Last updated on 28 Apr 2014
In seeking tales and informations [Henry VIII, Act V, scene 3]
For some time now* I have had problems with the notion of information. Not, please note, with this or that piece of information, but with the notion itself, especially in the natural sciences. In this age of computers and internets, we have taken to mistaking the thing described for the thing itself, and treat information as a property out there in the world, not a representation in our heads and language.
Let me set the scene. Back when Dawkins wrote about biology, he proposed the idea that genes were a special case of what he called the Replicator:
A replicator may be defined as any entity in the universe of which copies are made. [“Replicators and Vehicles” 1982]
Notice the word “copy”. I can copy things in a number of ways, from imitation to tracing, but Dawkins has a particular sense in mind, which he explored in an especially purple prose passage in The Blind Watchmaker (1986):
It is raining DNA outside. … [downy seeds from willow trees] The cotton wool is mostly made of cellulose, and it dwarfs the tiny capsule that contains the DNA, the genetic information. The DNA content must be a small proportion of the total, so why did I say that it was raining DNA rather than cellulose? The answer is that it is the DNA that matters… whose coded characters spell out specific instructions for building willow trees… It is raining instructions out there, it’s raining programs; it’s raining tree-growing, fluff spreading, algorithms. That is not a metaphor, it is the plain truth. It couldn’t be any plainer if it were raining floppy disks. [Chapter 5, p 111]
DNA, and the replicators they are a special case of, are information. This is not a metaphor. Similarly, physicists will occasionally assert the same claim about physical things. The physical world is just a mathematical construct, and things like electrons have only mathematical properties, says Max Tegmark, a physicist at MIT:
… all the properties that electrons have are purely mathematical. It’s just a list of numbers. So in that sense, an electron is a purely mathematical object. In fact, there’s no evidence right now that there’s anything at all in our universe that is not mathematical.
We get the “it” from “bit”. Some mischievous philosophers have even suggested that we do, actually, live in the Matrix, although what the Matrix lives in is unclear…
So, why am I unhappy? Is this wrong? I think it is.
To get at this I need to hit you with a little bit of natural philosophy from the Greeks, in particular from Plato and Aristotle. Plato famously proposed that the real world was the world of Forms, or ideas (the Greek word he used was ideai, from the root eidos, meaning appearance, something seen). Forms were more real than what you see around you. A physical circle is at best an imperfect instantiation of the real circle, which exists nowhere in physical form.
Aristotle, in contrast, explained the physical things in the world by supposing that they had matter, which filled space and gave weight (made from several admixtures of the four elements, two light and two heavy) which the scholastics called substance (substantia, meaning that which stands under), and form, the structure and mathematical properties of a thing. This matter/form dualism is called hylomorphism, from the two Greek words hule, meaning stuff (it originally meant “wood”) and morphe, or form. Hylomorphism was intended to be an alternative view to atomistic materialism, which had become a widely held (and generally atheistical) view in his day. Epicurus, his contemporary, had an entire philosophical school based upon the older Democritan atomism [see this excellent review just revised in the Stanford Encyclopedia].
Now hylomorphism was roundly demolished as a scientific hypothesis when Daltonian elements were named and investigated in the nineteenth century. By 1900, terms like “substance” (for matter that is propertyless apart from mass and extension in space) and “form” had taken on a largely philosophical sense that differed extensively from Aristotle’s own views. Instead, an increasingly elaborate atomism had won the day, far beyond anything Epicurus or Democritus had posited. The properties of things, including their mass and filling of space, were the result of fields in space-time.
And yet, a kind of hylomorphism remained, even in science. Biologists argued that form determined many properties of organisms in ways that could not be reduced to their parts, and this kind of thinking remained and was co-opted by the molecular biologists and geneticists in the 1950s, especially since around that time, computers were getting going and information was a hot topic (it had not been much prior to that time). Thus, we get the “Central Dogma” in genetics:
The central dogma of molecular biology deals with the detailed residue-by-residue transfer of sequential information. It states that such information cannot be transferred back from protein to either protein or nucleic acid. [Crick in 1958]
Read physically, this means only that the structure of the DNA molecule is not reproduced from the structure of proteins, a perfectly reasonable account of the molecular processes. But because Crick used the word “information”, some scientists, including Dawkins, took this to mean genes are informational entities, that “code” for organismic traits from the molecular level up to the entire organism and even beyond.
