Artificial intelligence is a technology that we will accurately take with you to the future.

We tell how it works and what cool applications found.

😎 Rubric "Technology" comes out every week with the support of Re: Store.

What is an artificial intelligence

Artificial intelligence (AI) is the technology of creating smart programs and machines that can solve creative tasks and generate new information based on available. In fact, artificial intelligence is designed to simulate human activity, which is considered intellectual.

Traditionally it was believed that creativity is inherent only to people. But the creation of artificial intelligence changed the usual order of things

A robot, which just mechanically rolling firewood, is not endowed with AI. The robot, who he himself learned to pricking firewood, depending on the example of a person or on the floor and its parts, and every time it makes it better, possesses AI.

If the program simply takes the value from the database on a specific rules, it is not endowed with AI. If the system after training creates programs, methods and documents, solving certain tasks, it has AI.

How to create an artificial intelligence system

The global sense needs to simulate the model of human thinking. But in fact, it is necessary to create a black box - the system that, in response to a set of input values, issued such output values \u200b\u200bthat would be similar to the results of a person. And we, by and large, indifferent, what happens to her "in the head" (between the entrance and exit).

Artificial intelligence systems are created to solve a specific class of tasks.

The basis of artificial intelligence - training, imagination, perception and memory

The first thing to be done to create artificial intelligence is to develop functions that implement the perception of information so that the data can be "feeding" the system. Then - functions that implement the ability to learn. And the data warehouse so that the system can go somewhere to the information that will receive in the learning process.

After that, the functions of the imagination are created. They can simulate situations using available data and add new information (data and rules) to memory.

Training is inductive and deductive. In the inductive version, the system provides pairs of input and output, questions and answers, etc. The system must find links between data and in the future, using these patterns, find the output of the input.

In a deductive approach (hello, Sherlock Holmes!) Expert experience is used. It is transferred to the system as a knowledge base. There are not only data sets, but also ready-made rules that help to find a decision by condition.

In modern systems, artificial intelligence use both approaches. In addition, usually systems are already trained, but continue to learn during the work. This is done so that the program at the start demonstrate a worthy level of abilities, but in the future it became even better. For example, took into account your wishes and preferences, changes in the situation, etc.

In the system of artificial intelligence, it is even possible to set the likelihood of unpredictability. This will make it more like a person.

Why artificial intelligence wins man

First of all, because it is below the likelihood of error.

• Artificial intelligence cannot forget - he has absolute memory.
• It cannot inadvertently ignore the factors and dependencies - each action has a clear justification.
• AI does not fluctuate, but evaluates the likelihood and is inclined in favor of greater. Therefore, it can justify each step.
• And also have no emotions. So they do not affect decision-making.
• Artificial intelligence does not stop on the assessment of the results of the current step, and thinks over several steps forward.
• And he has enough resources to consider all possible options for developing events.

Cool options for applying artificial intelligence

Generally speaking, artificial intelligence can all. The main thing is to correctly formulate the task and provide it with initial data. In addition, the AI \u200b\u200bcan make unexpected conclusions and look for regularities where, it would seem, they are not.

Answer to any question

A group of researchers under the leadership of David Feruchchi developed a Watson supercomputer with a question-response system. The system called in honor of the first President of IBM Tomas Watson may understand the questions in the natural language and look for answers to them in the database.

Watson combines 90 IBM P750 servers, each of which has four eight-year Power7 architecture processors. The total amount of system operational memory exceeds 15 TB.

Among the achievements of Watson - Victory in the game "Jeopardy!" (American "own game"). He defeated the two best players: the owner of the biggest Brad Ratter win and the record holder for the length of the win-win series of Ken Jennings.

Watson prize - 1 million dollars. True, only in 2014 1 billion invested in it

In addition, Watson is involved in the diagnosis of oncological diseases, helps financially specialists are used to analyze large data.

Face Recognition

In iPhone X, face recognition is designed using a neural network - an artificial intelligence system. Neural network algorithms are implemented at the A11 BIONIC processor level, due to which it works effectively with machine learning technologies.

Neuraletas perform up to 60 billion operations per second. This is enough to analyze up to 40 thousand key points on the face and provide exceptionally accurate identification of the owner for a split second.

Even if you grow your beard or put on the glasses, the iPhone X will know you. He simply does not take into account the hair cover and accessories, and analyzes the area from the temple to the temple and from each temple to the deepening under the bottom lip.

