OpenAI recruitment page screenshot

lol

“Necklace” unsupervised learning algorithm.

There are some entities which are easily comprehend by almost everybody, yet are very difficult to describe in details or give strict science definition. That is true for almost any term from the field of our intelligence, and term intelligence itself. Thought, feeling, understanding, consciousness, learning, recognition and the interest, the sense of beauty in image, form, melody, and the sense of humor. It is hard for intelligence to formulate these entities, same as it is hard for a snowman to answer “what is snow?” question.

I feel that “Interest” is the one among these entities, that is undeservedly deprived of attention from AI researches. And the one that I am long interested in.

In the 4 pages pdf paper below you can read my thoughts on the interest and how it can help to build unsupervised learning HTM.

https://www.dropbox.com/s/8nnn1w10yw90m2h/%E2%80%9CNecklace%E2%80%9D%20unsupervised%20learning%20algorithm.pdf?dl=0

Harmony.

Harmony, the musicality - is the simpliest and plane exposition for intelligence interest.

You got sequence of sounds, some of them are predicted by mind, some are unexpected. The right balance of these makes music being loved.
If you got too many unexpected sounds - the music is too complicated and feels like a noise.
If you got too many predicted sounds - the music is too dull and uninteresting to listen.

Also, it seems that here lies some mind delimiter clues.
Some old melodies are interesting, even if we do listen them for the hungred+1 time. But they not interesting if we listen them 3 times in a row.
It means that some part of the melody are great at forgetting it. And some are coming easy to predict. That is how greatest hits are made.
Because some part of the melody is forgetting quick, but looks like periodic and predicted when you hear it or play it.

If you ask me.

If I ever will interview someone to real-AI developer position, then wrong answers to some of these would be critical.

What is Human?
Human is a living being. Need details? Human is an mammal animal. If I hear someone disagree that, then I am asking back with the raised eyebrows: "Not an animal? What then? A mushroom or a plant?"

Is that possible to create mashine that will act similar to human intelligence?
Yes. No discussion with those who disagree.

What is the meaning of life?
Meaning is when someone have a meaningfull plan or reason in something. Like the meaning of saucepan is to cook sauces in it. When life orginated on earth, there was no intelligent life to have any meaning of anything. So the question is nonsense.

What is Intelligence?
The feature of a living being(so far) that helps forward them to behave better, according to their biological needs. I can't say where exactly intelligence starts on the line from unicellular life sensor-reaction behavior, to behavior of Albert Einstein. But I put this point closer to start, then most other people do.

You see what you look at.

You see what you look at.

The name of this blog contains one of the cornerstone specific of my approach to ai.
Nervous system of any living creature not just classifying, or sensing, or understanding. It makes moves. From the begining of the evolution this was and still the first function of any nervous system.
Sensing-response.

Most of the modern ai developers complicating the task by cutting off any active behavior. Forcing the mashine to understanding, to classifying, to making predictions being paralised! Can you imagine a paralised animal having any kind of intelligence?
Can you imagine a paralised newborn, able to learn, adble to feel the world around?

It is obvious for me, that understanding and learning process are coming from comlicate cooperation of sensory and behavior. It must be much easier to understand image on your retina, when you have the ability to control your eye! It must be much easier to classify sensory input from your ears, when you have the ability to make sounds!

To play, is much more interesting, than to watch.

Little close to create god. But what for?

Self-improved real artificial intelligence will finally self-improove himself to some beyong state. And therefore will be all-knowing. As the proverb says knowledge is power. So it will probably become all-powerful.

The only all-powerfull thing we can imagine is a god.

That is why we are little-close to create god.

What kind of a god would you create? What goals, what intentions, what missions would you gave to him?

The first answer coming to the most minds:

"Peace on earth and hapiness to all the people."

That could include victory over poverty, victory over every desease, ethernal life and all that.

That is what I call acetic bacterium manager.

Let's imagine that some acetic bacteries in some 5 liter canister was lucky to create some prosperic technologies. They was able to create intelligence creature, with the unlimited power. They create something like... you, for instance. And all they want from you is to maintain their 5 liter canister in perfect acetic bacteries conditions, for them to have "peace and hapiness".

Don't you think that will be too litle, for the creature like human?

At the same time we can't imagine goals for the creature like god.

I'v tried, and I asked my friends for any thought on the matter, with no result.

The only win-win answer I have is to set up his mission for knowledge. That in fact delaying of the answer, but there is some point in such delaying.

So. What that will be like.

My real AI will be a software.
It should "live" in some world, real or virtual.
It should be connected to a real time sensors, to have real time data.
And it should have some real time outputs, connected to devices, wich can operate the body in the world... thereby changing incoming data flow, from the sensors.
It will learn patterns, only those, wich it will be able to find. Patterns will be not restricted by time.
Patterns will not be restricted by data type: incoming, outcoming, and deep dots will be the parts of a pattern.
"It" just must be something recognizable and unique, to be a pattern.
The single-step(flat) pattern will be just a special case of time-sequence pattern with the zero time period.
The patterns will be in a fuzzy logics. And they will not be a "space distributed", in means of Numenta and followers. (No random initiate data please!, thank you).

It will have "good-meter" behaviour targets.
And it will plan it behavior with the help of reality modeling(consciousness), that will use curent data and the prediction, based on the patterns base. Plus some imagination.

And, the cake's special cherry, - "it" should be curious.  Yes, that is one of a generous real-general AI aspect. It should have curiosity and the abilities to act curiously.
And this curiosity will be a part of behaviour.

So, here it is, the result of a few years thoughts in a 5 minute text.