Richard Lewontin’s 1992 book The Doctrine of DNA is one that I reread every couple of years, discovering new things. The tone of weary omniscient scepticism grated on me for some years, but no longer does. After all, he was right to point out the limitations of the human genome project, and his rightness becomes more apparent every year.
There are two reasons for these limitations. The first is that the genome in itself does not predict very much about the organism. We need all sorts of extra-sequential information for that, as well as the rules by which that information is interpreted by the cell. You may, if you wish, see these rules as themselves contained in the genome. That doesn’t affect the argument. If you haven’t unpacked the rules, you can’t make sense of the information, and you certainly can’t discover what the rules are by studying the genome itself.
The second point is more general. Somewhere in Lewontin’s Marxism, there is a point about meaning, or about holism, which Dan Dennett and Sydney Brenner, seem to be approaching from different directions. Here’s Lewontin:
“It is not that the ‘whole is more than the sum of its parts’. It is that the properties of the parts cannot be understood except in their context in the whole.
Parts do not have individual properties in some isolated sense, but only in the context in which they are found. ‘The theory of human nature that searches for that nature in the products of genes in individuals and the limitations of individuals caused by those genes, or in the properties of an external world that are fixed and that cannot be altered except in a destructive way, misses the whole point.”
Compare this to Sydney Brenner, from my book, arguing against ’emergence’:
“The real thing I’m against now is this idea of emergent phenomena. Seems to me very mystical: they say that if you mix things that get to a certain level of complexity, then you get emergent properties. That’s to try to say the whole is more than the parts but, in fact, the whole is the sum of the parts and their interactions, that’s all; and our job is to find the interactions. Then we could compute the behaviour of the system.”
The crucial point here is the mapping between “properties” and “interactions”. Lewontin’s point, it seems to me, could be rephrased as saying that the properties of something are the sum of its interactions with the world. A property is a prediction of behaviour under certain circumstances. This kind of functional definition is pure Dennett. I did have a page reference and quote to back that up, but the book got tidied away … Get Lewontin from the library instead, and read that.
“Compute the behaviour of the system”. Right.
Brenner should have a word with one of the (by repute) two people in Intel who can hold in their heads a complete model of the latest processors, and ask them whether it helps much. It’s been the case for a while that the complexity of the things exceeds any ability to comprehensively map their behaviour. Intel has complete control of the design, pretty well down to the atom, a fabulous array of design tools and more brains in vats than you could eat with a very big spoon, but sho ’nuff the chips behave in ways that nobody can predict. When you fix the interesting problems that mean one sequence of instructions slows the beast down for a thousand cycles, similar behaviour is liable to pop up in a different sequence. It defies the abilities of Intel to codify or model: they can simulate and hope.
If our computers ran a billion times faster, then we could run a billion times more test patterns in simulation, and we’d find more of these, but it’s still not clear we could understand what was going on. Does this count as ‘computing behaviour’, or does it just mean we can make a fine-grained model good enough to exhibit the same emergent behaviour as we see in real life.
I think for me the most graphic demonstration that some things won’t be boiled down is Conway’s Game of Life (nerd cliche, I know). You can find out whether your starting screen will end up with a stream of gliders lolloping along in an infinite loop only by running the thing. You’re not guaranteed to find out any other way, and in fact you’re not guaranteed to find out in any case.
One can fully explain the rules of the GoL to someone in a minute. I haven’t tried the experiment, but I’d expect I could teach enough BASIC to a shiny novice in an hour so that they could write their own, and be confident that they were completely aware of each part of the design and implementation process. Inasmuch as anyone can ever possibly aspire to the condition, the Life afficionado is omniscient of the properties/behaviour of his universe and its contents.
It doesn’t help. You can predict a subset of futures, and I dare say that as with maths, physics, biology or whatever you can expand that subset as you get better, but the system will always be able to surprise you in interesting ways. At some point, we hand over ‘understanding’ to observing what the machines do and wondering.
Emergent properties might ‘just’ be phenomena born of our limitations rather than something inherent in the systems we observe, but I don’t think I can tell the difference and I do think that it is just as valid to call them emergent properties as anything else.
Meanwhile, did you see this? Does me crust in, it does.