This informal talk was given on 9 July 2014 to a small audience. Some sections include comments from the audience members. Topics include digital archival, biotechnology, and information science.
Gnomic means “full of instructive sayings”, but it comes from the Greek gnome, meaning “thought, opinion, maxim, or intelligence”. So “autognomic” means something that is self-explanatory.
I was thinking about how to preserve information. Either you make it so awesome that people copy and translate it, or you carve it into some tough metal, though then it won’t translate itself. So I was thinking about how you get around the problem of translation. Perhaps you can make it autognomic in some way?
I was thinking that you could probably do so even relative to some alien culture. You’d start by encoding the human genome somehow—probably just draw the actual molecules out, and then put a “key” by it. But then you have the problem of how you encode our thought-languages as well, and I was thinking, doing that is very difficult because it’s relative to our entire environment.
So then I was thinking about information content. What, for example, is the information content (size) of the human genome? I think it’s like 1GB or so, if you measure the DNA, but what about the language that the genome exists in? You might think of it as instructions for biological machines, a self-assembling 3d printer blueprint; but then, again, the environment is part of the language.
So I was wondering how you could measure the information content (size) of an entire language. You’d have to measure the information content of the environment too. I was also thinking about what a language is: it’s a collection of patterns that enables a subset pattern, one that’s never been seen before, to be interpreted relative to the collection of patterns. Perhaps, then, language is a meta-pattern.
So for example, DNA is a pattern. But when it’s “interpreted” by the environment, it’s done so relative to all the things in the environment. So for example there’s a single pair in the human genome that decide whether we’re lactose intolerant or not. That single “bit” (or two bits?) encodes information about lactose intolerance. Imagine how much you could write about lactose intolerance, and yet that’s encoded in a bit or two!
So it’s the relativity of the pattern : metapattern that really decides the information content, not the size of the pattern, i.e. the utterance in the language (which is human language biased; when you think about DNA, a pair for lactose intolerance is not really an utterance. I’m using pattern as the generic name).
I also thought about kinds of computing. So for example, the first computers are presumably conscious animals: Computer I. Then some of the conscious animals made another type of computer: Computer II. Very different kinds of computing, as far as we can tell, though we don’t yet know why. So I was thinking maybe there are entirely other kinds still!
Audience: You sort of need a compiler that compiles itself, right?
Author: Yes, something like that.
Audience: Well, then you need a bunch of primitives, and then a whole language built based on those primitives. Sort of.
Author: Primitives are interesting because human languages don’t work that way, and nor does DNA. But our Computer II programs do work that way. But it’s pretty weird, when you think about it. Imagine a dictionary where some words could literally not be defined in terms of others. That would be weird, but that’s what it’s like in Computer II.
Audience: I’m not sure about the “conscious” part for Computer I. I mean, well, wait, you mean the animals that made Computer I? Or that Computer I means conscious animal?
Author: Computer I means nature’s own computers. We don’t know how brains work yet, so we don’t know what level of animal computes and in which way. I was imagining that example from... not sure who it was. Penrose? About the planet made of broken TV parts. Maybe it was Hawking. And imagining that the tidal forces of the planet somehow assemble a working TV. That actually happened, in a way, with evolution creating Computer I. But it didn’t happen through random collisions. It happened through gradually more complex emergence of languages from the environment.
And it’s still happening. Our languages are not separate from that process. They’re a new aspect to the exact same process, whether natural language or computer language or something new. An interesting thing about DNA is that it doesn’t seem to compute, and yet it’s very informational. How does that relate? Maybe it does compute.
People are thinking about how to do computing with DNA. That’s fair enough. But is DNA currently computing, somehow?
Some researchers are calling the biological transistor a “transcriptor”:
“The creation of the transcriptor allows engineers to compute inside living cells to record, for instance, when cells have been exposed to certain external stimuli or environmental factors, or even to turn on and off cell reproduction as needed.”
It’s not clear how close the analogy is. It’s funny, because Zuse’s first computer was entirely mechanical; then he quickly switched to electricity and we’ve been using electricity ever since. It seems very natural! But now we’re wondering if biological computers might not be better after all.
