Five Popular Posts Of The Month

Sunday, May 12, 2019

Two notes on AI: what it is and what it is not.

I.
Let use your imagination (an ability only humans have).

Imagine you are in the Library of Congress and need to find a book that contains one specific sentence. You may get lucky, but most probably you die before you find it.
But you can hire thousands of interns, and they will be flipping through pages in books and comparing the sentence on a piece of paper you gave them with sentences in a book, and one will find the book rather soon.
This is exactly what AI does these days, and will do for many years to come; AI is just an intern with thousand hands and eyes, and a brain strong just enough to learn some patterns, although for AI learning takes much more time than for a human intern.
Everything that has a pattern can be digitizes and represented in a numerical form.
People identify a pattern.
People develop a procedure for teaching AI how to recognize a pattern, train AI to recognize a pattern (instead of old school - writing a code for selecting the right pattern; they train it as a circus animal), and use AI to search for the pattern it trained to recognize.
And AI can do it much faster than any human, even the smartest one – exactly like thousands of interns but alone.
This is no different from replacing thousands of diggers 
with shovels by one huge excavator.
And people who design neuron networks – the hart, or maybe we should say, the brain of AI - get paid hundreds of thousands of dollars.
Even though they do not have even a workable common definition of what they do – i.e. artificial intelligence.
They are trying to “teach” AI a common sense. Even though they do not have a working definition of a common sense.
I have a very big hole in my character, a real issue.
I happen not to like liars and pretenders (especially those who call themselves scientists; you would not be surprised if a politician would be a liar and a pretender). 
And currently all AI-involved developers known to me (by their writings) are both.
They pretend that things they don’t know about intelligence aren’t important.
AI technologists throw out from intelligence everything except “pattern recognition”.
And then they lie to everyone that that  
is the actual intelligence.


Sunflowers (as well as billions of other living species) follow the pattern of the sun traveling through the sky. It does not make it (and other species) intelligent.
Pattern recognition is not intelligence.
Pattern recognition is only a basis for having intelligence, like basis for walking is having legs. However, if one has legs but his/her nervous system is broken one will not be able to walk, as well.
And yet, there is no AI professional who would openly said that currently there is no AI, not even close, all they have is an advanced software with limited self-adjusting abilities that can mimic certain aspects of pattern recognition happening in a brain, and they don't even know how to make an actually intelligent machine because they don't know what intelligence is and how intelligence works.
Because admitting that would force them to adopt a new term instead of "AI" (and risk millions in funding).
Let use your imagination (an ability only humans have).
Imagine this conversation:
#1 (pointing at a bicycle) "This is a rocket!"
#2 "No, it's a bicycle"
#1 "A rocket can move from point A to point B. And this thing can move from point A to point B - this is a rocket"
#2 "No, a rocket can fly and this thing cannot, it is not a rocket"
#1 "I don't care, a rocket can move from point A to point B. This device can move from point A to point B - this is a rocket!"
Do you get the analogy?
AI technologists point a finger at their neural network devices and say "this is an artificial intelligence!" And I say "No, it's not".
Of course, you can say that this analogy is wrong, the difference between a bicycle and a rocket is not the same as the difference between a neural network and an actual artificially made intelligent device. And I say, yes, you are right. 
The difference between a bicycle and a rocket is much much smaller than the difference between any current neural network and an actual artificially made intelligent device.
That is why an actual artificially made intelligent device does not exist and will not be developed for a very very long time.
For people who have grown-up children, this analogy may be helpful. 
Parents celebrate every single achievement of their children - the first time the child stand up and stands for 5 seconds before falling on the floor, the first time the child walks across a room, etc. But a reasonable parent knows how far the child yet is from being called "an adult".
"Parents" of "AI" are very proud of their "child"; but that "child" is "just making its first steps and just learning how to speak", however its "parents" tell everybody "Our kid just got a PhD in rocket science".

II.
AI technologists are trying to teach AI how to learn on its own.
So far scientists do not know how to make people learn on their own. First, they need to figure out THAT.
It’s like that. Eventually AI may become intellectually as powerful as people (even the smartest ones). When that happens, we will have to treat AI the same way we treat people, the source of intelligence does not matter, it does no matter how does this individual get to life - biologically or manufactured. The treatment - education, culture, beliefs, etc. will define the behavior.
Maybe, in the future, all knowledge and skills and ideas and beliefs could be “recorder” in AI in one “mental” action, so its “education” will not take years, like for growing humans, but that again does not make big difference.
What matters is what knowledge and skills and ideas and beliefs will AI have in its brain.
And THAT depends on people teaching, educating, “nurturing” AI.
So, as long as there will be bad people who want to use guns, bombs, or robots, or complicated computer systems to do harm to others, there will be bad AIs representing a threat to all humans. But -
- and this should be a relief for all us -
Hence, all that talk about AI ethics is BS; and all those people who talk about it just enjoy all the attention. 
The confuse or deliberately substitute the ethics of AI (as an intelligent agent) with the ethics of using AI (by an intelligent agent).
The former does not exist.

