Sunday 1 September 2024

Neuroscientist Anil Seth Links Panpsychism To Integrated Information Theory

 

(i) Introduction
(ii) Integrated Information Theory and Panpsychism
(iii) David Chalmers on Conscious Thermostats
(iv) Is Information Fundamental?

Introduction

Firstly, let me quote the British neuroscientist Anil Seth offering his broad view of panpsychism. He writes:

“Panpsychism is the idea that consciousness is a fundamental property of the universe, alongside other fundamental properties such as mass/energy and charge; that is present to some degree everywhere and in everything.”

Seth then immediately tackles the subject of “silly” panpsychism when he continues with the following words:

“People sometimes make fun of panpsychism for claiming things like stones and spoons are conscious in the same sort of way that you and I are, but these are usually deliberate misconstruals designed to make it look silly. There are more sophisticated versions of the idea [].”

More relevantly to this essay, an important way of linking panpsychism to integrated information theory (IIT) can be seen when comparing Seth’s words that

[p]anpsychism is the idea that consciousness is a fundamental property of the universe”

to his words (written elsewhere) that integrated information theory

“also implies that *information itself* exists — that it has some definite ontological status in our universe — a status like mass/energy and electrical charge”.

Here we have two rival views as to what is fundamental in the universe…

Or are they rivals?

Integrated Information Theory and Panpsychism

Integrated information theorists don’t believe that all (physical) entities (or “systems”) instantiate consciousness. The main reason for this is that not every… thing has the required level of integrated information to be conscious.

In broad and general terms, most integrated information theorists (along with Anil Seth himself) emphasise complexity and integration. Thus, unlike panpsychists, they also focus almost entirely on biological brains.

Seth offers his readers his own technical take on integration, information and complexity when he cites the example of the molecules in a gas. He claims that this

“kind of system has maximum information — maximum randomness — but shows no integration at all, because every element is independent from each other”.

On the other hand, we also have Seth’s other example of a “crystal lattice”. In this case, “all the elements do exactly the same thing”. Thus, here we have

“maximum integration, but almost no information, because there are very few possible states that the system can be in”.

Interestingly enough, the physicist Max Tegmark also mentions gases. And he too uses integrated information theory to distinguish conscious matter from other physical systems such as gases, liquids and solids. Indeed, he backs up both Anil Seth’s and Giulio Tononi’s position when he tells us that consciousness is dependent upon “information, integration, independence, dynamics, and utility principles”.

Now let’s tackle panpsychism more broadly.

Panpsychism

The problem with arguing that consciousness (or experience) is integrated information, and that information is everywhere, is that even very simple objects (or systems) must instantiate (or contain) a degree of (integrated?) information. Therefore, such basic objects must also have a degree of consciousness. Or, in the language of integrated information theory, all such objects (or systems) must have a “φ value”.

Perhaps, then, we’ve entered the territory of panpsychism here.

Not surprisingly, Giulio Tononi’s position does touch on panpsychism — even if his position isn’t identical to that of panpsychists. That said, he’s written conflicting things about this particular philosophical ism.

For example, Tononi wrote the following words:

“Unlike panpsychism, however, IIT clearly implies that not everything is conscious.”

Despite all that, and to repeat, IIT has it that even basic objects have a nonzero degree of Φ. This would mean that consciousness is almost everywhere — if only to a rudimentary degree (as with the “proto-experience” of panpsychists).

In any case, the argument that IIT is not a kind of panpsychism is at odds with what the philosophers David Chalmers and John Searle believe. They do take IIT to be a form of panpsychism. [See here and here.] What’s more, the German-American neurophysiologist and neuroscientist Christof Koch (Giulio Tononi’s co-worker) has even claimed that IIT is a “scientifically refined version” of panpsychism.

In any case, if we accept a strong — indeed a necessary — link between consciousness and integrated information, then an ant or a virus must have a “non-zero degree of consciousness”

Indeed, this could be true of a thermostat too!

David Chalmers on Conscious Thermostats

In his ‘What is it like to be a thermostat?’, David Chalmers writes:

[Thermostats] take an input, perform a quick and easy nonlinear transformation on it, and produce an output.”

What does Chalmers mean by the word ‘information’ when it comes — specifically — to a thermostat?

Basically, heat and cold (i.e., all variations in temperature in a given environment) can be seen as bits of information. However, are heat and cold information for a thermostat? More relevantly, does that even matter in this IIT-panpsychism context?