You’ll notice that I didn’t really solve the autognomic information problem. But I did have one thought about it. It’s that when we come to understand the architecture of the brain better, we may be able to create a Rosetta stone that encodes our languages relative to patterns in the brain. In some sense we’ll be speaking the ur-language, because it’ll be the cognitive language—though I expect that this will once again challenge our definition of a language (just like, say, DNA does).
Audience: But again this relies on a set of primitives that need no explanation whatsoever, right?
Author: “Needs explanation” is a very misleading concept, because language acquisition doesn’t work by looking up words in the dictionary, yet we have a misleading idea or folk belief that it does. It’s a kind of cargo cult, I guess.
Actual language acquisition works by immersion in the environment in lots of different ways, and is very heterogenous. (We still don’t know the full ins and outs, partly because we don’t know what’s really going on in the mind).
Audience: I suppose that if you’re constructing a higher level concept based on primitives, the notion of primitive implies it cannot be constructed in terms of any lower level constructs. That’s what I meant by “needs explanation”.
Author: Computer II systems work in this way, with primitives and then compounds made from the primitives. But Computer I systems might not, and DNA seems to be very different—is it all primitives? Or what? I think what’s happening is that we’re mapping natural language onto Computer II systems, and we’re getting confused by our analogy, when it stops holding.
We use functions like verbs, we create data structures and think of them as nouns. Then we think of computational primitives as magic verbs, atomic verbs that can’t be broken down. But actually, I’ve been through this and smashed them down. It’s like the atom itself—it’s a misnomer! You can break apart the atom!
If you take some 6502 assembly instruction, say. That’s a primitive, but then you can look at the 6502 chip to see how it’s implemented. Lots of transistors, working together. So then I thought, hey, imagine if your only primitive was the transistor. But you can have other kinds of primitives. You can have SBNZ, for example, or any of the other OISC primitives.
Audience: Boolean logic for instance being the ultimate “primitive” then?
Author: No. Because then you think, well wait. If I can make a 6502 chip out of transistors, then I can also simulate it in SBNZ right? So then you realise that, actually, these “primitives” are not important. They’re a substrate for the meaning. It’s like saying that the letters “a-z” are the primitives for language. Suddenly, you realise that it’s phoney.
Audience: That’s just the Church-Turing thesis, right? What are you getting at?
Author: We’re using the Church-Turing thesis to investigate the nature of information. More specifically, of languages—where language is broadly conceived to cover, say, human natural language, programming languages, and DNA. So when you look at any of these individually, you can get misled by either the characteristics of that specific language family, or by misapplication of concepts from another family—and that’s what was happening here. We were getting misled thinking about the nature of the programming language family, thinking that “primitives” are an important concept somehow. But one way around that is thinking that the word “hello” must be composed of these primitives, h, e, l, and o. But of course it’s not. They’re not really anything to do with it. (And also, as you note, using the Church-Turing thesis to reason about it.)
Audience: Here referring to the biocomputer?
Author: Well, biocomputer could mean lots of things. This specific context is a specific subset of written human languages. This started in wondering how you measure the information content (size) of things, when actually what you’re doing is measuring them relative to an interpretational framework. But then even “interpretational framework” doesn’t quite apply to all languages. A machine interprets a program in one way; humans interpret language in another; the world interprets DNA in another.
Audience: The way these “computer in a cell” reports always strike me is that we’re forcing our primitive paradigms (Turing-machine-equivalents) down the neck of the complex dynamical system that is a cell.
Author: Yes!
Audience: We should be doing cybernetics.
Author: Yes, but it might be very hard. It might take us another thousand years to work out the basics, for all we know. Or maybe it’s easy. We can’t know until it happens. I expect that there are surprises still in store for us. David Deutsch wrote a great article about this. He argues that the reason we haven’t figured out AGI yet is that we don’t understand creativity. I think it’s more complex than that—there’s probably a whole slew of things we’re yet to understand.
Audience: Most of the time we can’t even comprehend how little we understand, now trite and sophomoric our conceptions are.
Author: I think what’s beautiful about our era is that we can see that the future of ideas must be exciting, because there’s so much we know tentatively.
