By Sylvain Magne - 2016
I have recently come across Georges C. Williams’ arguments for why genes and memes should be defined as information, in his 1992 book Natural Selection. I had seen quotes from Williams made by many proponents of the meme idea but had yet to see for myself how he argued that information is what genes and memes are made of. I found that Williams had a great insight into the true nature of replicators, maybe the most precise understanding that I have come across so far. However, I also believe that Williams made a slight but important mistake in his argumentation. The mistake is that Williams is confusing two very different understandings of the concept of information.
Although Williams’ mistake may not appear as a major issue at first, it may be one of the reasons why progress with memetics has been slow. Consequently, many memeticists have been trying to build upon this idea with limited success. As a result, still today, no one seems to agree on what a meme really is. Despite that, I was pleasantly surprised to see how close my views are to Williams’. I was also pleased to see that he took the time to try and argue his views, while others that I have read merely quoted Williams without reviewing or challenging his arguments. The major point that Williams was arguing is that genes and memes are best described as information. To be fair he was not the first one to make this point, but his contribution seems to be the most prevalent. Unfortunately, I fear there is much misunderstanding about what information is and, like others before and after him, confusions about the concept of information have lead him to wrong conclusions.
The parts of the book I am reviewing here are the following chapters:
2 - The gene as a unit of selection (page 10)
2.1 - A general model of selection in the codical model. (page 13)
Williams seeks to make a clear difference between two different kinds of entities that play a role in evolution. Those are the replicators and the products of replicators which he calls interactors. For example, genes and memes are on the side of replicators, while the bodies of living things are on the side of interactors. These interactors are also known as vehicles, phenotypes, or gene (and meme) machines. I personally chose to call them readers but for the sake of this article I will use the term interactors.
Here’s what Williams says:
(p 10) “I suggest that these terms should refer to two mutually exclusive domains of selection, one that deals with material entities and another that deals with information and might be termed the codical domain. A unit of selection in the codical domain would be a codex (codices is the only plural in my dictionary).”
So far I hadn’t seen any memeticist use the terms “codical domain” and “codex” in this context, other than myself (admittedly I haven’t read every article about replicators). I use terms such as “coding systems” and “codes” instead, but they refer to the same ideas. The Wiktionary defines codical as “Of or pertaining to a code or codex” (https://en.wiktionary.org/wiki/codical). Indeed, I am very pleased that Williams understood that genes and memes are in a world of codes. This is exactly what I have argued and it is crucial to making progress with memetics. However, in the same sentence, Williams uses a word that will only confuse the matter. This word is “information”. I want to argue that Williams’ understanding of the relation between information and codes is unfortunately mistaken. Williams seems to put too many conflicting concepts under the one concept of information.
First he seems to put in the same bag such concepts as codes and meaning:
(p 10) “In discussing the codical domain we use such terms as bits, redundancy, fidelity, and meaning.”
Indeed codes can be described accurately in terms of patterns displayed in a particular medium (or substrate). For example, codes are sequences of nucleotides, or arrangements of ink on paper or sequences of air pressures, or sequences of bumps and holes on a CD, etc. We can indeed measure their redundancy, or their entropy in bits. Their fidelity is related to such things as the resilience and durability of the medium or the coding syntax. All these values can be measured objectively and are independent from an observer. However, the meaning of a code is entirely subjective and depends on who or what is reading the code. The meaning of a code is not a property of the code, it is whatever the reader makes out from the code. The meaning can differ greatly from reader to reader. So the term “meaning” should be used carefully in this context. That is a first indication of Williams’ confusion of terms.
Then Williams describes the second domain as follows, again putting together conflicting concepts:
(p 10) “The material domain is described by color, charge, density, volume, etc.”
Indeed, charge, density and volume are, again, objective properties of the material world, but it is not the case with colours. I am being picky here because I am sure Williams may have just as well said “wavelength” instead of “color”, but there we have it, this is symptomatic of the whole misunderstanding between objective reality and subjective reality. Colours are subjective, and there is no way to know whether any of us experience the colour red in the same way. Wavelengths on the other hand are objective. I argue that information is on the side of subjective reality and therefore it cannot be said that genes, memes or any replicators are informational. I also argue that codes are on the side of objective reality and can qualify as replicators.