Or is it the case that the actions (i.e., cases of processing) which are carried out by the thermostat constitute information? Alternatively, perhaps it’s the physical nature of a thermostat (its mechanical and material innards) that constitutes its information.

In terms of the thermostat at least, surely information is information-for-us, not information for the thermostat itself. After all, a thermostat responds to changes in temperature because we’ve designed it to do so…

Nonetheless, whatever a thermostat is doing (even if designed), it’s still doing. That is, the thermostat is acting on changes in temperature. (When it’s hot, it does one thing. And when it’s cold, it does another thing.)

Thus, does a thermostat have (to use John Searle’s term) as-if information? Or does it have real (first-order) information? In other words, does the fact that a thermostat is designed by human beings automatically stop it from having experiences which are themselves determined by its informational innards and/or nature?

To move away from thermostats.

Does the fact that a computer (or robot) is designed by human persons — and created out of synthetic materials — create any necessary or automatic problems for artificial consciousness?

After all, humans are also — in a strong, if metaphorical, sense — designed by their DNA, and we certainly have experiences.

Thermostats are designed by human persons: do the former have experiences too?

David Chalmers also tackled (way back in 1996) the case of the artificial neural network NETtalk, and asked us whether or not it does (or could) instantiate conscious experience. He wrote:

“NETTALK, then, is not an instantiation of conscious experience; it is only a model of it.”

John Searle had something to say on thermostats too:

“I say about my thermostat that it perceives changes in the temperature [].”

This means that this is Searle’s way (as with Daniel Dennett) of taking an intentional stance towards thermostats. That is, we can treat them — or take them — as being intentional objects. We can also take them as as-if intentional objects.

On Searle’s view, then, the as-if-intentional nature of thermostats is derived from the fact that these inanimate objects have been designed to (as it were) perceive, know and act. However, this is only as-if perception, as-if knowledge and as-if action. (All this involves as-if information too.) Thus, such things are dependent on human perception and human knowledge. Yet such as-if perception, as-if-knowledge and as-if-action require real — or “intrinsic”— intentionality.

This must mean that Chalmers’ thermostat has a degree of as-if intentionality too, which is derived from (our) intrinsic intentionality.

Now let’s jump from conscious thermostats to conscious nations.

The following passage is Anil Seth’s reference to the China brain thought experiment:

“The China brain thought experiment considers what would happen if each member of the Chinese nation were asked to simulate the action of one neuron in the brain, using telephones or walkie-talkies to simulate the axons and dendrites that connect neurons. Would this arrangement have a mind or consciousness in the same way that brains do?”

So, according to Seth, at the other end of the scale it’s also the case that “an entire country [could] be conscious”. What’s more, if that were the case, then we’d also need to decide if “one country [could] be more conscious than another”.

Let’s now move on from integrated information, and simply tackle information itself.

Is Information Fundamental?

Anil Seth discusses the idea that integrated information theorists see information as being fundamental. He writes:

[A]nother weirdness of IIT is that by making the strong claim that PHI *is* consciousness, IIT also implies that *information itself* exists — that it has some definite ontological status in our universe — a status like mass/energy and electrical charge.”

There are problems with this position.

The science writer Philip Ball quotes the words of the physicist Christopher Fuchs to express some of these problems. Firstly, Ball writes:

“Christopher Fuchs sees these insights as a necessary corrective to the way quantum information theory has tended to propagate the notion that information is something objective and real — which is to say, ontic."

Ball then quotes Fuchs directly:

“‘It is amazing how many people talk about information as if it is simply some new kind of objective quantity in physics, like energy, but measured in bits instead of ergs. You’ll often hear information spoken of as if it’s a new fluid that physics has only recently taken note of.’"

Finally, Ball sums up this issue with the following words:

“In contrast, [Fuchs] argues, what else can information possibly be except an expression of what we think we know?”

That passage can be read as arguing that stuff gives off information, rather than stuff actually being information in and of itself.

Yet this position conflicts with what some philosophers and physicists believe. That is, such people believe that information is in no way mind-dependent. Indeed, they believe that information is information regardless of minds, persons, observers, experiments, tests, etc.

So Fuchs is (at least partly) at one with the philosopher John Searle in rejecting this hypostatisation of information.

That said, information may well become (what Searle calls) information-for-us for such information-based physicists. Yet it’s still regarded as information even before it becomes information-for-us.

There is a midway position here, as expressed by the theoretical physicist Carlo Rovelli.

Rovelli writes very loosely about information here:

[T]he white ball ball in my hand is black. We’re dealing with physical facts, not mental notions. A ball has information, in this sense, even if the ball does not have mental states, just as a USB storage device contains information [].”