Another thought I had was figuring out the average human genome, doing as wide a survey as possible, and then making that human. See what he or she is like. They won’t be average, because they’ll get so much attention! But then that spurred another thought: we should be able to use Computer II systems to model evolution much more quickly than nature does it. But perhaps I’m wrong about that—for example, being able to model the lactose intolerance bit. Imagine trying to do that. You’d need a computer basically encoding the whole environment. But perhaps we can do it to some level of statistical confidence. Or, at least find out some interesting things from such an investigation.
I’ve often wanted some kind of DNA fingerprinting for identification of species. Lke, derive the taxonomies from DNA itself. Don’t figure it out from other methods. But I was wondering how you’d compactly represent the various overlaps from the DNA. That’s quite a UX challenge! I expect that you’d discover that taxonomies aren’t as tree-like as we tend to think.
Audience: Well, we’re getting a grip on something seemingly-fundamental called “information”. We just can’t even define its most basic properties consistently between domains.
Author: Yes, that’s what I was trying to do today of course, partly influenced by the Baez and Stay paper. Baez and Stay’s paper extends my discussion of language into physics, topology, and logic.
Audience: I wish someone would do a day conference on that paper, or just the rough idea. But autognomy is an interesting idea. Like, all the effort put into the Pioneer plates. Is it really in vain? For communication, that is the coordination of internal representaiton in the basis of exchanged information, you need some shared context. The efficiency of discourse is a function of the wealth of shared reference. But we postulate universals, and we retrofit them into a kind of shared context. So we guess aliens will eventually be curious about and figure out hydrogen orbitals and pulsar frequencies, but it’s still a bit of a stretch.
Author: Yes, I was thinking about autognomy for anything human being founded on top of the human genome. But once you have the human genome, it’s still difficult to figure out what the human does, because there’s also some entanglement with the environment here on earth.
Audience: I think what we call the genome is one of these laughable toy-brick concepts. But in the most abstract sense imaginable, there is some shared heritage of genetic information, and that co-conspires with the environment in some very not-yet-fathomed ways to make the wondrous shared complexity of humans—or at least shared propensity towards the elaboration of remarkably similar complexities.
Author: Yes, one example that I like is the base pair that encodes lactose (in)tolerance. It probably does other things too, of course. Genetic polysemy!
Audience: With genes, we’re thinking of the states and not the correlations. If you think about quantum computers, the number of correlations grows so quickly with the number of things that have state. After what might be called a small handful, you cannot possibly simulate them interplay of correlations with all the classical computational resources of the universe. But we still talk about genes-for-X. People trying to write quantum algorithms do not talk about the qubit for b-search, or the qubit for fast-fourier-transforms.
Author: Yes. You’ll notice that this was the basis of my idea about measuring the information content (size). You can’t measure the states and call that the size of the information content. You have to measure the correlations! I was thinking “how big is the human genome?”, and that’s when I realised. But it’s still true that the genome itself has an information content size. It’s just that it’s not that interesting a measurement.
Somebody very recently came up with a definition of computing:
“This is how we can escape from falling into the trap of ‘everything is information’ or ‘the universe is a computer’: a system may potentially be a computer, but without an encode and a decode step it is just a physical system.”
“Consider the case of Egyptian hieroglyphics: after the loss of the language, and before the Rosetta stone was deciphered, did a hieroglyphic inscription perform a communication? It was potentially a communication, just as a physical system can potentially be a computer. However, until a decode was possible, it did not in actuality perform communication (no-one could read it). Once the language was understood, the decoding relation was in place, and communication could occur.”
The use of ideas of Information Theory for studying ‘language’ and intelligence in ants is very interesting. It appears to use information science to discover a new language family (in the way that DNA, computer programs, and natural languages are each a language family).
‘We analyzed the question of whether ants can use simple regularities of a “word” to compress it. It is known that Kolmogorov complexity is not algorithmically computable. Therefore, strictly speaking, we can only check whether ants have a “notion” of simple and complex sequences. In our binary tree maze, in human perception, different routes have different complexities.’
July 2014