Pretend intelligence: from chatbots to neural networks

People are constantly looking for a way to simplify their life and delegate some work to simple and complex devices. The desire to use our energy economically is inherent in us by nature. At the beginning of evolution, this led to the emergence of the simplest tools, such as a wheel, lever, hammer, shovel, etc. Over time, tools became more functional and convenient. But gradually their most important drawback emerged - the use of any tool required the presence of a person.
The solution to this problem became possible only with the advent of computer systems, when artificial intelligence ceased to be fiction. Back in the sixties of the last century, the US Department of Defense was working on methods of simulating mental activity for military purposes. Today it is already a reality, and technologies that imitate the work of the human brain are available to everyone.

 What is artificial intelligence and how does it work, what can be done with it and why a profitable business is impossible without it? Everything in order.
 It's hard to imagine life without AI. There would be no smartphones, no Tesla cars, Amazon sales would be many times less, and Netflix could not offer an interesting movie. Antivirus software would be weak protection without heuristic analysis, and anti-spam filters would let ads through and mailboxes filled with spam.

Many of us interact with Siri, Google Assistant or Alice on a daily basis. Such projects are complex programs that recognize human speech and understand its meaning. The development cost of this software is quite high. This is understandable, because a programmer must solve many problems: develop a voice recognition module, teach how to analyze an image and distinguish objects on it, and minimize the number of errors. And, the most important thing that an AI developer must do is to put in the code of the virtual assistant an adaptive behavior algorithm that changes depending on conditions. If the user specifies the time, an answer should follow with the current time; if the command to set the alarm is received, the corresponding settings of the new timer should be made, etc. Typically, a smart software developer describes AI behavior in several ways.

Scenario based machine intelligence.

The first option consists in compiling all possible scenarios for the development of events and indicating the responses of artificial intelligence in each case. This method is quite suitable for solving relatively small and simple tasks - for creating an automatic chat bot in a messenger (Viber, Telegram) or for automating payment systems. When it is necessary to organize an appointment with a doctor on the web page of a medical company, provide some background information or show a list of goods in stock - such a chat will be very effective.

 Adaptive AI

But, unfortunately, the method of setting up AI in some scenarios is far from always applicable.
Imagine artificial intelligence playing chess. It is impossible even theoretically to describe the development of all possible chess games. American mathematician, father of computer science Claude Shannon (the scientist who coined the word "bit" to denote a unit of information) back in 1950 calculated the minimum number of non-repeating chess games - 〖10〗 ^ 120.

This "Shannon number" is incredibly huge, it is many, many times greater than the number of atoms in the observable part of the Universe. With such a large amount of data, playing chess requires an algorithm built not on scenarios, but on adaptation to current conditions. The player makes a move, analyzes the alignment of forces, predicts further developments and chooses the optimal movement of the piece on the board.

 Neural networks and deep learning

 An even more complex third option for describing AI behavior is also possible - a self-learning algorithm, in which artificial intelligence is like a child striving to learn about the world. The most effective area of ​​machine learning is the use of so-called neural networks (deep learning). They are a mathematical model that describes the internal connections between the components of a simple processor array. Each processor works to receive and return a signal with the rest of the cells of the "computer brain". In simple terms, a machine algorithm seeks to repeat the work of the nerve cells of a biological organism. The more complex the topology of neural connections, the higher the level of its learning ability and the more difficult tasks it is able to solve.

The possibilities of a trained computer intelligence are practically endless. The more experienced such a system is, the more it resembles the mental activity of a living being.
A neural network can solve almost any analytical problem. She is able to search for a formula for a cure for a disease, perform forecasting of quotations in the financial market or make other analytical assumptions based on statistical values. Such a "superbrain" learns to recognize, without hardware, just one video, human movements (similar to motion capture technology) and repeat these movements on an animated model. In-depth analysis allows you to make a deepfake video by replacing one actor with another. The neural network can even learn to copy an artist's style or generate a human voice with the right emphasis and shades of emotion.
Today we are only at the very beginning of the development of deep learning technology. Its progress is not going as fast as we would like so far. This is due to the high level of investment and considerable requirements for computing power. However, progress is visible - the current level of complexity of machine intelligence is quite sufficient to delegate such intellectual tasks as searching for information on the Internet or even writing a short essay.

Business chatbots

You are probably know the pop-up window in the corner of the screen "We are glad to welcome you to our online store. How can I help you?" Such virtual assistants significantly improve the quality and speed of customer service. Plus, it's a good tool for building sales funnels. The program can determine the interests of the client no worse than a human sales assistant and offer him relevant goods or services. Such tools easily integrate with social networks and popular messengers such as Telegram, WhatsApp, Viber. This way of automating business has long been the norm, and companies that still underestimate the capabilities of technology risk incurring losses in the coming years, including losing their business.
A simple automatic chat installed on the site significantly reduces running costs. It works free of charge 24 hours a day, without requiring monetary compensation or vacation. By taking on the "monkey work", this module reduces the burden on the rest of the firm's employees. Machine chat responds more quickly than a real operator. A chatbot devoid of the “human factor” never makes mistakes - accepting an application, it will not make a mistake in the customer's phone number or address, and so on.
 In addition to the obvious uses for automated chat, there are many other uses that the business project manager may not be aware of. Those who are far from web technologies should get competent advice on how best to use a chatbot. For example, a virtual assistant can promote a brand on social networks: offer to participate in contests, inform about ongoing promotions, and also form a loyal audience by sending out interesting stories, viral videos, funny gifs, and so on. The bot can search for clients on social networks using targeting, as in contextual advertising, taking into account the geography, time and interests of people.