Randomness and Early Learning

I’ve recently been spending a lot of time with a baby, and one thing I’ve noticed is that much of their behavior is effectively random. Their motions are often erratically motivated, and even when purposeful, are jerky to the point of distraction (e.g. they may lose their balance while sitting and lean forward to discover their feet and change their whole plan). Their utterances seem unprompted, and do not correspond in timing or content to the utterances of those around them or to stimuli except in rare cases.

However, these random actions and utterances provoke very non-random responses in the adults that witness them. This strikes me as expected. Babies are exploring a world they know nothing about. Random input seems like a reasonable heuristic to use to explore a completely unknown thing. Imagine yourself trying to understand a box. You don’t know how to open it, you don’t know what’s inside, you don’t know who made it or how or why. A reasonable approach would be to pick up the box turn it over in any which way, rotate it, look at it from different angles under different lights, weigh it in one hand and then the other, shake it – to exert random stimuli upon it and then to observe how it behaves.

This is what babies are doing to the world and to everything they find in it. They start from a place of much greater ignorance. In the box example, we understand what a box is, what it means to open a box. We have certain expectations and learned understandings about the box even before we examine it. But we exert random stimuli on those parts with which we are unfamiliar. Similarly a baby exerts random stimuli on everything.

Where does this randomness come from? It must be generated in the brain, but it need not be generated by a dedicated brain system. We might compare it to a computer that uses the motion of the mouse to generate random numbers. In the same way, a brain could generate random stimuli by sampling its other systems.

It also seems as though the ability to produce random stimuli could be selected for. In exploring a box or any other puzzle, the ability to creatively explore the puzzle, and the degree to which our exploration is constrained or not by our presuppositions, and the manner in which it is so constrained, can affect how successful we will be at solving the puzzle or understanding the Box. A system that produces good randomness, and does so at the right time and for the right puzzles, could mean an individual solving a problem that leads to e.g. more effective hunting, etc.

This seems particularly true for very young children. The right mix of random actions could mean faster acquisition of the building blocks of understanding, which could have significant long-term cognitive consequences.

Kind of reminds you of the Philosophy forum, doesn’t it. :sunglasses:

Formative years for children, 0-3 yr. Coordination (in mind & body) is huge in accomplishing random actions, which spills into memory for mind/body. The sharpest memories would forego random actions first.

Merely the act of shining a light into a dark room, sprays the room with random colors and intensities. It is from pure randomness, obscurity retreats (aka statistical variation).

and this is what I meant by your sterling additions to this website. This is solid work.

This is gonna sound legit nuts

You could flip this idea on its head and think… “how can I associate this seemingly random behavior with something meaningful?” While adults tend to control their behavior… specific gestures (e.g. “waving”, “peace sign”, “middle finger”… etc… have intention and meanings)… I don’t think babies are aware of these. Perhaps a more clever actor could make sense of these less intentional gestures… paralleling/correlating them to other things that might be going on in the environment.

I’m putting my foot down on this one

That’s a good question, but I think actually random, non-meaningful actions are being used. Two points on this: first, babies aren’t great at ‘meaning’. I’ve been amazed to witness the full scope of things that babies don’t seem to understand. They don’t know how to fall asleep. They don’t have a concept of object permanence. Perhaps most importantly, they don’t have a sense of bodily integrity. At some point, they gain enough muscle to bend at the waist, and they literally discover their feet, without realizing they they are their feet. When they start to act intentionally, they don’t know at first that their right hand can reach things on the left, or that their right and left work together, and you’ll often see what look like cross-purposes between the left and right hand. All of this suggests that a lot of what’s going on is random (or effectively random), and looking for the meaning behind it is a fool’s errand.

Second, you can see randomness employed in adult humans as well, albeit a randomness constrained by understanding and expectation. Returning again to the box, an adult’s exploration of the box will be best understood by reference to at least some effective randomness. They will appeal to their understanding of what a box is, that a box has an unseen interior and usually opens in certain ways, etc. They will also appeal to a sort of intuitive Bayesianism to guide their exploration, so they look first for an obvious opening, then an obvious lid, then a less obvious lid, etc. But once these are accounted for, there is still some randomness: what part of the box did the person look at first? Which way did they spin it? If two possibilities are approximately equally likely, which did they attempt first? They will use randomness to break ties, to make decisions in the absence of information, and to otherwise fill in gaps. And since by hypothesis, the adult is exploring an unknown box, there will be gaps to fill where meaningful and intentional actions are insufficient to justify every step in the exploration.

