Define the world so that novel concept can be understood spontaneously.

Don’t be smart

Smart usually indicates the ability of identifying causal relation. Someone is smart if he could solve difficult problems or make accurate predictions, presumably.

Recent AI researchers and companies emphasise on how smart their algorithms or systems could be. They developed algorithms that could help us solve many complex(high-dimensional) problems effectively and sometimes it exceeds human performance significantly. Many believe that a system is possessing human level intelligence if it could be as smart as human. This assumption indeed guide us developed many useful tricks. The only problem is:

we are not smart

Being smart is not an essential constituent of human beings. Most people, at most of the time, are astonishingly stupid. We are looking for patterns all the time and most of these patterns has absolutely nothing to do with causal relation or could only be effective within very limit domain.

We are good at learning patterns comparing with other species. But only a small proportion of patterns represent causal relations and learning patterns is different from identifying causalities. In fact, during the past tens of thousands of years history of Homo Sapiens, we had not developed an efficient way of identifying causal relation until the most recent couple of hundred years.

Being smart spontaneously is a high level function , it is a result of composing pattern learning, intention, memory, imagination and many other lower level abilities in the RIGHT way. I believe that the pattern learning ability is one of many foundational abilities of human(biological) level intelligence.

The way we learn pattern is different from many typical pattern learning algorithms. Actually, it is the ability to define novel pattern or more precisely, it is the ability to define novel. it combines/bridges ontology and epistemology. In this document I will discuss some unique properties that are necessary for a system to reach human(biological) level learning ability.

No Free Lunch

A general-purpose universal optimization strategy is theoretically impossible,
and the only way one strategy can outperform another is if it is specialized to the objective function 
    -- "Simple Explanation of the No-Free-Lunch Theorem and Its Implications"

In our daily life, there are many intuitive examples which reflect the notion of NFL theorem.

Normally

• we use coffee grinder to make coffee not juice extractor,
• we would like to wear soccer shoes while playing football not slippers.
• paratroopers are equipped with parachute not umbrella.
  

The fact that specific task requires specific solution is the most common knowledge which we take it for granted, and all these solutions are basically specialized optimization strategy for specific objective function.

In every use case of all these specific strategies, there is an essential component:

HUMAN

• WE identify the objective requirements.
• WE decide or design what strategy will be used.
• WE cooperate with all kinds of strategies designed by us.
• WE are chefs.
• WE are drivers.
• WE are pilots.
• WE are captains.
• wE are astronauts
  

The best way to guarantee the well functioning of a system is to integrate HUMAN into the system. - WE coordinate the work of different sub-systems. - WE are the inevitable core and ultimate information source which support every specific strategy under exceptional circumstances.

We are the general purpose universal optimization strategy.

SO, does the existence of HUMAN disprove the NFL-theorem?

Assume

• There is an universal optimization strategy Z.
• There is not any specific ”objective function” exists before Z is being implemented.

Maybe

We are this optimization strategy Z, so we could define the existence of all objective functions, furthermore, we could draw the conclusion of NFL-theorem.

Information Representation

To be continued …