Church-Turing thesis states (in a formulation from this article) that any effectively calculable function can be computed using a Turing machine. (For a description of a Turing machine, see here.) Now it seems that what this thesis really meant can probably only be appreciated by studying its original logical/mathematical context. In his SEP article on the C-T thesis, Jack Copeland first traces the development of the ideas associated with the thesis in the pioneering work on calculation and computing by Turing, Church and others beginning in the 1930’s. Then, Copeland spends much of the rest of the article objecting to how the thesis has been misunderstood and misused by philosophers and computing theorists. The C-T thesis did not purport to say that all physical systems, or even all machines regardless of architecture, could be simulated by a Turing machine. It certainly did not prove anything of the sort. (This discussion reminded me of similar debates over the philosophical applicability of Gödel’s incompleteness theorems beyond their original context – see old posts here and here.)

Nevertheless, as long as one is careful not to inappropriately invoke the authority of the original C-T thesis, one can explore more expansive versions and try to evaluate their validity.

Physical Versions of the C-T thesis

In her review article on Quantum Computing, Dorit Aharonov presents two versions of the thesis. First, (p.3), she presents a simple “physical” version: “A Turing machine can compute any function computable by a reasonable physical device.” She says this is something which cannot be proven, but that no known counterexamples exist. In particular, quantum computers are not believed capable of computing functions non-computable by a Turing machine.

She quickly then notes that: “However, in the theory of computation, we are interested not only in the question of which functions can be computed, but mainly in the cost of computing these functions.” [Emphasis original] The way this is evaluated is by noting whether the computational resources needed rise as a polynomial or an exponential function of input size. It is the former which form the set of tractable computations.

After also discussing the superior efficiency of a probabilistic version of the Turing machine, she presents another thesis to consider (p.4): “The modern Church thesis: A probabilistic Turing machine can simulate any reasonable physical device in polynomial cost.” We have a great deal of evidence, though not proof, that this thesis is contradicted by quantum computers.

The Advantages of Quantum Algorithms

Aharonov explains first, that quantum computers can simulate classical computers, at little loss in efficiency. On the flip side, it appears classical computers can simulate quantum computers but only at exponential cost. What we really want, though, is a positive demonstration of how far quantum computers can outperform their classical counterparts.

The a priori expectation might be that the ability to manipulate qubits, which can be in a superposition of states as opposed to just to two states, would lead to great increases in computing power. Because of the necessity for conducting a measurement to extract results (collapsing superpositions), however, the power of this idea is muted. Other, more subtle sources of limitations on quantum computing are discussed later in the paper.

Despite this, however, many investigations into quantum computing over the years have found quantum algorithms which improve efficiency. One of these algorithms, Shor’s, gives a polynomial algorithm for factoring integers where all known classical algorithms have exponential cost, thus crossing the crucial boundary. It must be pointed out that as of yet there is no proof that a classical polynomial algorithm for factoring is impossible.


Most of what I’ve discussed in Aharonov’s article above comes from the introduction. In the ensuing 60 pages she goes into more detail about the nature of computers and computing, various models for quantum computing algorithms, the issues of noise correction and fault tolerance, and some of her own ideas of what quantum computing theory says about the boundary between quantum and classical regimes in physics.

On the key question of what quantum computers can do better than classical ones, one is left with the impression that the question is much more subtle than might first be imagined. We have some exciting theoretical results, but perhaps fewer than might have been anticipated on a naïve expectation. At the same time, it seems we’re still in the early phase of growth in our knowledge of the field. A lot of interesting work and new developments lie ahead. (The engineering efforts underway toward building quantum computers and the challenges they face is another interesting topic).

Let me return to the question about what all this might mean for the philosophy of mind, assuming (as I do) that the quantum level grounding of biology contributes meaningfully to the mind’s function. I think a modest conclusion is called for. The fact that a classical computer can simulate a quantum computer only at an exponential cost suggests that the project of simulating a human mind is impractical, though not blocked in principle. This conclusion is broadly consistent with my philosophical stance regarding the simulation project.

[P.S: after drafting this I recalled there was a good debate (thanks to Tanasije and Mike) on the simulation topic in the comments to this May 2008 post on Russellian theory. My memory is awful.]

Tuesday, March 03, 2009

John Heil Gets Very Close…

…to solving the mind/body problem.

John Heil’s 2003 book From an Ontological Point of View consists of roughly three components. The first part is a thoroughgoing critique of analytic philosophy based on the central role it gives to language. The second part is his positive ontological account: this is centered on his theory of power-properties. The third section discusses the application of his ontology to philosophical problems, notably the mind/body problem.

Being a blogger, I’m only going to discuss one aspect of how his account of properties (as part of a substance/property ontology) gives him a leg up on solving the mind/body problem, although I think he ultimately falls short of the goal. For an appreciation of his overall approach to philosophy, see this paper by Ross Cameron and Elizabeth Barnes. A Notre Dame book review penned by Gary S. Rosenkrantz is here. (There is also a whole volume of collected responses to the book available -- I have not read this).

