On cuteness
Our old friend Debra visited us this weekend and, as we went along, took a bunch of pictures which make up the bulk of the new slug of photos tagged “day eighteen”. Thanks to her for letting me put them up.
Beginning his third week of life, Dante has started becoming responsive to his environment. He is beginning to track people with his eyes a little bit, lifting his head toward things he wants for a few seconds at a time, and reliably getting his hands to his mouth whenever he feels the need. It all points to a time when he will have the ability to feed and move himself, rather than relying on baby appeal to move adult muscles in his service… but for now, his first endogenously motivated activities just serve to highlight how cute he is, and provoke nearby grown-ups to more strenuous activity on his behalf.
I understand that everything that my son does at this point in his life is part of an instinct that babies have evolved to efficiently mine their parents for resources, and that his “cuteness” along with my deep need to respond comes straight from the stupidest mammal part of my brain, being a manifestation of parents’ instinct to be exploited to a greater or lesser extent by their offspring. I can’t help but marvel at how deep our biases, in the learning theory sense, run — that is, at how much information about the nature of the world is simply encoded into our bodies before we even have to make decisions, or to learn new behaviors or policies. Well underneath the level at which I can reason about my reasons, I simply know that a crying baby requires attention, and it’s very stressful to be around one. Conversely, it’s very calming to be around a happy baby. Consequently, babies and adults form a homeostatic system that stabilizes on happy babies and calm adults, which presumably has good evolutionary consequences.
This sort of instinctual model of the universe, present and rich even before learning can take place, is a theme that’s run through my thinking about machine learning in the decade I’ve been working on it, making me think that the conventional wisdom about high-bias ML systems like expert systems (said CW being that those systems are old and busted) is completely wrong. Successful living and learning systems in the real world carry an immeasurably great bias in every aspect of their design: most feet are flat because the ground mostly is flat too, and they’re mounted on the bottom of a creature because a part designed to touch the ground needs to be below all the other parts… and that’s just feet. How much more bias, subtle or profound, must inform the architecture of a learning organ? And why should we expect “purer” learning systems, drained as completely as possible of bias, to succeed in making predictions about a world so complex that real organisms, embodying huge amounts of background information, can barely cling to life?
Any theory of human nature based on a blank slate reflects an astonishing naïveté and lack of insight. Careful consideration of what is packed up in our notion of “cute” reveals its irreparable lacunae. John Locke is considered the concept’s most important proponent among Anglophone philosophers; I think he must never had any children (edit: sure enough), which, married with a lack of introspection, might have allowed him to convince himself that there is no immutable human nature, that humans have no instincts. But how would such impoverished creatures survive and learn in such a phenomena-rich world?
Update: Coincidentally, Jen Kirsch of Juror2, who drew es&f’s banner, recently put up a new piece with a very expressive infant in it. She gets a lot of mileage out of her sticks, which are evocative and imaginative and, yeah, cute.
Leave a comment