The latter is "trivial", or at least no different from the ethic of using an atomic bomb, or a hammer.
So, AI ethics has nothing to do with AI.
It’s about people using sophisticated tools for god or bad.
People!
Currently what media call AI is just a computer software with limited self-adjusting abilities. No more. It has relatively advanced pattern recognition abilities. Relatively to other software. Maybe soon even relatively to some not very smart and educated people. But calling it intelligence is like calling a gorilla human because it also has legs and arms.
There is no much of a difference between an old-fashioned computer and AI so far.
People establish a pattern, describe it, put the description into numbers, but in the end instead of writing for a computer a program for recognizing the pattern among other structures, they train it to recognize that patter like circus trainers train circus animals do tricks.
BTW: the distance from current AI to real Intelligence is much more than the distance from a gorilla to an educated human. So, humanly thinking machines are decades ahead of us. I stress - machines. Because a biological AI can already be created - just take an egg and a sperm and put in a host (soon it could be a test tube). But the dream is making a machine that is stronger and faster than a human and at least as intelligent. "Faster and stronger" is already here. But the last part will take decades. And the main reason for that is that AI scientists do not know and do not want to know what intelligence is. Hence they cannot model it. ▢◼️Q.E.D. 

"А те хто нихочуть таскать люмень будють таскать чюгунь."

A food for thoughts: 
When one wants to say a sentence, it is never formed in full at once in one's mind. At first, in some part of a brain, just an intention of what should be said is being formed.
Everyone who wrote at least one essay, not mentioning a book, knows this feeling. Now one knows from the beginning the exact words he or she will be writing from the start to the end of the book. 
Then that intention guides the formation of words from the first to the last. That process is happening in a different part of a brain. 
it is a very important notion:
1. one part of a brain is responsible for generating initial intention.
2. another part of a brain is responsible for the exact formulation of the text guided by that initial intention.
Again - TWO different parts of a brain are acting in concert to generate a textual thought.
NONE of the current neural networks is even close to mimic this type of functioning.
And, as I said many many times before, no one in AI field even thinks about this.
The neural networks they design these days a much more complicated than ones ten years ago, but they a still extremely primitive relative the structure of a brain - the structure that no one really knows yet.
Anyone who wants to understand the structure of intelligence needs to start asking people of all professions: actors, teachers, chefs, drivers, ... - what do they feel or think right before they feel or think that what they feel or think (but only if this sentence makes sense for them).
For the fans of "The Game of Throns" the following analogy may be helpful.
What is the significance of Bran become the three eyed raven? Of course, he was the bate for the Night King, there was no other way to defeat his army. But he also was the only one who could find the truth about Jon Snow. And that explains why Jon Snow could ride a dragon. And that explains why the dragon did not kill him when Snow killed the Mather of Dragons - for the dragon it was between two "dragon kings", anyone else would have been burned. But George R. R. Martin did not know all the details at the beginning of the book. All he knew was he did not want to kill Jon Snow in the end. That is why in his mind from the very beginning George R. R. Martin made Jon Snow a Targaryen. And envisioned Bran as the one who would revile this for us. And all what George R. R. Martin knew from the beginning is that he would make Jon Snow to kill the Mather of Dragons and he would let him live after that. All the actual details came later during the writing. 
Currently, there is no one on the whole planet who knows how such a process "idea - first, details - later" can be modeled. But this very process "idea - first, details - later" represents one of the most important features of intelligence: i.e. an ability to create a solution to a problem one has never solved before (and represent this solution in textual or audio symbols). Creating a solution starts from generating an idea, and then that idea gets more detailed representation, tested and rejected, accepted, or modified.

There is such profession as a movie critic; a book critic; a play critic, etc.
I have established a new professional field, and I am the only member of that field, so far - an AI critic.
No one likes critique, hence no one likes critics. However, in other fields the role of critics has been established as important - that is the sign of the maturity of the field. 
In AI people just ignore any critique altogether.

I have no time to put all my writing on AI in a single and edited piece, but they all provide different insights on the matter and available here: https://www.cognisity.how/p/arti.html