Likewise, physicists like Tegmark, Wheeler and so on, who say that the physical world is “just” information can be read as saying that there is physical structure in the world. This is harmless.
But this is not how “information” is interpreted usually. Instead, we get the idea that information is itself a kind of universal property that underlies all physical things. Information, which is the modern equivalent of Aristotle’s morphe, has become the equivalent also of Aristotle’s hule. And this is where I part company.
When scientists talk about the information content or the informational entropy of something, they can mean several things. They can mean the entropy of the string of characters that are used to measure or describe that thing, like a mathematical description of a process, or a sequence of symbols like the G, T, A, and C, of DNA. But DNA is not composed of G, T, A, or C. It is composed of molecules, nucleotides, that bear the symbols as names, and they have properties that mean that occasionally they do not follow the mathematical or semantic descriptions of these names. For example, there is a “fifth nucleotide”, 5-methyldeoxycytidine (5-mC), which can pair up in 5′-CG-3′ dinucleotide positions. 5-mC is a molecule in methylation molecules, which themselves modify the expression of the DNA. The informational entropy (or information content) of a sequence is therefore just a measure of how the DNA is represented. 5-mC can even breech the Central Dogma.
The point here is that the representation abstracts away from the physical properties of the molecules. Measures of the informational entropy are therefore actually measures of the abstract representations, not the things themselves. But suppose we had a simulation or representation of the things right down to the level of quarks (if that were physically possible): would the informational entropy of the objects be identical to the physical properties? Would we have the physical informational entropy of the objects? I’ll get back to that. First I’d like to consider some of the other meanings of “information” in science.
Another meaning of “information” is the semantic meaning: what one thing (e.g., a gene) represents (e.g., the phenotype). This is the “information as signal” view, based loosely or strictly on Shannon’s Communication Theory account. One thing “refers” to another (in Shannon’s theory, the received message “refers” to the sent message). This, as Shannon noted, is not a theory of the content of a signal. After all, a gene sequence does not represent the phenotype by describing it. A similar view is Wiener’s notion of information as control, the cybernetic account. It is very hard to think of these kinds of information underpinning the physicists’ view above. Here, the properties are just informational, or as they put it, mathematical. Program-style accounts of genes are in this class.
A third kind of information is the information, or rather the accuracy, of measurements. This is called “Fisher information” after its originator. It is roughly the point on a curve of measurements where the second derivative is zero, or where the error curve is flat. This cannot apply to either physical or biological information, as it is a measure of how well and closely we can measure a physical system. Ironically, it is in my view the only actual physical sense of information, since it requires a physical state to be measured, and a physical system to do the measuring.
So let us get back to the physicists’ claim that the universe is just information. I have argued before [see note] that if an electron has mathematical properties, this is not the same thing as saying that the electron is just a mathematical object. An analogy might make this clear. Suppose I program my computer as an orrery, a simulation of the solar system. If I do this, the computer represents the mass and physical constants as numbers, and processes them according to the mathematical equations of physics. But that solar system in my computer doesn’t have the mass of a real solar system (luckily for me, and everyone else on earth). Instead it has an abstract mass, and the ways the abstract sun and planets interact is, well, abstract. A mathematical description of a system like the solar system is abstract. Apart from instances of that description in physical objects like heads, paper or computers, it exists nowhere in space or time. Consequently, abstract properties do not cause anything in the physical world.
Moreover, the abstractions must leave something out. As the genetic A, C, G, and T leaves out the actual physical properties of adenine, cytosine, guanine, and thymine, which can do things occasionally the symbols cannot, even the most well specified and detailed representation of a thing, at least above the fundamental building blocks of the universe if we ever can say we have them, will leave out properties and capacities we are not interested in representing, and so they will, sometimes but inevitably, deviate in their representation from the actual things. And if we have the fundamental objects (quantum fields?) of the universe, we could not compute the system without first constructing a computer capable of dealing with the whole system, and for a universe, that would have to be a universe-as-computer.