More directly on the theme of observers or scientists:

“But the effective way of continuing to exist in a changing environment is to manage correlations with the external world better, that is to say, information; to collect, store, transmit and elaborate information.”

Thus, information can be seen as being fundamental, and it can be tied to minds or observers too.

What’s more, some readers might have spotted that these passages from Rovelli tie in with Philip Ball’s earlier fundamental question:

[W]hat else can information possibly be except an expression of what we think we know?”

In other words, these commentators certainly don’t believe that information is (to use the words of Christopher Fuchs again)

“simply some new kind of objective quantity in physics, like energy”.

Saturday 31 August 2024

Anil Seth: Consciousness ≠ Integrated Information

Integrated information theorists postulate a literal identity between consciousness and integrated information. The British neuroscientist Anil Seth rejects this identity.

The British neuroscientist Anil Seth puts the integrated information theory (IIT) position at its most simple when he tells us that

“on IIT information — *integrated* information, Φ — actually *is* consciousness”.

According to the neuroscientist Giulio Tononi, the mathematical measure of that integrated information (in a system) is symbolised by φ (phi).

It can be presumed that many people (at least the ones who think about this issue) would claim that consciousness (to use a metaphor) contains information. That is, a conscious state has (or it instantiates) informational content. However, is consciousness itself information?…

An integrated information theorist may now simply ask:

What is this consciousness which contains information?

Alternatively:

If you take the informational content of consciousness away, then what are you left with?

As just stated, IIT is an identity theory that postulates a literal identity between consciousness and integrated information…

But not so quick!

Giulio Tononi actually believes that consciousness doesn’t equal just any kind of information. However, any kind of information (embodied in a system) may be conscious — at least to some degree.

Of course, we need to know what information actually is…

But, for now, we still can’t simply say that

consciousness = information

Instead, we must say:

consciousness = integrated information

Anil Seth himself makes it clear that integrated information theorists are identity theorists when it comes to consciousness and integrated information.

For example, Seth claims that such theorists treat consciousness “like temperature”, which is “mean molecular kinetic energy”. Thus:

temperature = mean molecular kinetic energy

is syntactically similar to

consciousness = integrated information

Seth doesn’t accept the second identity above. However, he does see the theoretical and experimental importance of information and integration in consciousness studies. He simply doesn’t see their joint instantiation as being equal to, or identical with, consciousness.

Anil Seth as an (Old-Style) Identity Theorist?

Anil Seth also refers to the (old?) identity theory in the philosophy of mind. Or at least he does so tangentially and in passing.

Seth does so when discussing his “preferred philosophical position”, which is “physicalism”. Seth writes:

“This is the idea that the universe is made of physical stuff, and that conscious states are either identical to, or somehow emerge from, particular arrangements of this physical stuff.”

So, according to Seth, physicalists can be identity theorists, or they can embrace emergence (i.e., and still be physicalists).

In detail.

In a video debate with Donald Hoffman (see here), Seth also stresses his newfound interest in emergence and top-down causation. This hints at the fact that he actually opts for the “emerge from” (i.e., rather than the “identical to”) option when it comes to the relation between consciousness and physical stuff. (Can emergent features themselves be identical with physical stuff?)

So Seth not only discusses the literal identity of consciousness and integrated information (which he doesn’t accept), he also hints at his own identity between “conscious experiences” and “neural mechanisms”. He writes:

“They key move made by Tononi and Edelman was to propose that if every conscious experience is both informative and unified at the level of phenomenology, *then the neural mechanisms underlying conscious experience should also exhibit both these properties*.”

More explicitly:

“That it is in virtue of expressing both of these properties that neural mechanisms do not merely correlate with, but actually account for, core phenomenological features of every conscious experience.”

What are readers to make of Seth’s claim that “neural mechanisms do not merely correlate with, but actually account for, core phenomenological features of every conscious experience”?

Arguably, this isn’t an explicit expression of an identity relation between conscious experiences and neural mechanisms. After all, Seth does use the words “account for”. Thus, those who stress the neural mechanisms which “correlate” with the “phenomenological features of every conscious experience” also say that the former account for the latter.

So is Seth also arguing that these neural mechanisms actually are the phenomenological features of every conscious experience? In other words, are the latter identical to the former?

That said, there are various grammatical phrases which Seth uses which point to the fact that he believes that his own position is not an identity theory… of any kind.