Randomness is much more apparent in babies, because they have so much less else to go on, but it never goes away completely. Similarly, it’s most important early on, and if it is selected for, that selection will be strongest early use of randomness, where that use dominates, though some selection pressure will remain.

Keep in mind that a part of their random motions are due to not yet resolved neurological associations. A part of learning coordination is learning what to inhibit from motion while inducing into motion something else. It’s learning to govern through direct experience, trial and error.

Yes, James, that’s another point responsive to Incorrect that I neglected to mention: much of what babies do is random because they aren’t good at doing what they intend to do. But as I mentioned in the OP, sometimes that randomness leads to important breakthroughs, by generating an unintended action which can them be reinforced or inhibited through positive or negative associations.

True, but not to be taken to an extreme. :sunglasses:

“I don’t believe anything that I read or anything they say, so I am just going to screw everything up until I learn through The School of Hard Knocks.”

Have you ever seen a baby throw an object they were holding to the ground and watch as the adult dutifully picks it up and returns it to the baby.

The baby again throws the object to the ground (and most babies do this at a certain age) and the adult once again returns it to the baby.

What is the baby ‘thinking’. HA!

Probably the same thing that a dog thinks when you throw a ball or stick, “Do it again … Do it again … hee hee…”

The question is, why to cats just look at you like your an idiot. :confused:

JSS wrote:

I know what you are saying and I have been 'the idiot". Perhaps because we walk dogs to parks and beaches and take them with us on our camping trips, (with preference over our cats) and in the process train them to fetch a stick or a ball, to sit and stay, mainly for our own convenience. Yet some cats can be trained and they do fetch and play, especially the Siamese breed and there are the cats that ‘fetch’ by presenting their dead or injured prey, much to the chagrin of their owners.

Here’s a question: do you think, possibly, that even random actions put forth by the baby in question might only be actions it takes because you’re the one viewing it; that it might have other actions if viewed by others and that your viewing it and its reactions to the action will help define it as it grows at the same time defining you? What if you struggle in your mind to take a different tact and instead of the perspective you’ve been putting forth thus far in watching the child in question, you tack on a different perspective to see results? Maybe put some music in the background or read some books while around the child?

‘Babies’ are adventurous when they know the mother is nearby. The word ‘babies’ can be connected with 0-12 months. after this they are toddlers. What age is being discussed here?
Development for most has milestones at various ages. If a baby is neglected these milestones may take much longer to reach.

Even tiny infants express anger when the mother does not respond quickly enough to it’s hunger. They have a range of emotions they use almost immediately as this is their only means of communication.

A baby’s personality is formed and ready for action from the moment of delivery. They are not benign mounds of flesh they quickly learn how to manipulate, for the survival instinct is strong and active.

I don’t think randomness comes into the equasion, if babies don’t follow the pattern of development used for the majority, something is amiss.

What say you mothers out there?

Edit

What say you mothers/fathers out there?

I was very pleased with myself when an AI blog I follow had a post titled “Better Exploration with Parameter Noise”, discussing the role of noise in reinforcement learning. I probably shouldn’t be too pleased, since my ideas here are pretty basic compared with this post, but still a little validating. And it highlights that humans are reinforcement learners, and many baby behaviors should be expected to favor learning (and maybe we should be granting rights to other reinforcement learners…).

Agreed. There is the unanswered question, of relation between randomness and order in terms of inherited traits. Whether, there are pre-existing neurological channels upon birth, or, that all channels are blank, slated totally on learning after birth, is still not entirely based on reliable data, as of now. As long as this question remains, the factual basis of levels of randomness, temporally, and temporarily remains on hold.

That instinctual learning cannot be definitely disassociated from learning, through analysis of what part of the learning curve is due entirely to behaviorally set reinforcement, may present the primary obstacle to secondary considerations. Until some corresponding data suggests otherwise, this will still present a grey area, although the 1-3 years of primary determinative aspects have been doubtlessly been agreed upon.