When physicists or philosophers say that we are living in the Matrix, or equivalent statements like the properties of atomic and subatomic objects are merely mathematical, they make a classical mistake, even worse than getting involved in a land war in Asia. They are mistaking the representation of a thing for the thing. The late medieval scholastics like Lombard knew this error and named it long before Saussure: the sign is not the thing signified. The word is not the world. If we are living in the Matrix, what does the Matrix live in? We know of no information processing system that is not, itself, physical.
This is the New Hylomorphism. Information is, as a commentator on Antievolution.org said, seen by Intelligent Design proponents as a kind of caloric or phlogiston. But it isn’t. It causes nothing at all. An abstraction cannot cause a physical process, and to think otherwise is a category error, unfortunately common among theoreticians as well as Intelligent Designists.
The notion of “information” in genetics is an honorary one. It can only mean causal specificity, not anything involving “real” information (on this, see Griffiths and Stotz’ Genetics and Philosophy). And since we have no real reason to adopt hylomorphist views on the real world any more (atomism, or its linear descendants, won the battle), one has to wonder why some scientists and some philosophers think it necessary to reintroduce form as information. Replicators are not informational objects; they are molecules and systems of molecules (Griesemer 2005, Waters 2000). For this reason I much prefer the notion of a “reproducer”, which is a physical entity (or class of entities).
It is time that we stopped making this mistake in science. It is time to give up on hylomorphisms, old or new. In the end, these metaphors (and they are metaphors) only mislead us.
I think that is enough about information from me [too much information].
Crick, Francis H.C. 1958. On Protein Synthesis. Symp. Soc. Exp. Biol. XII, 139-163.
Dawkins, Richard. 1982. “Replicators and vehicles.” In Current problems in sociobiology, 45-64. Cambridge UK: Cambridge University Press.
Dawkins, Richard. 1986. The blind watchmaker. Harlow: Longman Scientific and Technical.
Griesemer, James R. 2005. “The informational gene and the substantial body: on the generalization of evolutionary theory by abstraction.” In Idealization XII: Correcting the Model. Idealization and Abstraction in the Sciences, edited by Martin R. Jones and Nancy Cartwright, 59-115. Amsterdam: Rodopi Publishers.
Griffiths, Paul, and Karola Stotz. 2013. Genetics and philosophy: an introduction. Cambridge: Cambridge University Press.
Tegmark, Max. 2008. “The mathematical universe.” Foundations of Physics 38 (2):101-150.
Waters, Kenneth. 2000. “Molecules Made Biological.” Revue Internationale de Philosophie 54 (4):539-564.
- See Is information essential for life? No., Why not information?, Mechanism, informationism and Ockhamism, More of me in Spanish, and information again, Information and metaphysics, Informational caloric, and Descartes before the horse – does information exist?
I like Gregory Bateson’s definition of information as a difference that makes a difference best for its aptness and versatility. I also remember that Dawkins somewhere calls genes difference-makers, when he struggles with the definition of the gene (that what codes for an amino-acid-chain, an enzyme or that what gets transmitted unchanged etc).
A chunk of DNA only carries information, if variants of that chunk exist and they have different phenotypic effects.
Hence, information is not a thing, but when things are different and have different effects, then the difference in the causes contains information about the effects.
I don’t think that’s hylomorphism, but it’s surely easy to reify information in some abridgments.
Very nice. Substantive, or was that accidental?
Found an error, you have “as bract” instead of ‘abstract’ in one spot.
“Replicators are not informational objects; they are molecules…”
And that which makes a replicator or reproducer different from other collections of molecules is? And that which is transcribed and translated in genetics is? The genetic code is a code. Sure it is impressed on material, but the same form on different materials.
There is a relationship between Shannon and thermodynamic entropy, though I don’t pretend to understand the arguments around Landauer’s principle. Fisher Information is tightly connected to Shannon information.
The point of the “Matrix” analogy is that the map is indistinguishable from a territory when you are inside it – the substrate for the map can be utterly different in nature.
(breach, not breech ;))
I’ve often thought that ‘information’ has no ‘reality’ too. Numbers are just a handy way of describing the world, and words just a limited way of categorizing a universe of ‘stuff’ into smaller parts so that we can deal with the smaller parts.