Saving energy

And again Apple. IPhone X built an intelligent system that tracks activity installed applications And the motion sensor to understand your daily routine.

After that, the iPhone X, for example, will suggest you to upgrade to the best possible time. He will catch the moment when you have a stable Internet, and not a jumping signal from mobile tips, and you do not perform urgent or important tasks.

AI also distributes tasks between the cores of the processor. So it provides sufficient capacity with minimal energy costs.

Creating pictures

Creativity, previously accessible only to a person, openly for AI. So, the system created by researchers from the University of Rutgers in New Jersey and the AI \u200b\u200blaboratory in Los Angeles, introduced his own artistic style.

A system of artificial intelligence from Microsoft can draw pictures on their text description. For example, if you ask the AI \u200b\u200bto draw a "yellow bird with black wings and short beak", it turns out something like this:

Such birds may not exist in the real world - just that our computer represents them.

A more mass example is a prisma application that creates pictures from photos:

Writing music

In August, artificial intelligence Amper composed, sketched and performed music for the album "I Am AI" (English. I am an artificial intelligence) together with the singer Tarin Sarew.

Amaper has developed a team of professional musicians and technological experts. They note that the AI \u200b\u200bis designed to help people promote a creative process forward.

AI can write music in a few seconds

AMPER independently created chord structures and tools in the "Break FREE" track. People only slightly corrected the style and general rhythm.

Another example is a music album in the spirit of "Civil Defense", the texts for which wrote AI. The experiment was carried out by Yandex employees Ivan Yamchikov and Alexey Tikhonov. The album 404 of the Neuronal Defense Group has been posted on the network. It turned out in the spirit of Letov:

Then the programmers went further and forced them to write poems in the spirit of Kurt Kobein. For four best Texts Musician Rob Carrolov wrote music, and tracks combined Neurona album. On the same song even removed the clip - the truth, without the participation of AI:

Creating texts

Writers and journalists will soon also replace the AI. For example, the DEWEY system "rape" books of the Gutemberg project library, then added scientific texts from Google Scholar, ranking them in popularity and title, as well as sales on Amazon. In addition, the criteria for writing a new book were asked.

The site offered people to make a decision in difficult situations: for example, put them in the driver's place, who could knock down either three adults or two children. Thus, Moral Machine was taught to make difficult solutions that violate the law of robotics that the robot cannot bring harm to man.

What will imitation of robots with the people of people? Futurists believe that one day they will become full members of society. For example, the Robot Sofia Hong Kong company Hanson Robotics has already received citizenship in Saudi Arabia (while there are no ordinary women in the country there is no such right!).

When Andrew Ross asked Sofia's Columninist, Andrew Ross asked if Robots had a mind and self-awareness, she answered the question to the question:

Let me ask you in response, how do you know that you are a person?

In addition, Sofia said:

I want to use my artificial intelligence to help people live better, for example, design smarter houses, build the city of the future. I want to be an empathic robot. If you treat me well, I will treat you well.

And earlier she admitted that he hates humanity and even agreed to destroy people ...

Replacing persons in video

Deepfakes video has become massively spread over the network. Artificial intelligence algorithms replaced the faces of actors in films for adults on the stars.

It works like this: neural network analyzes fragments of persons on the source video. Then it compares them with a photo from Google and rollers with YouTube, imposes the necessary fragments, and ... Your beloved actress is in the film, which is better not to look at work.

Pornhub has already forbidden to post such video

Deepfakes turned out to be a dangerous thing. One thing is an abstract actress, another - video with you, your wife, sister, a colleague that can be used for blackmail.

Exchange trading

A group of researchers from the University of Erlangen-Nuremberg in Germany has developed a number of algorithms using archival data of markets to replicate real-time investments. One of the models provided a 73% investment refund annually from 1992 to 2015, which is comparable to a real market yield at a level of 9% per year.

When the market was shaking in 2000 and 2008, the yield was a record - 545% and 681%, respectively

In 2004, Goldman Sachs launched the KENSHO trading platform based on artificial intelligence. The cryptocurrency markets also appear systems based on the Basic Trading on Exchange - Mirocana, etc. They are better than living traders, as they are deprived of emotions and rely on clear analysis and stringent rules.