For example, take Seth’s words “neural basis” (i.e., as found in the clause “they made claims about the neural basis of every conscious experience”). Thus, if x is the basis of y, then x and y can’t be one and the same thing. (Seth’s words “underlying mechanisms” may also rule out an explicit identity.)

Giulio Tononi on Integrated Information

We can cite Giulio Tononi again as an example of someone who believes that consciousness (or experience) simply is integrated information. Or, perhaps more accurately, he believes that consciousness is information as it’s processed and integrated by brains and, perhaps, non-biological systems.

Thus, if that’s a statement of identity, then can we invert it and say this? -

information = consciousness

As stated earlier, Tononi actually believes that consciousness doesn’t equal just any kind of information. However, any kind of information (embodied in a system) may be conscious — at least to some degree.

Technically, not only are systems more than their combined parts: those systems have varying degrees of “informational integration”. Thus, the higher the informational integration, the more likely that system will be conscious.

Mere Correlations!

Anil Seth is certainly unsatisfied by (as the popular phrase has it) “mere correlations”. Or, at the very least, he wants to offer his readers more than that. He writes:

“The deeper problem is that *correlations* are not *explanations*. We all know that mere correlation does not establish causation, but it is also true that correlation falls short of explanation.”

Instead, Seth wants something that will close the “explanatory gap between the physical and the phenomenal”.

But does all that mean that Seth is — again — offering us identities?

Seth continues:

“But if we instead move beyond establishing correlations to discover explanations that connect properties of neural mechanisms to properties of subjective experience…, then this gap will narrow and might even disappear entirely.”

The philosopher David Chalmers tackled this issue many years ago (i.e., in 1995). Indeed, he even christened it with a new technical name: “structural coherence”. Chalmers himself wrote:

“This is a principle of coherence between the *structure of consciousness* and the *structure of awareness*.”

Yet, later, Chalmers also notes the problems here:

“This principle reflects the central fact even though cognitive processes do not conceptually entail facts about conscious experience [and] not all properties of experience are structural properties.”

It also needs to be stressed that Chalmers was noting structural coherences between conscious states and what he calls the “structure of awareness”, not between Seth’s “subjective experience” and “neural mechanisms”.

Despite those differences, we can say that if x is coherent with y, then x and y still can’t be one and the same thing. Thus, again, we don’t have any literal identities here.



Friday 23 August 2024

Rudolf Carnap: Truth Is Relative to Convention

 “The conditions of the truth of sentences in a system need not be found outside the system [or convention], but must be provided within it.”— Rudolf Carnap

To put the following essay in context, let’s firstly quote two passages from Rudolf Carnap’s ‘Empiricism, Semantics, and Ontology’, which was published in 1950:

“If someone wishes to speak in his language about a new kind of entities, he must introduce a system of new ways of speaking, subject to new rules; we shall call this procedure the construction of a linguistic framework for the new entities in question.”
“To accept the thing world means nothing more than to accept a certain form of language, in other words, to accept rules for formulating statements and for testing, accepting, or rejecting them.”

The German philosopher (a logical positivist) Rudolf Carnap argued that truth is an “internal question”. That is, truth is internal to specific “conventions”.

The following is a broad definition of conventionalism:

“Conventionalism is the philosophical attitude that fundamental principles of a certain kind are grounded on (explicit or implicit) agreements in society, rather than on external reality. [] Although this attitude is commonly held with respect to the rules of grammar, its application to the propositions of ethics, law, science, biology, mathematics, and logic is more controversial.”

So why is any given convention chosen in the first place?

More specifically, why did Rudolf Carnap himself adopt the “framework” of what he called the “thing world”? Why not one based on abstract objects or the pronouncements of goblins?

It can be argued that truth-exterior-to-convention (at least in some form) was always lurking in Carnap’s background — i.e., despite his protestations against “external questions” on the outside of all (or any) conventions.

Again, surely there is some (form of) truth that’s antecedent to the adoption of a Carnapian convention.

What’s more, is it true that Carnap’s adopted convention “works”, “provides results”, “solves problems”, “handles experience”, etc?

Pragmatists and/or instrumentalists must face these antecedent questions - even if they’ve already adopted various conventions.

All this also parallels the late-19th-century stance of the American pragmatists.

For example, is it true that belief P is (as William James put it) “better for us to believe”?

Is it true that pragmatism itself is better for us to believe?

Thus, can we apply the pragmatist test of truth to pragmatism itself without begging the question?

Surely there must be some kind of truth which is, in this case, external to pragmatism.

We can then ask a Carnapian similar questions.

What comes first: “the world” or the convention?

