Common sense and the well-designed child

30 Jan 2012AI

I recently read an interesting article by John McCarthy that set me thinking. The paper is called The Well Designed ChildJ. McCarthy (2008). The well-designed child. Artificial Intelligence 172, 2003-2014. Free online version here. and talks a bit about about the old nature versus nurture debate: to what extent is a newborn child a blank slate, and to what extent are human intellectual capabilities intrinsic and instinctual?

To me, the whole question of nature versus nurture has always seemed odd. It’s clear that both aspects are important. We are all limited to one extent or another in what we can do purely by the physical parameters of our existence: as much as I would like to teleport myself to Europa to frolic in the sunless sea beneath the ice, I can’t, and never will be able to. Equally though, the potential that exists in all of us at birth can be squandered, and without education and opportunity, none of us can make of ourselves all that we might. So, we need a bit of both.

Perhaps the source of the debate lies in observation of other creatures: beavers don’t go to Beaver Engineering School to learn how to build dams, and spiders don’t attend art college to learn how to make beautiful webs. These abilities are intrinsic, evolved, part of what Richard Dawkins called the “extended phenotype” of these creatures, behavioural manifestations of developmental and neural structures that are passed down from generation to generation. Humans don’t appear much like that. We do have to go to school to learn how to build dams or weave webs or to do any of the thousands of things that we do. The question then is: what intrinsic capabilities are we born with? What is on our slate, since it seems as though it ought not to be completely blank?

Human babies don’t seem all that capable, compared to the young of other species. That’s something of an illusion though, since humans give birth to developmentally very premature infants, mostly as a result of evolutionary compromises to allow us to have large brains combined with an upright posture. I don’t know exactly what human developmental stage would be a good one to use for comparison with other species, to try to decide what intrinsic capabilities we might have. However, there are three fairly obvious capabilities that do spring to mind. These are the capability to learn, the apparently intrinsic facility for language acquisition, and a basic grasp of what, for want of a better term, I’ll call “folk physics”.

The first two are aspects of human development that are quite well known and explored, although none of the questions are yet really settled, as far as I can tell. I’d like to concentrate on the third, this idea of “folk physics”. What do I mean by that? I mean the kind of common sense knowledge of the world of physical objects that we all have–two objects cannot occupy the same space; if one object passes behind another, it does not cease to exist; objects further away appear smaller; gravity gives a definite up-down orientation to the world. You can come up with lots more examples. These things seem almost too obvious to state, but they are still knowledge of a kind that either has to be intrinsic or acquired, and they constitute a large base of shared facts that we all need and use for living in the physical world.

What McCarthy’s paper really set me to thinking about though, was what might minds be like that had different sets of common sense to our own? Uncommon sense, if you like. A couple of examples will give an idea of what I’m talking about. First, think about an entity evolved to live in vacuum, in a micro-gravity environment. Human-default common sense physics seems to be Aristotelian more than Newtonian, but that’s hard to imagine for a creature that never feels air resistance and doesn’t have a large local gravity field to give it a sense of up and down. Just for an example, how would such a creature refer to directions? No up and down may lead to no left and right either, although near and far seem as though they should still be usable. How about referring to directions towards items, rather than abstracting into categories like up and down? That might work, but it would have a fundamental impact on the way that such entities might use language to refer to each other and to objects and places. It’s interesting to speculate what sort of idiomatic usage might arise from this different perspective on the world.

Second, another example, more remote from human experience in many ways. What sort of common sense, and what sort of mind, might an entity inhabiting a computer network (the internet, say) develop? Here, there is no “physics”, since there is no physical. There are no physical directions at all, although the environment has a rich topological and metric structure. The more I think about it, the more I find myself at a loss to imagine what the existence of such an entity might be like. It led me to think that true artificial intelligences, if we ever succeed in creating them, may be even stranger than they are usually portrayed in fictionCharlie Stross’s intelligent lobster spam filters might be an exception here on the insufficient strangeness front....