This said, don’t get me wrong, I generally agree with Williams that we should establish a dichotomy between the physical world and the codical world. This had to be done and Williams made the effort to argue for it. I have myself argued this point even further. I have explained that there are mathematical arguments for why we can regard the entire universe as being made of codes and machines, and for why it is relevant to the study of evolution. For the full argumentation please visit the following link:
However this dichotomy between codes and matter is not necessarily exactly the most relevant with regard to evolution in particular. There is a slightly more relevant dichotomy which consists in separating replicators (a special type of codes) from other codes and interactors. I will come back to it later in this document.
So to sum it up, yes there is a difference between the world of codes (codical domain) and the world of physical reality (material domain), and this difference is important. Now if Williams kept it at that, I think it would have been good, but Williams also used the term information, and with information comes trouble.
Here’s what Williams says about information.
(p 10) “Information can exist only as a material pattern, but the same information can be recorded by a variety of patterns in many different kinds of material.”
(p 15) “Information, unlike objects, can be recognized as having meanings of various kinds, including moral meanings.”
This is probably the most common view of information among memeticists today. However, this is strongly misleading. If we are to take this statement literally it sounds like information is really something that hops from medium to medium, and therefore such concepts as moral meanings would be able to move around. This idea is very much part of our culture today. Everybody will tell you that we live in the age of information and that internet is the information highway. However, this common view of information as something that travels is naive and simplistic. This view has two sides. One is about the nature of information, and the second is about the apparent continuity of information, let’s take a closer look at those.
Is Williams right when he says that information can exist only as a material pattern?
Information is much like the colour red. Everybody can see red, we all agree it is there and yet there is no such thing as “red” in the real objective world. Information also exists only inside our subjective virtual minds. Red is our interpretation of a certain pattern of light that we project back onto the world. It takes an observer for information to exist just like it takes a person for “red” to exist. If there is no observer, there is no information.
Here’s what actually happens when we supposedly put information into a pattern. Because our brains are good at recognising patterns, we can learn to associate patterns with mental representations such as ideas or memories, emotions, mental images, etc. For example, the word “tree” can trigger images of trees in our minds. So when we wish to communicate a mental representation, i.e. we want someone else to have the same subjective experience, we simply craft a message by reusing the very same patterns that we have already associated with those ideas. For example, we will say or write the word “tree” again. The patterns that we then create do not contain our ideas, they only contain patterns that can trigger those ideas. Then we pass on this message full of “triggering patterns” onto another person and we can only hope that this person will be triggered in the same way that we have been.
We need to make a clear distinction between the objective world and the subjective world. In the objective world there are patterns, and in the subjective world there is what we make of those patterns. So when a message exists, depending on which end of the message you find yourself dealing with, information can be either the interpretation intended when crafting the code or the interpretation made when reading the code. Those two don’t always match.
Why would Williams think that the same information can be recorded by a variety of patterns in many different kinds of material?
There is an undeniable continuity between Don Quixote as a book and Don Quixote as a computer file. This perceived continuity across media is a mental projection of the fact that we know that one version can be transcribed into the other version, and more importantly, into a version that we can understand. That is why we are quick to think that “they all contain the same information”. But as we have seen there is no such thing as an information that can be contained in a pattern, simply because information is subjective.
For example, what are we to make of such a pattern as “911”? The person that crafted this pattern may have intended to trigger the idea in our mind of a tragic terror attack. However, when we read it, something very different may come to mind. We might be thinking that this is an emergency phone number, or we might think that this is a famous sports car’s name. The reason we can interpret this differently is because it does not contain the information that we think it does. The meaning of this relies entirely on the mind of the observer. The information it contains is subjective and unique to each person receiving the pattern.