Metaphysical realists would answer, The world.

Carnap would (or might) have answered, Conventions

Or would he?

C = Convention

What would have made Carnap (or anyone else) adopt C¹ in the first place?

Carnap couldn’t have adopted C¹ from within C¹. It follows that his reasons for adopting C¹ weren’t internal to C¹. Therefore, when he adopted C¹, he must have been outside of C¹.

So where did these external reasons come from?

Didn’t truths, facts, reasons, evidence, data, etc. external to C¹ determine the choice of C¹?

Again, surely such reasons for a Carnapian rejecting his current convention couldn’t have come from C¹ itself. Thus he must have had good reasons for adopting C¹ which weren’t actually part of C¹.

So were they part of, say, C²?

Or are there many concurrent conventions which the Carnapian is internal to?

We may indeed require conventions, language games, conceptual schemes, etc. However, are we internal to just one of them?

More particularly, if conventions determine what we think and what we say, then conventions must determine our adoption of other conventions.

However, there’s something strange about this conclusion because conventions (or conceptual schemes) are often supposed to be “self-contained” (i.e., “closed universes”). At least many people have believed that. Yet this can’t be true if, for example, we can move from C¹ to C².

Thus, in the contemporary literature, why are conventions (or conceptual schemes) deemed to be so powerful, restricting and important?

We may indeed need conventions in the Carnapian sense. However, we can still jump from one convention to another. Conventions aren’t, therefore, anything like the Kant’s categories which a priori determine how we must perceive, experience or talk about the world. (Relevantly, the French philosopher Michel Foucault believed in what he called the “historical a priori”.)

Conventions are thoroughly contingent and adaptable.

On another point.

Was Carnap’s “adopted framework” a posit itself (i.e., a posit within which he posited other things)? Wouldn’t Carnap’s adopted framework itself (as an abstract object) need to have been posited via another adopted framework?…

And so on.

If this isn’t a real regress, then how do things get started?

If it’s not possible to posit anything without an adopted framework (or a convention), then perhaps it isn’t possible to posit an adopted framework unless that too belongs to another adopted framework. Contrary to this, if it’s possible to posit a convention which isn’t itself posited by another convention, then perhaps we can posit “medium-sized dry goods” without an adopted framework or convention.

So a thing from Carnap’s thing-language may be as autonomous as the adopted framework itself.

To generalise and repeat.

If conventions are necessary for the postulation of objects, events, facts, etc., then conventions themselves (as abstract objects) may need to be posited by other conventions… ad infinitum.

In detail.

What convention does a new convention belong to?

It can’t be an unmoved mover or cause of itself.

So does a convention need another convention to legitimise it as a genuine abstract object ?

If things, truths, facts, evidence, etc. are “relative to” conventions, then what are conventions themselves relative to? Are they relative to themselves? Have they come into being ex nihilo? Or are they relative to meta- (or second-order) conventions?

In that case, what are these meta-conventions relative to?

The same kind of problem can be seen at a smaller scale.

A convention (C¹) may offer the following statement:

Statement S is assertible according to C¹.

The assertibility of S is fine. However, what about the entire meta-statement directly above? What is the meta-statement assertible according to? The following? -

The meta-statement is assertible according to C¹, in which S is assertible according to C¹.

This is a more refined regress than the earlier regress. Or we could have the following:

The following is assertible in C¹: “‘…’ is assertible in C¹”.

Again, here we have an example of self-reference, which was earlier applicable to the convention itself.

External Truth Again

The possibility that there may be a tacit (or implicit) commitment to external truth (i.e., in relation to Carnapian conventions) was hinted at earlier. Yet Carnap himself had his own position on truth being entirely(?) internal to conventions.

Carnap wrote the following in his Introduction to Semantics:

“The semantic ‘definition’ [of truth] is not a definition of truth, but a criterion of the adequacy (accord with our intentions) of a predicate for the concept of truth within a given system. ‘True’ thus becomes a predicate applicable by the rules of a system to sentences of the system. [] In pure semantics the conditions of the truth of sentences in a system need not be found outside the system but must be provided within it.”

In Carnap’s scheme, then, just as the truth of the thing-language is an internal question, so too is the question of truth itself before it’s applied to a thing-language… or to anything else for that matter.

Carnap also referred to the “adequacy” of the “truth predicate”, rather than simply saying it “works”, “provides results”, “solves problems”, “handles experience”, etc. Yet now all the previous arguments about the adoption — and status — of a Carnapian convention are now applicable to Carnap’s internalist account of truth itself.