General human algorithm for learning Commonsense

 General human algorithm for learning Commonsense –

Whenever we notice any data, we notice properties of that data upon some reflection. For example, suppose a kid visits your house and sees a big, brown cupboard with books inside it. Say, the “data” he notices in it is – ‘there are books kept inside a cupboard’. (Suppose the cupboard is transparent).

Now, there are an enormous number of properties to this “data”. That, the books are kept on shelves in the cupboard. That, the books are vertical in the cupboard. That, they are touching each other face to face in the cupboard. That some books in the cupboard are red, some blue, some brown. That the books are on 4 layers of shelves of the cupboard. That, the books are inside a door of the cupboard with spherical knobs. Etc.

Notice that these are all properties of the “data” – ‘there are books kept inside a cupboard’.

Now, the kid leaves your house. He goes on with his life. He goes to school, to his tennis coaching, to the playground in his building, eats, sleeps etc.

 

Consider some of the  above properties of the “data” –

1.       books are on shelves in the cupboard,

2.       books are vertical in the cupboard,

3.       books are touching other books in the cupboard,

4.       books are on 4 shelves in the cupboard

Each of these properties get stored in the mind in a “Dual Format” –

Format 1) The whole set of properties associated with the “data”, and

Format 2) Each property, OUT-OF-CONTEXT from the “data”, whereby each becomes one different, general instance of a property of the key words (books, cupboard) in the “data”. Each becomes a data-property piece, wherein this new data now here is the keyword in the original “data”.

Let me explain. Lets see (2) in detail. For example, the 4^th property in the list above – books are on 4 shelves in the cupboard – gets out of context from the “data” and gets stored as a separate instance of “a cupboard having a property that it has 4 shelves”. So now there is an isolated, out of context data-property piece stored in the mind - ‘a cupboard having 4 shelves’.

So now the mind has a list of properties of the “data”, and also, isolated, out of context data-property pieces like ‘a cupboard having 4 shelves’.

 

Lets focus our attention back to (1) – the list/set of properties of the original “data”. When the kid sees another instance (and further, more instances) of the same “data” – ‘books kept inside a cupboard’, firstly, other lists of properties get enlisted in those instances, each for an instance. From across these lists of properties of the various instances of the same “data”, the corresponding properties ((i.e. orientation of the books, the touchability of the books, there being shelves etc.) gather together in the mind (owing to the sheer correspondence) and if the “values” of the properties (like say, the books being ‘VERTICAL’) get repeated across a lot of these instances, a piece of commonsense expectation about the world – that books are kept vertical in cupboards – gets formed in the mind.

Lets extend this to (2). Sometimes the kid might see something like just books on a table, or say in a bag. Here the same above process repeats. Remember, he has an isolated, out-of-context piece like ‘books being vertical’ in his mind. So ‘Books’ (the keyword in the “data”) becomes the new data with a property-value of its being ‘VERTICALNESS’. This will be compared similarly as in the process above, with the values of the orientation of the books seen now (table, bag) and checked for repeated-ness and consequently for the formation of a commonsense expectation about the world. (In this particular case, no particular commonsense will be learned since he sees books on a table being horizontal and vertical in bags. May be something like - books are kept vertically OR horizontally wherever they are.) 

COMMONSENSE "INFRASTRUCTURE" IS IN-BORN

COMMONSENSE "INFRASTRUCTURE" IS IN-BORN -

The essence of what anything means lies in what it means to us, which in turn lies in our 'EXPERIENCE' with it. Now, firstly, you cannot even talk about something which you haven't "had an experience of" in some way. Even if you conceive of an arbitrary, imaginary word, nothing like which exists in real, you still have experienced the phonetics of it.

What is the essence of 'steps' (of a staircase)? You would immediately narrow down and say - they take us up (or down), bit by bit. Now, there are so many aspects to steps - 1) the planes are parallel 2) they are made of marble or stone or whatever 3) they keep shifting to the right (seen across) 4) each is made of perpendicular dimensions etc. But these things are not quite the "essence" of steps. The essence is as is said above. How does one get to it? When one actually uses steps he gets the feeling - the experience - of being elevated or descended. He also moves ahead, but that experience is had even while normal flat walking. So the core of the essence of the meaning of steps is the bit by bit elevation or descendence, which occurs to us when we experience them.

Moving on further, can we experience everything? Jupiter and Saturn? Yes, as kids, we see pictures of them in the encyclopedia. Now, extending the above point in a sense, the bulk of every person's experience with a picture of Saturn is going to be the same. That there is a sphere and that there are circular rings around it. Some specifics may differ, but there will be a common core to all kids. This is where I am hinting at and going towards commonsense. Nobody is going to perceive Saturn as - "there is an arbitrarily shaped patch on the ring-plane, then moving above and to the left there is nothing (air), then there is a solid curved patch, then to its left there is a similar colour and material solid patch etc. So there is a commonality in the basic way in which all of us perceive things around us. There is some common perceptual machinery with which we absorb the data around us, to a certain minimal extent. This is a function clearly of our embodiment - combination of the locations and arrangement of our heads and other organs. This basic machinery contributes to WHAT and HOW we perceive the reality around us. Hence this cannot be separated from what we call our 'commonsense'. Yes, the fact that food is served in plates is not inborn, but rather learned and acquired, but the basal machinery to acquire that, which is inseparable from the common SENSE itself, is inborn.

This shows up in our experience with things around us, what they mean to us, and thus in what the essence of their meaning, and casually speaking, just their meaning, is. The observed is a function of the observer. Human commonsense is "human".

A Micro-theory of learning Commonsense Motion

 A Micro-theory of learning Commonsense Motion - 

We all know that - 

If I push an object to the left, it will move to the left.

If I push an object to the right, it will move to the right.

If bent/turned along a curve, it will move curvilinearly.

.......

How do we acquire the above commonsense?

When a kid plays with his toy car or blocks, he hits/pushes/throws them in various directions. They move (accordingly). 

One preliminary thing one could say is that the kid has (acquires) a broad sense that wherever he will throw the toy (with the corresponding intention), it wont go in a drastically otherwise manner. That is, if he throws the toy ahead, it wont jump back over his head to his back. Or if he pushes it to the left, it will not jump up. 

So, coming back to the original question of how we learn this, the point is - How does the kid suspect in the first place that there is a correlation between the direction in which he pushes the toy and the direction in which the toy moves? (both of which are only not drastically different).

While moving the toy car, the kid gets a certain sense. It learns something. What?

A kid gets this sense - how I will "CONTROL" the block (car) with my hand, so will it move.

'Intention' + 'Physical Adjustment' (= f(intention)) = 'Control' => 'Way in which the object moves'.

As the kid grows even further, and gets investigative,

Differentials (which are micro-actions and micro-effects) of this above equation are what give the kid the "finer" laws like - 

i) if pushed to left, it moves to left

ii) if turned curvilinearly, it turns along a curve

iii) if lifted upwards, it moves upwards

Lets see these micro-actions / differentials of the above equation : 

d(intention) = slightly intended 'movement', in any sense - say to the left or right, or curvilinear or up or down

+ d(physical adjustment) = a slight push/bend in the corresponding intended sense - to the left/right/down/up/curve

= d(control) = d(intention + physical adjustment)

=> d(way in which the object moves) = slight resultant observed motion of the car - slightly to the left/right/along curve/down/up

The point is this - as the kid grows and matures to do intentional investigations, he observes these above "differential" laws (which are easy for him to test experimentally through micro-actions and seeing over correspondingly short intervals of time, their micro-effects).

Can any 'machine learning' can teach/imbibe this sense into a computer?