There is a better way to describe the apparent continuity of codes, and that is “transcoding”. Because patterns can be transcoded into other patterns, when two patterns can be mutually transcoded, we can call them transcodes. Each transcode is linked to all other trancodes by a chain of transcoding that we can call a codical chain. I have used the term “causal chain” before and Williams has himself used the term “chain of transcriptions” (p13), but I have come to like Williams’ use of the term “codical” for it is more specific to codes, so I might adopt his term “codical domain” and add the new term “codical chain” to the lexicon.
Now, in general, I don’t mind if we pretend that information travels. It can make things simpler. It is indeed convenient to be able to say “Let’s gather some information” or say “that’s a good piece of information”. However, memeticists will need to be careful when doing so, or should try and avoid doing it altogether. There may be some better terms we could use.
The meaning of information has changed over time and I tend to prefer an older understanding of the term. This is what I found on wikipedia about the etymology of information (my emphasis):
“The English word was apparently derived from the Latin stem (information-) of the nominative (informatio): this noun is derived from the verb informare (to inform) in the sense of "to give form to the mind", "to discipline", "instruct", "teach". Inform itself comes (via French informer) from the Latin verb informare, which means to give form, or to form an idea of.”
This idea that we can form the mind by using codes is pretty much exactly how I see it. Here, information is not an entity as we see it today, but an act, the act of “in-forming” as in “giving form to the mind”. I suggest that instead of saying that a code (or pattern) contains or carries information, we should say that a code is informative, in the sense that it has an informational potential. It means that a code can cause meaning in an observer’s brain with a certain probability. A codical chain would have the property to preserve the informational potential of a specific code until it is copied into its original form. However two transcodes may not have the same informational potential. Indeed, two transcodes are not the “same code”. People can read a book but no one can read its digital version until it is printed again. A digital file has a high informational potential for a computer but not for a person. “Tree” is highly informative for English speaking people while “arbre” is highly informative for French speaking people.
If I had to make up my own modern definition of information for memeticists, it may look like this:
“Information is a subjective representation caused by a particular physical pattern.”
On a side note, some memeticists have described information as being abstract. There is much confusion around this term as well, but I don’t mind using the word “abstract” as long as it is understood in the way as it is described in the online Oxford Dictionary. They define abstract as “existing in thought or as an idea but not having a physical or concrete existence”. This definition seems rather in line with the understanding that information is subjective. However I would like to point out that what happens in our minds is still based on concrete physical processes.
It is quite unfortunate that information can be understood in quite a few different ways. It can really confuse things. For example, information has a different meaning in Physics, Information Science, Computer Science, Information Theory and everyday life. In memetics we usually refer to information in two different ways. On one hand, the popular understanding of information is all about meaning and subjective interpretation. On the other hand, in Information Theory it is all about codes, predictability, fidelity and objective reality. I want to emphasise that information theory is not concerned with meaning or interpretation. Let’s have a look at some definitions from the online Oxford dictionary.
This is how they define Information Theory:
“[Information is] the mathematical study of the coding of information in the form of sequences of symbols, impulses, etc. and of how rapidly such information can be transmitted, for example through computer circuits or telecommunications channels.”
And here are some definitions of information related to Information Theory:
“2 - [Information is] what is conveyed or represented by a particular arrangement or sequence of things. [...]"
“2.2 - [Information is] (in information theory) a mathematical quantity expressing the probability of occurrence of a particular sequence of symbols, impulses, etc., as contrasted with that of alternative sequences.”
Definition 2 is close to how Information Theory understands information. Note how it says nothing substantial about information, it only describes it by its medium. It is a bit like a car maker saying that a “person” is “what drives a car”. That would say pretty much nothing about a person, because it is seen from the perspective of the car maker. Information Theory doesn’t really know or say anything about information, it does not care whether information is a subjective concept and whether it has a meaning or not. Information Theory is only concerned with codes and such things as the amount of bits they contain.
Definition 2.2 makes that fact quite plain. I admit that it is rather confusing that Information Theory is not actually a theory of information. Shannon’s theory is also known as “Communication Theory”, which I think is a much better name for it.
So, the Theory of Information will be of no help when trying to study the intended or perceived meaning of a message. Also, those who think that replicators are made of information (and not codes) will not be able to make use of Shanon’s theory. Remember that the car engineer does not know who drives his cars. So if we understand that replicators are codes, then we can use Shannon’s Theory of information to study replicators.