Not the greatest insight in the world, perhaps, but I dislike biologists talking about how ‘evolution has designed us..’ and then agreeing that ‘design talk’ is just a comfortable way of describing something we have no ready metaphor for. No wonder daft ideas gain credence when we don’t talk about things accurately.
I give these types of debate an ‘Armageddon Test’. If there were no people, or sufficiently complex organisms, would ‘information’ still exist without an observer? I think there is a vital distinction between hearing a tree fall and the pressure variations generated by an ‘unobserved’ tree.
What is missed when DNA is converted to a digital sequence and analyzed or modified and inserted back in the organism or when it is synthesized and put in the organism?
It seems that most of the work being done is at the CAGT abstraction level. Converting from and to actual DNA is a mere implementation detail; a technique. That information could travel through any number of representations and be converted back and would work, right? (see Venters latest book). Maybe that’s true for some definition of work but it seems that industry is proceeding with that assumption.
If something is lost in the A->D and/or D->A conversion, how should that affect our understanding of gene therapy and gene synthesis?
My answer would be, that in addition to anything known, there could be other effects. But what is known?
John, What do you make of Burian and Kampourakis’ gene concept in Kampourakis’ book ” The Philosophy of Biology: a Companion for Educators”:
That’s a pretty fair and exhaustive definition as I read it. I haven’t yet read that chapter. I must do so.
They do have a much longer version – just wondering about their use of information here – they don’t use that word in the longer definition.
This is just specificity, which is fine. The Crick version was originally specificity of base sequence, which he termed “information”, meaning structure. The term “information” can easily be unpacked that way.
Very nice. Another example of the return of the repressed in modern thought is the way that the Singularity version of what it means to be person cheerfully modulates Cartesian dualism in a hylomorphic key. The soul isn’t a separable substance temporarily located in the pineal gland, but it is a separable pattern of electrical impulses that can inhabit a super Internet just as well as a brain.
And of course, as I once noted on this blog, physical differences make a difference…
I read this and thought: Ontological argument
I agree that Fisher’s measure is one of physical magnitudes and structures. I also agree that what Luciano Floridi has called digital ontology is flawed. Interestingly, Floridi rejects the digital ontology inspired by Konrad Zuse and Wheeler’s ‘It from Bit’ ontology, but then goes on develop an informational metaphysics wherein data that reduce to abstracta (relational entities or any non-uniformity in any abstract or concrete medium – including non-extended matrices and noetic spaces (a Kantian influence)) are foundational in the ontology. I think that this is wrong also, and that what belongs at the foundation is the structure of the quantum fields that you have mentioned.
However, I think that there are some problems with regarding information as only a metaphor. More importantly, there are some problems with regarding information as only a measure or the value of a measure. It is not clear to me that regarding information as real in the ontology necessarily delivers hylomorphism, although your argument has salience and I agree with some of it. I think that there is a middle ground.
In physics as well as the special sciences, measures are of some quantity or magnitude, and the quantity is usually a physical one. You bring this out with your comments about Fisher information:
“A third kind of information is the information, or rather the accuracy, of measurements…it is a measure of how well and closely we can measure a physical system. Ironically, it is in my view the only actual physical sense of information, since it requires a physical state to be measured, and a physical system to do the measuring.”
However, then you assert that measures of information in genetics and other physics are measures of representations of abstracta:
“The point here is that the representation abstracts away from the physical properties of the molecules. Measures of the informational entropy are therefore actually measures of the abstract representations, not the things themselves.”
Fair enough, since you also indicate that information is differently defined and conceived of buy different scientists:
“When scientists talk about the information content or the informational entropy of something, they can mean several things.”
This is the motivation for the kind of pluralism about information endorsed by Floridi. He ascribes it also to Shannon (but I think that this is a mistake). It may not be that this pluralism means that there is no central conception of physical information to be had, but instead that abstract measures are not in fact the information.
Another thing to consider is that information in Shannon’s theory is considered to be the objective reduction in frequentist uncertainty. However, what if probabilities themselves are in fact physical?