Powers and Qualities

For Heil, properties are powers (or dispositions). Properties are distinguished by their contribution to the causal powers of their possessors. There are many potential advantages for this approach in constructing a realist account of ontology and causation. One advantage as it relates to philosophy of mind is that the ubiquity of powers offers a natural home for intentionality in the world. Powers have an “aboutness” – they are always directed towards their manifestations (I discussed this at greater length in my posts on George Molnar’s powers ontology).

But the biggest innovation in Heil’s account compared to other treatments of power properties is his treatment of qualities. Qualitative properties are usually considered to be separate from dispositions. They are often identified with the so-called categorical properties of objects. Heil proposes that qualities and powers are the same thing. He calls this the identity theory: “If P is an intrinsic property of an object, P is simultaneously dispositional and qualitative… (p.111 – page references are to the paperback edition)” Furthermore, this is not to be seen as a dual aspect theory: P’s nature is truly and simultaneously dispositional and qualitative. (Heil gives credit to the late philosopher C.B.Martin for this idea).

Heil’s defense of this identity theory was persuasive, in my opinion. He starts by giving some homely examples (size, shape, color) of how objects can be seen as having effects by virtue of their qualities. He then notes that while physics is silent on the subject of qualities, this in no way contradicts the identity theory. If middle level objects have qualities, it makes sense that their constituents do as well. (Heil strongly objects to ontological accounts which feature distinct “levels” of reality).

Qualities in the Brain

Turning to the mind/body problem, we can see that if properties which are both dispositional and qualitative make up the world then there is no special problem with the fact that conscious experience seems to possess qualities. Still, Heil wonders, could it be that these qualities are just qualities of, say, neurological activity? He notes that the qualities we should focus on in considering this question are not the simply representational qualities of experience, but the “diaphanous” qualities associated with having an experience. The qualities of experience outstrip their representational qualities, and these residual qualities are key to the question.

P.229: “The what-it-is-likeness of conscious experience stems from the nature of the representational medium…” as well as what is being represented. Heil makes use of a thought experiment based on an actual apparatus developed for the blind whereby a camera impresses images via pressure on the skin of the back or chest. After a while, the subject using such a device mostly forgets the nature of its implementation and just processes the images. But certainly the residual, non-representational experiential qualities are different between vision and this touch-analogue of vision.

So, do these “what-it’s-like” qualities differ from material qualities, or can we say that they are no different? Can we locate them in the brain? Well, Heil asks, why not? Perhaps these qualities are just are neurological qualities.

What about the First-Person Aspect?

The remaining question relates to the different way we are acquainted with these putative neurological qualities. P.234: “…we do seem to have something like ‘direct acquaintance’ with neurological qualities.” This means there still is a distinction here between the what-it’s-like qualities and other material qualities. Heil gives us an account that deals well with the qualitative character of experience -- after all qualities are ubiquitous in the world. But what about the subjective character of experience (my term)?

One might ask: is this subjective character also ubiquitous? Alas, Heil doesn’t want to take that leap. After all: “You might be worried that a conception of this kind leads to panpsychism or worse. (p.234)” Heil wants to maintain that all of the qualities of experience are perfectly ordinary qualities of brains. He concedes that it remains to be understood how this could be the case. But he hopes we’ll know more in the future as we learn more about neuroscience.

There’s one additional passage where Heil ponders this issue of the subjective character of experience (pp.237-9). He discusses the privacy of mental states, and acknowledges the difference between being in a state and observing the state. But he objects to treating this as indicating an ontological divide between subjective and objective properties. Conscious people are, after “objective” in the sense of being natural entities in the world. But again: “What, then, distinguishes conscious states from those that are not conscious? (p.239)” Could it be that there is a different functional role? This can’t be an explanatory strategy for Heil, since causal and qualitative roles are constituted hand-in-hand. I give Heil credit for raising the issue, but for him the privacy/subjectivity issue remains unexplained.

What about Zombies?

In his Chapter 20 (pp.240-249), Heil looks at the conceivability, or “zombie” argument against materialism. The bottom line is that zombies are not conceivable given Heil’s ontology: “The zombie possibility arises only against a particular ontological background, one according to which powers and qualities are only contingently related. (p.248)” Heil is right -- IF you define the zombie problem as being about qualia, rather than being also concerned with the subjective character of experience. As an aside, this is why I personally don’t use the term qualia when discussing the mind/body problem: first-person experience possesses both qualitative character and subjective character – discussions of qualia often ignore the latter issue (on this point see also a quote from philosopher Uriah Kriegel in this old post).


So, Heil comes close to solving the mind/body problem. By virtue of an ontology which places qualities in the world, the qualitative character of conscious experience is accounted for. The other dimension to experience, its subjective character, is left without an account. I think the only way to address this is to place subjective points of view into the natural world as well.

PS: My thanks go to Gualtiero Piccinini of the blog brains for his post on C.B.Martin, also suggesting Heil as a route to helping to understand Martin. This prompted me to read the Heil book, which I had been meaning to do. Martin's book is in the queue.