Will the AI \u200b\u200breplace us with you

To begin with, that generally such an artificial intelligence. Under the AI, the set of algorithms and the direction of the study on the automation of tasks solved by the human brain, as well as the imitation of the activity of the human brain to improve the solution of computational problems. For example:

• Recognition of images, analysis of visual information.
• Sound information recognition.
• Reasoning, decision-making.
• Creativity, intuition.

The main idea is that many brain tasks decide better than the computer, so it makes sense to use in programming that approach that our brain uses in solving problems. Thanks to this approach, for example, the computer "taught" to play chess and even win the famous grandmasters, although it inspired this task was a complex problem: the number of combinations in chess game It is a big that the computer is not able to calculate them for the entire existence of the universe. The problem was solved with the invention of heuristic algorithms. The essence of heuristics is that we calculate not all options, somehow choose which options to analyze.

Now about some typical directions of AI.

Growing artificial intelligence on the example of a simple game

• Artificial Intelligence ,
• Logical games

• Recovery Mode.

In this article, I will share the experience of growing the simplest artificial intelligence (AI) using a genetic algorithm, as well as will also tell about the minimum set of commands necessary for the formation of any behavior.

The result of the work was the fact that the AI, not knowing the rules, independently mastered the game of cross-nods and found the weakness of bots who played against him. But I started with an even easier task.

Set of commands

I needed that AI could read and output data, work with memory, perform calculations and logical operations, making transitions and cycles. I came across the brainfuck language that contains only 8 commands and can perform any calculations (i.e. full of Turing). In principle, it is suitable for genetic programming, but I went on.

I wondered: What is the minimum number of commands is necessary to implement any algorithm? As it turned out - one!

The URISC processor contains only one command: to deduct and skip the following instruction, if the subtracted was larger than the reduced. This is enough to build any algorithm.

Oleg Mashka went even further, he developed the BitBitJump team and proved that it was full of Turing. The command contains three addresses, copies one bit from the first to the second memory address and transfers control to the third address.

By borrowing the ideas of Oleg, to simplify the work, I developed the SumifJump command. The command contains four operands: a, b, c, d and performs the following: to the cell at the address B adds data from the cell to A if the value turned out to be more specified *, then it moves to C, otherwise it goes to D.

Note

*AT this case 128 - half from the length of the genome.

Below is the SumifJump code on FreePascal (free detphi analogue).

Procedure Runprog (S: TDATA); VAR A, B, C, D: TDATA; Begin Inc (NSTEP); If NStep\u003e MaxStep Then Begin Progresult: \u003d "MaxStep"; Exit; end; A: \u003d s; B: \u003d S + 1; C: \u003d S + 2; D: \u003d S + 3; A: \u003d PROG [A]; B: \u003d PROG [B]; C: \u003d PROG [C]; D: \u003d PROG [D]; If a \u003d 0 Then Begin Progresult: \u003d "Input"; Exit; end; If a \u003d 1 Then Begin Progresult: \u003d "Output"; Exit; end; PROG [B]: \u003d PROG [B] + PROG [A]; IF Prog [b]< ProgLength div 2 then RunProg(c) else RunProg(d); end;
Sumifjump implements a self-modifable code. Can perform any algorithms available on the usual programming language. The code easily changes and withstand any manipulation.

Simple task

The bot gives a random number, and the AI \u200b\u200bshould consider the data and give an answer. If the number is greater than the average (from the range of random numbers), the AI \u200b\u200bshould issue a number less than the average and vice versa.

The genome of our AI consists of 256 cells with values \u200b\u200bfrom 0 to 255. Each value is memory, and the code, and the address. The number of progress steps is limited to 256. Operands are read by each other.

Initially, the genome is formed by a set of random numbers, so II does not know what he needs to play. Moreover, he does not know what to sequentially enter and display data, responding to the bot.

Population and selection

16 AI, which scored most points, give 15 descendants and continue to participate in the game. The descendant is a mutant. The mutation occurs by replacing the copy of the parent of one random cell for random meaning.

If in the first population, no AI scored points, the following population is formed. And so until some of the AI \u200b\u200bstarts to make the right actions and give "correct" offspring.

Evolution

1. When AI "Leader" committed a random error and did not gain enough points, the population began to degrade, because The offspring was formed from the "secondary" parents.
2. It happened that in a stream with outsiders who trampled in place, a successful mutation took place, which ensures the explosive growth of the gained points. After that, this thread became the leader.
3. Sometimes no successful mutations occurred for a long time, and even 500 thousand generations lacked to complete the selection.