I think that Shannon’s measure is one of physical information, not a measure of physical magnitudes that itself is information. Shannon sometimes refers to the measure value as the information, but he also both refers and alludes to the information within the system being measured. This is brought out by his adoption of Von Neumann’s terminology and concept of entropy, and of Boltzmann’s conception of entropy. Boltzmann’s conception is probabilistic, but the measure is one of disorder in a physical system. This looks a lot like a reference to a form of sorts – I prefer to refer to structure. You say that when physicists equate structure with information it is harmless. However, I do not think that it is inconsequential.
Physicist B. Roy Frieden has produced a unification of Shannon’s measure and that of Fisher (extreme physical information). It is known to include some mathematical bugs, but is otherwise a successful attempt.The emphasis is on recognising the structure and properties of a physical system as data that is information. That Shannon’s measure can be hybridised with what you refer to as the only physical measure of information i interesting and perhaps not inconsequential.
Scientist and Philosopher of science Sahotra Sarkar has called for a unification of conformational structure and dynamics of molecules and molecular interactions in biosynthetic processes. His argument is that the properties of proteins and phylogenetic traits are governed not only by sequence structure, but by such things are conformation, stoichiometric properties, and the dynamics and degrees of freedom associated with them. Moreover, it very much seems to be the case that sequence structure can be taken to reduce to these conformational properties to a significant extent. Arrangement of what reduces to fields seems to matter. Form of a kind is perhaps not out of the game, but not as some additional separate feature.
Then there is the conception of algorithmic information put forward by Andre Kolmogorov. His measure of information is the length of the shortest program that will produce a given data object. This seems to be about sequence at first. However, Kolmogorov was a Brouwerian intuitionist and materialist physicalist. He did not think that infinity was real. There are problems with Brouwer’s intuitionism, of course. But Kolmogorov’s objects were always physical strings/sequences. The measure was of something physical. The structure of the data object matters. One can deliver a Kolmogorov measure of a string that does not exist, but according to Kolmogorov such is not ‘real information’.
Then there is the question of the structure of quantum fields themselves: it very much seems to be (arguably) the reductive basis of physical properties.
I am giving too much information even more than you at this point, so I will leave it there. However, I do not think that structure is meaningless in terms of a basis of information, and if it is physical structure then there seems to be a case for the information to be physical.
Bruce, I am in furious agreement with nearly all that you say here. Although I am unsure if AIT is a generalisation of Shannon information, Fisher information or whatever, or vice versa, Chaitin’s version of Kolmogorov information seems to me the best version. But I would cavil at a few comments you made.
1. I was referring to the symbolisations of DNA (GTAC). They are a kind of measurement, I guess, in that they identify the molecular structure to some degree of precision. But attempts like Gatlin, Schneider, and others measure the knowledge of the genes, not the reality of the molecules, and so they are epistemic measures rather than ontological.
2. Structure can be simply called “structure”, without the need to call it information with all the connotations I object to here. If you want a term for genetic information, I referred to Griffiths and Stotz’s view of structural information for genes. This can, I think, be generalised to any measured state. Entropy is a metric, yes, but it is a metric of physical states.
If we could cleanse “information” of the metaphorical implicature it now has then I would perhaps have no problem, because scientists would not tend to hypostasise the metric.
I left a lot out of my discussion, because after all this blog is not for specialists and I was making a point. In the end, though, I think information is either causal structure (so call it that) or it is a mathematical abstraction. Either way, we have to distinguish epistemology from ontology.
(*Apart from a couple of typos, my correction for rushed production (small dependent really wanted to go home) is that Sarkar is in fact after a unification of structure-conformation, dynamics, and sequence for a conception of information for the molecular biosciences (I work with this for chapter 5 of my thesis in which I present an analysis of a unification and a calculus for expressing the information in an entire causal pathway in biosynthetic processes).)
Sorry if I targeted the wrong audience, but everyone seemed to be keeping up nicely to me.
Paul Griffiths is my supervisor (and a better supervisor for this material there isn’t).The project of my thesis has been to use scientific metaphysics (Ross and Ladyman but without the broad reduction to statistical constructs which seems a lot like another attempt to make the ontology mathematical in a real sense and not just an epistemic one) to identify a central conception of information. I mention it in my paper ‘Information is intrinsically Semantic but has No Alethic Value’ (in the next edition of Synthese), so here it is:
P1. Information is what I call CICS (causally induced configuration of structure).
