John Searle's Chinese Room Argument

He's fed up with talking about this. I used to have a copy of the little book he wrote called Minds, Brains & Science. Apparently the AI at Google has read it too and gets it completely:

How does stuff like this happen? Most people won't even notice what I'm saying even if I point out that the Google search changed the title. The book was really called Minds, Brains & Science, not Minds, Brains & Programs. (But maybe that was a later edition or something?) I asked Google and it said:

John Searle's book Minds, Brains and Science did not become "Minds, Brains & Programs"—rather, they are two separate, distinct works published a few years apart:

• "Minds, Brains, and Programs" (1980): A landmark academic paper published in the journal Behavioral and Brain Sciences. This is where Searle first introduced his famous Chinese Room Argument to challenge the concept of "strong AI".
• "Minds, Brains and Science" (1984): A short book published by Harvard University Press (derived from his BBC Reith Lectures). It offers a broader introduction to the philosophy of mind, covering the mind-body problem, free will, and artificial intelligence (drawing upon ideas from his earlier paper, among other topics). 

They were never the same title; one is a 1980 journal article and the other is a 1984 book series based on lectures.

Echoes in a Chinese room! 

Jon Doyle in his 1980 PhD thesis A Model for Deliberation, Action and Introspection mentioned Searle on the very first page, where he has a footnote saying what he means by "Program":

You can read more about the work of Brian Cantwell Smith and others in Nada Amin - Metacircular Interpretation ad infinitum. William Byrd also talks about some advice his brother passed on concerning the dangers of seeming too innovative, and some very fancy Quines he meta-programmed. See William Byrd on Relational Programming and Quines. 

David Marr, in Vision: A Computational Investigation into the Human Representation and Processing of Visual Information (1982) points out that formal representation is something more than just a text. It is a text with an associated effective method for transforming that text into something meaningful:

Representation and Description

A representation is a formal system for making explicit certain entities or types of information, together with a specification of how the system does this. And I shall call the result of using a representation to describe a given entity a description of the entity in that representation (Marr and Nishihara, 1978).

For example, the Arabic, Roman, and binary numeral systems are all formal systems for representing numbers. The Arabic representation consists of a string of symbols drawn from the set (0, 1, 2, 3, 4, 5, 6, 7, 8, 9), and the rule for constructing the description of a particular integer n is that one decomposes n into a sum of multiples of powers of 10 and unites these multiples into a string with the largest powers on the left and the smallest on the right. Thus, thirty-seven equals 3 x 10¹ + 7 x 10⁰, which becomes 37, the Arabic numeral system's description of the number. What this description makes explicit is the number's decomposition into powers of 10. The binary numeral system's description of the number thirty-seven is 100101, and this description makes explicit the number's decomposition into powers of 2  [1 x 2⁵ + 0 x 2⁴ + 0 x 2³ + 1 x 2² + 0 x 2¹ + 1 x 2⁰ = 32 + 0 + 0 + 4 + 0 + 1 = 32 + 4 + 1 = 37].  ln the Roman numeral system, thirty-seven is represented as XXXVII.

This definition of a representation is quite general. For example, a representation for shape would be a formal scheme for describing some aspects of shape, together with rules that specify how the scheme is applied to any particular shape. A musical score provides a way of representing a symphony; the alphabet allows the construction of a written representation of words; and so forth. The phrase "formal scheme" is critical to the definition, but the reader should not be frightened by it. The reason is simply that we are dealing with information-processing machines, and the way such machines work is by using symbols to stand for things--to represent things, in our terminology. To say that something is a formal scheme means only that it is a set of symbols with rules for putting them together--no more and no less.

To which Searle would reply, if I understood anything in that book, something like:

What makes you think it is the machine that uses symbols to stand for things and not the intelligent human being who knows how the machine works and who intentionally programs the machine in such a manner that the symbols it mindlessly cranks out have the particular significance of whatever things the computation is supposed to represent? 

Notice that thirty-seven is a decimal quine, as is forty-two, but six times seven is not. 

There was a nice talk by Pat Hanrahan at the RacketCon in Oakland last week:

53:39 Matthias is right about this. But that said, there must be some advantage to two people using the same concrete syntax when they discuss some specific problem face to face, in front of a blackboard, say cf. discussion at 1:05:02. Interesting that he says when psychologists use the term representation they invariably mean there is something concrete out in the world that the subject is experiencing. Aristotle would understand that. What do the psychologists think of as a presentation then? English punctuation is all very well, and suggestive to some English speakers (probably only a minority though), but if one person is Chinese and the other is Greek maybe they'd both be better off inventing some common domain-specific syntax. It is possible that we could develop algorithms to compute intermediate representations, given two separate ones. As to how vital Unicode is, I disagree. The last thing the world needs is a committee policing what are digits and what are letters and telling people what sort of letters, digits and symbols they can use to say what. You can still use the code as a semi-universal character interchange, but you need to make sure you are not restricted by it. The way to do that is use abstract syntax trees to represent the shapes, meanings and codes, and provide the necessary algorithms to convert these between representations. You also need to be able to communicate with whales and dolphins who don't use written symbols, unless you count bubbles.

Subscribe to Racket.

This is all coming up here because I have been trying to write down a response to this talk Propositional calculus and the nature of reality by Samson Abramsky which he gave at an Isaac Newton Institute seminar in 2012 called Semantics and Syntax. In this talk he mentions the measurement problem in Quantum Mechanics, and shows how it is because of an interpretation of frequency of observations in a collection of experimental runs as an a priori probability produces what he calls a logical Bell inequality, which is an upper bound on the amount you can learn from a model of unitary evolution of a system. He goes on to suggest that this might mean there is a kind of fully abstract interpretation of Quantum Mechanics which I think goes back to the ideas of William Wootters' which were the subject of his PhD (1980!). See Optimal Information Transfer and Real-Vector-Space Quantum Theory by William K. Wootters (2013).

