Artificial Intelligence 2020 - is there progress? - page 58

 
Maxim Dmitrievsky:

as long as you don't get a "wi-fi tip."


No... You shouldn't have put a sign on the drone that said "Russian Post" and everything would have been fine.

 
Andrey Dik:

No... They shouldn't have put a sign on the drone saying "Russian Post" and everything would have been fine.

I wonder if a drone costs 1.2 Mio, how many times they have to be smashed against the wall to get the monthly budget of Buryatia

60,000 times \ 12 = 5,000 drones per month

I think I exaggerated the scale of the tragedy
 
Maxim Dmitrievsky:

I wonder if a drone costs 1.2 mio, how many times they need to be smashed against a wall to get the monthly budget of Buryatia

60,000 times \ 12 = 5,000 drones per month

I think I exaggerated the scale of the tragedy.
I wondered how many of these copters Buryatia Post can produce in 3 billion years? That's 18 trillion copters. Each one has a payload of 2 kg. That's 36 trillion kilograms. That's 36*10^14kg. The earth weighs 6*10^24. Unfortunately, the chances of saving the earth this way are slim.
 
Why do we think AI does not exist globally?
Is it possible to believe that man is the pinnacle of knowledge and only man can create AI.
Everything has already been created before us.
AI has long been embedded in chromosomes. If someone doesn't know about it, even the best scientific journals won't help.
===
AI, as we understand it, gives the crooks an excuse to control the behavior of the population. And that has serious consequences for the world's population, and let's face it, each and every one of you...
 
Andrey Dik:

Why... not at all.

The engine itself is of little interest, but as an interface between a real AI (if it ever will be created) and a human, it is very good, i.e. analogous to the speech centre of the brain.

Of course, I was kidding a bit. The software is generally very worthy and it will have a lot of practical applications. And the percentage of 90% of users, not really understanding how it works, will seek (and surprisingly find) in his opuses the highest meaning.
 

A neural network is a thing that divides feature space into regions during training. Based on this property it is possible to build (teach it), various applied tasks, such as classification, approximation, optimization, clustering, etc. Of course, each such task can be solved by the specialized mathematical apparatus - to approximate by the least squares method, to optimize by linear programming methods, to cluster by the Voronoi partition, etc., but the NS does it itself, knowing nothing about the mathematical apparatus that lies in the mathematical solution of the task. The main thing is to choose correctly a training target and correctly configure the feature space.

Another remarkable quality of NS, explaining why it is so widespread in nature, is its natural simplicity and that even the simplest primitive architecture, a la Rosenblatt one-layer perceptron - can realize many tasks, of course, under restriction on linearity of division in the feature space. But this limitation was only reached by evolution hundreds of millions of years after the appearance of the simplest NS.

 

It is possible to imagine that everything happened approximately so - after a rather complicated structure of a cell was formed in the course of evolution, receptors for lightness, temperature, salinity, etc. began to appear in it on the one hand, and on the other hand, executive mechanisms allowing to change the place in space depending on the state of receptors. Connecting signals from receptors through a certain adder and transmitting commands to the output from it to the acting mechanism, here it is a miracle: a prototype of the elementary NS appears.
.

Its weight coefficients were selected and fixed by natural selection. This is how reflexive behaviour evolved. As organisms become more complex and control is specialised in a single organ - the brain - behaviour becomes more complex. In principle, large complexes of reflex systems can realize very complex behavior, but its optimization requires selection over tens of thousands of generations and millions or even billions of individuals.

To eliminate this contradiction, the next revolutionary step is taking place - the brain is capable of conditionally reflexive behavior. Now, in the course of a single lifetime, the individual may optimise his own behaviour in response to his environment.

An evolutionary 'arms race' ensues - more and more complex behaviour is selected for in predator/prey systems. At some point, in order to manage complex behaviour and prioritise targets, there is a need for a specialised interface in the brain - the mind - to predict the behaviour of fellow species and make assumptions about the presence of enemies or food.

