Evolution proves that algorithms can solve any problem
 Some intelligent chatbots, yesterday.
|
You will see much scepticism in these pages that the claims habitually advanced in the cause of artificial intelligence are all they crack up to be. This is not just predicated on the quality of LinkedIn’s AI-generated question prompts — though they are certainly articulate evidence for the defence. Happy work-iversary!
We say that algorithmic processes — even clever ones — are simply incapable of responding to unexpected events in complex systems. Crisis management requires imagination, creativity, and the ability to quickly construct a narrative — qualities not possessed by any artificial intelligence known to the world today. Expecting preconfigured algorithms to solve novel problems is like expecting Newtonian mechanics to explain the very events which, by their existence, falsify Newtonian mechanics. One needs to construct an entirely new model.
Here is the prosecution’s clincher:
Aha, but the very imagination, creativity and narrative construction skills you point to, are human, and the imaginative sapience of homo sapiens is itself the product of an algorithm: the one encoded into evolution by natural selection.
Here the prosecution appeals to Darwin’s Dangerous Idea: it was evolution by natural selection, after all, and only evolution that operated relentlessly for 370 million years, from when the first legged fish crawled out of the primordial ooze and onto the shores of a new, terrestrial world. That single algorithm transformed those little flippy-finned mudsuckers into the highest type of sentient being yet known in this neighbourhood of the Galaxy: the ISDA ninja. And, to the best of our current thinking, all an ISDA ninja is doing is using her brain, and that is purely algorithmic, we see that human natural intelligence is an algorithmic process, created out of an algorithmic process.
So how can you say algorithms can’t be intelligent?
Herewith, the case for the defence:
Not “can’t”. Aren’t.
It is not that algorithms cannot in principle generate general intelligence — though the “Darwin’s dangerous idea” arguments are a bit hand-wavy — but that the particular ones you find in artificial intelligence as it is currently developing — and, no, it isn’t “evolving”, which is kind of the point — won’t. Not just any algorithm is capable of self-awareness — a good thing, or you would spend more time in meaningful communication with your carrot cake than is necessarily healthy. They are a special kind of algorithm. Some folks have some ideas about what their special qualities may be (Douglas Hofstadter thinks the key may be recursivity).[1]
Accepting for a moment that the evolution by natural selection algorithm can generate intelligence, consider how staggeringly slow, destructive and wasteful it is.
The century of the self
Hofstadter is at his best when when he addresses the reflexivity of human consciousness — the magic that emerges courtesy of the strange loop whereby the human perceives itself inside the universe it constructs, and where that working narrative must allow for, in order to explain, one’s own causal impact on the universe. This sets off an infinite loop which creates magical artifacts all by itself.
In Roland Ennos’ recent book The Wood Age he gives a good example:
Early apes, manoeuvering through the treetops, developed a concept of self, because they realised their bodies changed the world around them by bending the branches they stood on.
The only way you can explain the movement of those branches is by reference to your own presence.
It is hard to see a dematerialized computer, operating in a virtual space, having to solve that same problem, other than at the quantum level (it need hardly be said that quantum elements are not the same as machine consciousness: that would be a reductionism too far).
Machines are more like Arthur C. Clarke’s sentinels, watching dissociatively. They purport to describe the world as it is without affecting it: they monitor, measure, observe, process and give back but not to themselves, and not for themselves. To do that would be to colour their observations about human interaction, which would be to defeat their commercial purpose. Machines were developed to execute operations, quickly, without error. There is no room for error, judgment, variation or interpretation. Machines are designed not to narratise. They process symbols in mechanical operations. They do not read.
Kazuo Ishiguro’s Klara and the Sun, which I happened to read straight after The Wood Age, makes the same point through the android’s unusual segmented spatial perception. It prompts the question: to what extent is consciousness a function of our own peculiar evolved perceptive apparatus? It seems to me it must be. When we design artificial intelligence with a blank slate, and especially if we come at it from a Behaviourist machine learning angle, which cannot have been the route to consciousness of human evolution, deprecating as it does the notion of an inner consciousness altogether, we would make different design choices, arrive at functional intelligence different way and these might lead to profoundly shifted articulations of ”consciousness”, if indeed any kind of consciousness emerges at all.
See also
- Daniel Dennett’s impressive, but over-blown, Darwin’s Dangerous Idea
- Evolution
- Why is reg tech so disappointing?