I mentioned earlier how Williams was trying to establish a dichotomy between the world of codes (codical domain) and the world of matter (physical model). The aim of this effort was to differentiate between the world of replicators and the world of their machines (interactors) in order to better understand the genes and memes as codes (codices). Separating between codes and machines is not necessarily the most relevant from the perspective of evolution. Indeed, not all codes are replicators. Evolution is more concerned with establishing the true units of selection, the replicators.
For codes to be preserved, they rely on their ability to be transcoded and copied. When it comes to genes we are used to thinking in terms of germline. A gene in a gamete will be copied into a gene inside a body, quite a few times, until it is copied back into a gene in a gamete. However memes go through many different types of codes. They can go from brain codes to sound codes, to ink on paper codes, to computer codes, to electromagnetic codes, etc. The question is; which of those are true replicators? Most memeticists today tend to assume that each “transcode” of a meme is also a meme. Although some memeticists argue that memes are only in brain codes. However Williams is the only one I have read so far who suggested that a transcription may only be considered a replicator when it has been encoded in the same medium as the original (and with the same patterns I would add). This is how he puts it:
(p 12) “While I think it important to keep separate the codical and material domains, I would not insist that the codex concept completely replace the idea of a replicator. Perhaps the term replicator could apply only when the same message is copied into the same medium.”
Williams then gives a couple of examples but does not elaborate much further on this. I think Williams got very close to the right idea here. It may look like we are arguing over meaningless details here, but this is very important in my view. Williams himself said (p 12) “This is a topic that warrants serious thought.” Indeed, we need to push it a little further to see why it is important.
I also believe that not all codes found on those codical chains qualify as replicators, except my argumentation is slightly different and I argue that we can determine exactly which codes in those chains are the real replicators. I have discussed this point before although I have described those codical chains as causal chains. You can read the article at the following address, chapter “8 - Copying, Transcoding and Causal chains”:
What we come to realise, I argue, is that it is the codes which the “interactors” exchange that are the real potential replicators. Indeed the interactors are all unique and varied, and while they may use different ways to store codes inside them or inside artefacts, when it comes to communication, the codes exchanged need to follow a common protocol. It is the fact that codes need to be recognisable by all interactors that allows them to have the right characteristics to be replicators. That is why memes will be carried by sound waves or light waves because sounds and light is what we use to communicate. In the realm of computers, replicators can have many media such as light waves, sound waves, radio waves, electrical signals, etc. And again, the true replicators are not inside the computer’s memory discs but are traveling between computers, where they comply with inter-computer communication standards. If we apply this reasoning to biology, we could find that only the genes that are stored inside the gametes actually qualify as replicators. It just so happens that the way bodies store their code is the same as the code they use for reproduction. In that sense, biology is a special case.
We need to make a clear distinction between transcoding and copying. We easily claim that a digital version of Don Quixote is a copy but it isn’t, it is only a transcode. Again it is our subjective perspective that compels us to see them as copies, but in the objective world, and in the eyes of evolution, they are two different entities that compete with each other. This is the case with synonyms, this is the case with two ways of making the same knot, this is the case also with any word and its translation into other languages. A good theory of evolution is concerned with explaining objectively why some patterns in nature are reoccuring and how they evolve. Therefore, making this difference is crucial to memeticists.
I hope I have made it clear that information is an often misunderstood and a misused concept, particularly among memeticists. Information is best understood as a subjective experience that is confined to our minds (and computers’ “minds”) and does not travel into the outside world. The only thing that travels between brains are codes. Codes can be said to be informative for they have a potential to cause meaning. Codes can be transcoded into other codes and form long codical chains of trancodes. However, these transcodes are not to be considered equal in the eyes of evolution as they compete for survival. The codes that end up being truly copied, i.e. with the same medium and the same recognisable pattern, are the replicators. Replicators are typically found to be among codes exchanged between interactors. Information theory can be useful for memeticists because it studies some of the codes’ properties.