8: Information and Semantics

Table of contents

1. Note
2. Review
3. New questions for the field of Philosophy of Information
4. Dretske’s Approach (as overviewed by Philosophy of Information, sections 3.3-3.5)
   1. Environmental Information (section 3.3)
   2. Equivocation (section 3.4)
   3. Digitalization and semanticisation (section 3.5)
   4. In sum

---

Note

Today, we’re going to start by finishing up our discussion of information theory’s impact on biology, and then I’d like to make sure we have a significant amount of time to begin reviewing the work we’ve done over the past few weeks. So please make sure to look at the posted exam review sheet before class tomorrow. (And of course please annotate the review notes as much as possible).

This means we will not have a lot of in-class time to discuss today’s focus reading () on Dretske’s an attempt to explain the “naturalization of semantic content”

Nevertheless, there are couple of bullet points on the review sheet related to this reading and the material described below.

In short, what I want you to have a general sense of is the basic debate taking place in the field of philosophy of language and philosophy of information. This debate is closely related to the question of whether machines are thinking when they process information or whether humans are doing something special when they “think” that machines simply cannot do.

One side of the debate is captured by John Searle’s famous “Chinese Room Thought Experiment” for which there is a video below (please make sure to watch this video).

Dretske’s response is an attempt to explain how a “natural/automated/machine” understanding of semantics might be possible. His account (as explained by Ilari) is admittedly quite difficult. My notes have tried to break down the reading into smaller chunks, but it is still quite challenging. As goal, try to simply come away with a general sense of how Dretske think semantics could be understood in “natural” way, and therefore why such an understanding, contra Searle, need not be a unique possession of human beings.

---

Review

We’ve spent some time now thinking about information as a kind of abstract “medium” that allows communication across what we’re previously seen as incompatible communication channels.

We’ve spent time over the last week or so thinking about how this abstraction allows for a common language (the language of information) to approach problems in what previously seemed to be diverse fields, for example, psychology, physics, and biology.

What made this possible, of course, was the kind of abstraction performed by Shannon. His ability to identify information as something identifiably distinct from both the meaning transmitted via this information and from the matter that encoded this information was what made this kind of universal processing possible.

New questions for the field of Philosophy of Information

But this abstraction has also raised new questions for the still emerging field of Philosophy of Information and related fields like the Philosophy of Language and Philosophy of Mind.

The central problem or question lingering for philosophers working in this field is: if information is something divorced from meaning (or “semantics”), how is it that human beings, as receivers at the end of the communication channel, receive, not just information, but meaning.

These receivers are referred to in our reading as “semantically enabled agents”.

Again, one of the features of Shannon’s new idea of information is that, because it abstracted away from semantics, it allowed us to see communication as something that ocurred without “semantically enabled agents”. This is what allowed us to see many seemingly diverse problems as instances of a common problem of communication. The removal of “semantically enabled agents” allowed for communication across diverse media at an extremely low cost and at astounding speed.

Our reading captures this fairly well when its says:

However, we also observe that the use of the word “information” is not confined to such cases involving semantically enabled senders and receivers of messages. For example, we speak of information being transferred when referring to the exchange of data between fax machines, or between clients and servers that transmit data by means of internet protocols. In both cases, the machines involved do not understand the information in the way we do. Moreover, we use the word “information” even for systems that are not constructed by human engineers, like the transfer of genetic information between chromosomes of the cell nucleus and the ribosome that produces the polypeptides according to the information encoded in the messenger RNA. In a wider sense, the fact of the past existence of animals can be communicated to a sedimentary layer, which preserves this fact as a fossil, or, in other words, as a record, which in turn can be considered as a kind of information. Likewise, the life story of a tree is communicated to its rings, and thereby they hold information about the tree’s age. In these examples there are no senders and receivers making sense of symbolic language, but we still comfortably describe the examples by referring to “information transfer”.

The idea that information can be communicated apart from semantics is the first step in a research known as “the naturalization of information”.

The second question is whether this “naturalization” can be extended to the second step in human communication, namely to the extraction of meaning or semantics too? Here we might ask questions like:

• Can this process be automated?
• Is it something machine’s can do as well?
• Or is a “magic process” that humans alone are capable of?
• Or are human beings the only type of “semantically enabled agents”?

Before we consider why Fred Dretske thinks this might be possible, let’s look at why some people think it is not possible. A common thought experiment invoked against this idea of “naturalization” is John Searle’s thought experiment known as the “Chinese Room”.

You can also read more about his experiment here in the article, “The Chinese Room Argument”, Stanford Encyclopedia of Philosophy

Now let’s turn to some ideas of the ideas that suggest the “naturalization of semantics” might be possible.

Our reading hints at a test for answering the question of when we could say that someone or something has understood semantics (p. 44, bottom).

What is this possible test or method for determining whether semantics and not just information has been communicated?

The reading points to the “problem of reference”.

If a receiver can receive information and then accurately locate a correspondence between this information and a referent, then we might be inclined to say that this is a case in which “semantics” have been communicated.

A key question lies in what is required of information and a receiver in order to have this effect.

It’s worth pausing here to highlight that these are question that have actual applications in various fields of active research and development. Our author writes:

Understanding the fundamentals of this problem is beneficial to appreciate current debates on neuro-physiology (“When does the transduction of information stop and when does the interpretation start within the brain?”), artificial intelligence (“Can we build semantically enabled artificial agents?”), and metaphysics (“Is information the ultimate substance of the universe?”).

Our goal here is only to get the smallest glimpse at what is going on in these fields, while hopefully alerting you to the fact that these areas of research exist. A quick google search and you will find a vast array of readings that will let you explore these topics further.