Pretty basic. In the thesis I end up having to develop a version of physicalist reductionist non-eliminative ontic structural realism (NOSR). I defer to QFT in order to follow the scientific metaphysics rules, but I do not end up with statistics a the bottom like Ladyman and Ross do. Ladyman and Ross’.The structure at P1 has to be physical. I have a lot of arguments from scientific metaphysics (think along the lines of Devitt’s ‘Putting Metaphysics First’ as well as the physicalist NOSR ideas of Ladyman, Ross, and French)
Re. your 1 above. This is a very astute comment and I see the difficulty.
I believe that Sarkar’s call for a unified conception of information in the molecular biosciences is a call for a more complete description of the semantic information content of the gene. (In true annoying philosophical style I now have to question your use of ‘epistemic’ as an apparently metaphorical reference to the information content of a physical structure, but I will assume that it is about making the material palatable for those not nuts enough to obsess about it like I must. If you mean that the measure is constructed relative to the epistemic needs of the scientist, then I think that it is still a measure of the real – in the abstractive sense that you mention that can yield differently constructed measures of information – or perhaps I should say ‘information measures’ so as not to be pre-emptory.)
As you said/inferred, genetic (DNA/RNAs) sequence reduces in large part to selected parts of the entire physical structure: it constitutes an abstractive (in the engineering and scientific sense of abstracting out details to emphasise important features for solving a problem or modelling a system). Recently (80s-90s) have been discovered of the existence and importance of chaperone molecules to ensure that protein folding executes correctly. The details are all about conformation and dynamical degrees of freedom: that is where some of the additional information is. Then there is the newish discovery (relatively speaking) that there extremely large numbers of possible dynamic folding pathways (mapped in most cases using energy landscapes) to get to a native protein configuration because of the enormous conformational and stoichimetric degrees of freedom involved in folding just a few arms of a macromolecule.
Then there is the information of the molecular interactions to be considered. That of the chaperone molecule with the protein, of the enzyme with the DNA in splicing etc. If information has something to do with causal pathways then these are certainly part of those. Can things go wrong with the native protein if something goes wrong with them (noise)? Yes. This suggests that there is information in the conformation and dynamics as well as the structure (I am sure that I am effectively teaching you how to suck eggs – but I am ordering my own thoughts and hopefully it is helping anyone brave enough to track along).
If we use ‘epistemic measure’ to refer to a measure of the information in a gene, then a measure of conformational (Sarkar), dynamical (Sarkar), interactive (Me), causal-pathway (all of us – I think), and sequence structure(s) would be such a measure: but the ‘epistemic’ information content is part of that – but also constitutes capturing what I refer to as an encoding of selected physical information. We are just selecting more of the physical information in the measure (my calculus does this).
Re. your 2. above. I wholeheartedly agree. I think that structure is just structure. Here’s an interesting liminal penumbra type question, however: If a structure is configured a certain way, then is it the same structure? Or is configuration just structure? If I talk about structure, is talk of the configuration of a structure just tautological. I am going with no. Thus refer to P1 for the definition of information. There are language games to be had, but I can use any terms. Take some quantum fields. Say that they are the structures. Take the same fields and shove them around. You have reconfigured them. There does seem to be a distinction (between the field structure and its configuration) that is real and matters. The information is the causally induced configuration of a structure. It can represent something of the structure (including dynamical structure of the interaction and the causal pathway)
Yeucch. That’s a lot of metaphysics (scientific hopefully) for cleaning out a metaphysical implicature. I am trying to make sure that my QFT is good enough to ensure that I am not flying a big fluffy armchair. Fingers crossed.
I am enjoying the hell out of your blog.
“information is either causal structure…or it is a mathematical abstraction”. In the genetic context, the actual structure of DNA is not essential to its function, in that we could imagine (in fact we actually have, given the redundancy) a different nucleotide code giving the same amino acid at the corresponding position in the encoded protein. In the evolutionary sense the engine of change is coding error (which again takes particular forms that depend upon the specifics of the informational storage medium), although the success of the resulting mutated protein (or transcriptional regulatory motif) as a tool for selfish gene survival is out in the brute facts of the outside world. The further complication is whether we need to take a population level view to really see how the “Genetic Algorithm” is solving the problems of life.
Comments are closed.