After tens of millions of years, the evolution of this interface and the enhancement of the ability to generalise led to the emergence of consciousness - self-consciousness, the 'I'. In my opinion, it is only after the emergence of the self-consciousness property of an AI that it will become a non-biological mind.

 
sibirqk:


The evolution of this interface, the enhancement of the ability to generalise, after tens of millions of years, has led to the emergence of consciousness within consciousness - self-consciousness, its own 'self'. In my opinion, it is only after the emergence of self-consciousness in an AI that it becomes a non-biological mind.

Self-consciousness is also only an assumption so far, it has not been proved and shown to be exact or not to be exact in a tree. Already today it is possible in principle to set a task for any system to evaluate itself. A car, what about the engine. Alice, how many people were satisfied / dissatisfied based on the survey. And there you have it self-awareness))) in a simplified form.
Alice how do you rate yourself, let me see. Wow, just ace how))))) 999% want me))) Oh, man, thank God it's not Vanya.)

(All right, that's it.)

 
sibirqk:

A neural network is a thing that divides feature space into regions during training. Based on this property it is possible to build (teach it), various applied tasks, such as classification, approximation, optimization, clustering, etc. Of course, each such task can be solved by the specialized mathematical apparatus - to approximate by the least squares method, to optimize by linear programming methods, to cluster by the Voronoi partition, etc., but the NS does it itself, knowing nothing about the mathematical apparatus that lies in the mathematical solution of the task. The main thing is to choose correctly a target of training and correctly configure the feature space.

Another remarkable quality of NS, explaining why it is so widespread in nature, is its natural simplicity and that even the simplest primitive architecture, a la Rosenblatt one-layer perceptron - can realize many tasks, of course, under restriction on linearity of division in the feature space. But this limitation was only reached by evolution hundreds of millions of years after the appearance of the simplest NS.

Thank you. As of late, I've started to understand it that way. Repeated exposure of neurons (probably) changes their conductivity, which eventually leads to the formation of a trait - i.e. the invariant part of the "reflected" object within the neuronal matrix. Then, "younger" features of phenomena and forms are imprinted because they are more polymorphic than the invariant (also due to repetition, i.e. learning), and the structure of the object is formed, which begins to branch hierarchically within the network. This is, as it were, the beginning of a classification of envelope forms of invariant. I've just started to understand it a bit...

 
sibirqk:

It is possible to imagine that everything happened approximately so - after a rather complicated structure of a cell was formed in the course of evolution, receptors for lightness, temperature, salinity, etc. began to appear in it on the one hand, and on the other hand, executive mechanisms allowing to change the place in space depending on the state of receptors. Connecting signals from receptors through a certain adder and transmitting commands to the output from it to the acting mechanism, here it is a miracle: a prototype of the elementary NS appears.
.

Its weight coefficients were selected and fixed by natural selection. This is how reflexive behaviour evolved. As organisms become more complex and control is specialised in a single organ - the brain - behaviour becomes more complex. In principle, large complexes of reflex systems can realize very complex behavior, but its optimization requires selection over tens of thousands of generations and millions or even billions of individuals.

To eliminate this contradiction, the next revolutionary step is taking place - the brain is capable of conditionally reflexive behavior. Now, in the course of a single lifetime, the individual may optimise his own behaviour in response to his environment.

An evolutionary 'arms race' ensues - more and more complex behaviour is selected for in predator/prey systems. At some point, in order to manage complex behaviour and prioritise targets, there is a need for a specialised interface in the brain - the mind - to predict the behaviour of fellow species and make assumptions about the presence of enemies or food.

After tens of millions of years, the evolution of this interface and the enhancement of the ability to generalise led to the emergence of consciousness - self-consciousness, the 'I'. In my opinion, it is only after the emergence of the self-consciousness property of an AI that it will become a non-biological mind.

A very beautiful concept, but I would add that the highest form of Mind, on a par with self-consciousness, is mental activity, without which man is like a very intelligent animal. So that is what the psyche is - beyond our comprehension...