Dretske’s Approach (as overviewed by Philosophy of Information, sections 3.3-3.5)

Environmental Information (section 3.3)

As discussed, the question of communication is theorized as twofold.

The process of communicating information and the extraction of semantics.

While separate, it is assumed that they are related.

Thus, the concept of “environmental information” is inspected first.

So, as our text states:

“Environmental information covers the first part of the naturalization, the natural transfer of information. The tree’s age is communicated by its own physical structure. But an observer must still interpret the structure to extract the information”

“Environmental Information” seems to cover the transfer of information, but we need to question how an agent can interpret that information, identify a reference, or identify a meaning (which is perhaps three ways of saying the same thing).

Equivocation (section 3.4)

Dretske first shows that even humans (“semantically enabled agents”) often communicate information without semantics.

What example is taken from Dretske?

“On the other hand, an event or state of affairs that has no meaning in any conventional sense may carry substantial amounts of information. An experienced poker player can read the signs; he can tell, or be reasonably certain, when a bluff is in progress. His opponent’s nervous mannerisms, excessively high bet, and forced appearance of confidence reveal, as surely as would a peek at his last card, that he did not fill his inside straight. In such a situation information is communicated, but the vehicle that carries the information (opponent’s behaviour and actions) has no meaning in the relevant conventional or semantic sense.” (Dretske, 1981) See,

In such a process, Dretske highlights the importance of a direct coupling between two states which he says is “unequivocal”.

But what then does “equivocation” mean?

“Equivocation occurs if several different events at the source are conflated into one observed event with the receiver, thereby rendering it impossible to infer exactly which event at the source produced the event with the receiver.”

This is useful to understand Dretske’s difference between “causal transfer” and “information transfer”. When there is “equivocation”, then these multiple states at the source can be said to be a “causal transfer” but not an “information transfer” (see p. 46).

It is helpful to remember that for Dretske: “every information transfer is a causal transfer, but not every causal transfer is an information transfer.”

In sum:

“At any rate, it should be clear by now what it means for environmental information to pass through a channel that is equivocation-free: it means that information is not irreversibly lost.”

We are still not talking about the extraction of “Semantic Information”, but the idea is that if there is going to be “Semantic extraction”, we’re going to need “unequivocal information transfer”.

As our text says at the end of section 3.4

“Equivocation-free transfer of information therefore prepares the way for the extraction of the content from the signal.”

Digitalization and semanticisation (section 3.5)

Even knowing that semantic extraction requires “unequivocal transfer” is still not enough to get us to “semantic extraction” because environmental information can often communicate unequivocally without the need for semantics.

Thus in addition to “unequivocal transfer” we need something more. It is at this moment that Dretske wants to speak of “intentional states” stating:

“Any propositional content exhibiting the third order of intentionality is a semantic content.”

Obviously, we need some clarity about we he means by an intentional state and what it means to be the third order.

The above quotation gives us some clues: an intentional state is a “state of mental contents”, a way of having and arranging different mental beliefs at any given point in time. The state of these “mental contents” may or may not justify different inferences or downstream behaviors.

Another clue is found when the reading says:

“the experience of a sound we hear is intentional with respect to the object that produces the sound, not with respect to the vibration of our ear drums, although the latter is implied by our having the experience.”

So for example, when my ear senses the various vibrations of a bird singing, and then I go the next step to form an idea or intention, I am hearing the “a song”, then I am in a particular intentional state.

But by itself:

“an intentional state is not sufficient to form a semantic structure, i.e. one that is capable of forming a belief state.”

The suggestion is that there is another step to be taken. By itself, the intention that I am hearing a song (t is in state f) may not yet be sufficient to form a belief, or what might be said otherwise, to extract semantic meaning from my intention. For example: “this is a bird’s song” (t is G)

Merely, holding the intention that there is a song (or t is in state f) may not be enough draw out the semantics (t is G).

What is it then specifically about what Dretske calls the third state of intentionality that allows us to draw this connection and form a belief that the first state of intentionality does not allow?

The final step is the development of a distinction between communicated information about matter of fact, “a bird is making melodic noise over there” (t is f) and type level information, the concept of a “song” allowing one to say, “this is a bird’s song.”

“Now, with regards to how natural information can give rise to symbolic communication, another important idea is required: the difference between a piece of information about a particular matter of fact, and the corresponding content the information has on a type level.”

Intentional states of the third order create intentions about these “types” and form beliefs about the relationship of individuals to concepts.

These concepts can be agreed to ahead of time, or they can emerge organically through conventional practice. (I would call this a ‘protocol’)

“A meaning is a symbol connected to a concept on type level. The symbol can represent a concept in the sense of a genuine convention, but this is not necessarily so, since a symbol can gradually acquire its meaning.”

The point is: once these concepts have a emerged, intentional states can follow which correlate one state of affairs with a concept classification. This is, in a sense, Dretske’s attempt to explain the extraction of meaning or the extraction of semantics “naturally” rather than assuming the human mind has magical powers to understanding “meaning”.

As the reading state:

“Knowledge of concepts is a form of type-level knowledge, and that is what is required for symbolic communication.”

In sum

I recognize this is hard reading. Once again our goal is to get a sense of the debate. So even if every move Dretske is making is not clear to us, let’s step back a bit from the details and try to get a sense of the whole?

• What is Dretske trying to explain?
• What view is Dretske opposing? Why does this opposing view think the “naturalization of semantic content” is impossible?
• What does Dretske think will be a sign that someone or something understands semantics/meaning?
• How does the invocation of type-level information become a way for Dretske to describe when something has understood a “meaning” (This is the probably the most difficult question for us, but, given our goals, the precise details of this move are the least important.)