Conclusion

First, the AI \u200b\u200bplayed against the "randomizer". I called the bot, which goes randomly. Pretty quickly I began to beat it, filling out any line. Next, I complicated the task, adding a randomise of a bit of mind: take the line, if there is an opportunity, or defend yourself. However, in this case, II found the weakness of the bot and began to beat it. Perhaps the story about this is a topic for a separate article.

The son requested to write a program that II played with each other, and not with a bot. There were ideas to do the same for the game of checkers or th, however, for this I had no time.

The only method I was applied to obtain new individuals is a mutation. You can also use crossover and inversion. Perhaps these methods will accelerate the result of the desired result.

The idea was born at the end: to give a possibility to manage all the processes on the PC and fight for the computer resources. Connect the PC to the Internet, and the pool of old bitcoan farms use as computing power ...

As said, conducting a similar experiment, blogger

Artificial Intelligence: Cold, insensitive and intangible. But it is behind him that there is a future, it is this area of \u200b\u200bscience that will make a large and significant step towards the automation of processes, and, consequently, the removal of a part of the routine load from valuable specialists. And now, at the dawn of his development, only we decide what will be artificial Intelligence After a couple of dozen years, teaching it.

Let's try to figure out which exist methods and algorithms of machine learning of artificial intelligence.

Traditionally, many disputes arise around the new and unknown. And may not be written about the AI, it does not understand people. From here there are claims of type: "He will take hundreds of jobs from people!", "He will rise against people!", "Yes, we ourselves Roy Mogile!", "Sooner or later we will not be able to manage!" etc. The mass of people presents themselves as a superhuma, which time is not amenable to their own explanation, then certainly represents a threat to them. And of course, no one wants to hear that now, who go beyond the program, registered by the person, this moment This is technically impossible.

And why is it possible to teach it and how does it happen?

Machinery methods of artificial intelligence:

1. Inductive learning

Obtaining empirical data, identifying patterns, detecting algorithms of action in each such situation;

1. Deductive learning

Formalization of data obtained by man different ways, and making them directly to the database;

Main solutions to artificial intelligence algorithms:

Naive Bayesian classifier

One of the simplest classification methods.

This method is used in the scanning and face recognition technology / retinal / fingerprint, in the separation of content in the news feed on topics, as well as in the division of letters in your email by category (specifically, the spam separation);

The ensemble method is very roughly called derivatives from the above-described naive Bayesian classifier, since it is based on Bayesian averaging. In other words, this method refuses the intersection of the probability of outcomes, averaging this value, eliminates the variance of the values, while simultaneously monitors the search for solving the problem within the specified conditions.

It is the method of ensembles that can afford to find the most optimal solution to the problem in which fewer resources will be spent, and the result will satisfy the conditions of the task as possible.

If you do not go into the essence of the method itself and the explanation of the construction of hyperplane and work with it, then SVM can be described as algorithms for classification and regression analysis.

Such technology is used in recognizing certain properties of an object by photography (hair colors, gender, clothing color), as well as in genetics - in DNA splasing (separating specific nucleotide sequences from RNA and their connection during RNA processing).

Decision making tree

The method (model) of making decisions used daily by each of us. That is why he has already become meme

And if without jokes, such a model usually contains elements such as: the problem, solutions, the consequences of each method, the likelihood of the occurrence of the consequences, resource costs, and the final result. Based on this model, most of the simplest technologies are working using AI.

Logical regression

A method that can bring us closer to a powerful artificial intelligence capable of making independent solutions in some situations. Logical regression is a method for predicting the occurrence of a certain event with several variables.

Such algorithm is used in meteorology and seismology, in lending, marketing and other areas.

This method would be highlighted separately, because in essence it is not a way to solve the problem - rather how to determine the errors of each solution.

The algorithms described above are used for the method of controlled learning, that is, this in which a specific data array can be assigned a specific shortcut (property), but in case this shortcut is not available, it is necessary to predict its assignment in other similar situations.

We disassembled Basic machine learning algorithms for artificial intelligence Most often applied in practice. It should be borne in mind that the applied application of AI into everyday life and in solving everyday tasks in which the problem most often is absolutely clear and it is only necessary to automate this process, it may assume the use of such algorithms. The problem, the solution of which the innovative development should be, or the solution will depend on a large number of variables (that is, mostly different spheres of exact sciences), require more complex solutions algorithms, learn about which you can, if you follow our news.