See the following papers:

• The Sheaf-Theoretic Structure Of Non-Locality and Contextuality by Samson Abramsky and Adam Brandenburger
• Logical Bell Inequalities by Samson Abramsky and Lucien Hardy 

Other posts of mine obviously related to this investigation are:

• Jacob Barandes' Talk at ESA Quantum Technologies Conference 2025 
• 50 Years of the Roche Biochemical Pathways Poster
• Fredrik Nordvall Forsberg on Linear Logic 
• James Cussen interviewing Cognitive Neuroscientist Peter Tse 
• John D. Norton - Claims that information is fundamental are mistaken: The thermodynamic myth at the heart of information theory 
• Happy Birthday Raymond Solomonoff
• Popper's Non-dualist Theory of Objective Knowledge

The abstract of Abramsky and Hardy's 2012 paper reads:  

Bell inequalities play a central role in the study of quantum non-locality and entanglement, with many applications in quantum information. Despite the huge literature on Bell inequalities, it is not easy to find a clear conceptual answer to what a Bell inequality is, or a clear guiding principle as to how they may be derived. In this paper, we introduce a notion of logical Bell inequality which can be used to systematically derive testable inequalities for a very wide variety of situations. There is a single clear conceptual principle, based on purely logical consistency conditions, which underlies our notion of logical Bell inequalities. We show that in a precise sense, all Bell inequalities can be taken to be of this form. Our approach is very general. It applies directly to any family of sets of commuting observables. Thus it covers not only the n-partite scenarios to which Bell inequalities are standardly applied, but also Kochen-Specker configurations, and many other examples. There is much current work on experimental tests for contextuality. Our approach directly yields, in a systematic fashion, testable inequalities for a very general notion of contextuality. 

There has been much work on obtaining proofs of Bell's theorem `without inequalities' or `without probabilities'. These proofs are seen as being in a sense more definitive and logically robust than the inequality-based proofs. On the [other] hand, they lack the fault-tolerant aspect of inequalities. Our approach reconciles these aspects, and in fact shows how the logical robustness can be converted into systematic, general derivations of inequalities with provable violations. Moreover, the kind of strong non-locality or contextuality exhibited by the GHZ argument or by Kochen-Specker configurations can be shown to lead to maximal violations of the corresponding logical Bell inequalities.  

The abstract to the above 2013 paper of Wootters reads:

Consider a photon that has just emerged from a linear polarizing filter. If the photon is then subjected to an orthogonal polarization measurement-e.g., horizontal vs vertical-the photon's preparation cannot be fully expressed in the outcome: a binary outcome cannot reveal the value of a continuous variable. However, a stream of identically prepared photons can do much better. To quantify this effect, one can compute the mutual information between the angle of polarization and the observed frequencies of occurrence of "horizontal" and "vertical." Remarkably, one finds that the quantum-mechanical rule for computing probabilities--Born's rule--maximizes this mutual information relative to other conceivable probability rules. However, the maximization is achieved only because linear polarization can be modeled with a real state space; the argument fails when one considers the full set of complex states. This result generalizes to higher dimensional Hilbert spaces: in every case, one finds that information is transferred optimally from preparation to measurement in the real-vector-space theory but not in the complex theory. Attempts to modify the statement of the problem so as to see a similar optimization in the standard complex theory are not successful (with one limited exception). So it seems that this optimization should be regarded as a special feature of real-vector-space quantum theory.

Did Wootters know about Abramsky and Hardy's paper of the year before? Did he read it? Did they read his 2013? What did they each think? Then listen to Jacob Barandes' ESA talk above where he explains why the Hilbert space representation of quantum states is an absolutely crazy notion of anything anyone would ever try to pass off as a representation of something real, .... Did he read Wootters paper? Did he read Abramsky and Hardy's? What do they each think of these ideas? 

Do Abramsky, Hardy, Wootters and Barandes know about the Port-Hamiltonian formulation of open thermodynamical systems which Frederic Schuller has studied? See Lyapunov Stability in Dynamical Systems. Can somebody please point out in a few crisp English sentences why the stochastic formulation of Barandes does not accommodate a thermodynamic physical model like the one described by Janusz Badur and Piotr Józef Ziółkowski? Because to physicists it smacks of instrumentalism and it'll never get past their chief of thought police? See David Albert Talking Complete Nonsense. 

And it goes on and on. Why so much? Because we keep inventing new representations and we don't take the care to investigate the algorithms we need to transform them into one another so we lose track of the invariants: Samaneri Jayasāra - Longchenpa ~ The Enlightened Mind ~ Dzogchen.

Subscribe to INI Seminar Room 2.

In this 35 page essay I have written I end up saying that Isaac Newton was wrong. So what? Einstein proved Newton was wrong! Well I end up showing Einstein was wrong too, because the only bit of Newton's work he kept was the bit that was the mistake, the rest was sound as a bell. Too innovative by far. And Newton's mistake? All his ideas he got from Aristotle except the one he invented himself which was wrong: what goes on in the heavens is not the subject of empirical study because we are on earth, not in the heavens. Aristotle's Physics, book one, part one. Newton didn't do his lessons. 

But it's OK, you don't have to listen to me. 

This song was such fun, but now it just makes me sad.

Subscribe to Nick Cave & The Bad Seeds.

See Noriko Miyakawa's Work With David Lynch: 

Subscribe to David Lynch Theater.