Neural Networks for Dummies: Artificial Intelligence vs.Human Mind

Neural Networks for Dummies: Artificial Intelligence vs.Human Mind

The phrases “artificial intelligence”, “machine learning”, “artificial neural networks” in recent years are increasingly heard in the reports of scientists and news media, and conspiracy theorists are sounding the alarm, predicting an imminent uprising of machines in the near future. How soon the electronic processor will surpass the human brain and whether it is worth attacking the washing machine now – we understand further in the material.

Neural networks in simple words

Avoiding complex formulations and heaps of terms, a neural network is most easily described as a device that functions on the principle of the human brain. Neural networks can be of very different types (convolutional, recurrent, direct propagation, etc.), differ in the type of task (analysis, forecasting, pattern recognition, etc.) and in their structure, but in any case it is a kind of a mathematical model presented in the form of software and hardware and with the goal of learning to draw conclusions from updated data like a person making a particular life decision.

Neural networks

What’s the difference between a regular program and a neural network?

The main difference between a neural network and a conventional computer program is the ability to learn. That is, the result of the neural network operation can be based on data that did not exist at the training stage.

A brief history of the emergence of neural networks

People began to think about building neural networks with the development of neurophysiology. The more information scientists received about the processes taking place in their own brains, the broader prospects opened up for neural networks. The main problem until the end of the 20th century was the lack of serious data sets – researchers had to mainly work with texts, build semantic maps of languages, teach programs to recognize syntactic structures, etc.

Everything changed with the advent and spread of the Internet. Today, huge amounts of data are available for training neural networks, including multimedia libraries. Thanks to such a base, people can invest less labor in creating programs using the learnability of neural networks. For example, the PROMT translation software has invested a huge amount of labor-hours of linguists and programmers, while the Google Translate service in the near future will be able to provide a better result with much less effort on the part of the authors, based on a gigantic text base.

Examples of neural networks

Artificial intelligence and neural networks are already quite firmly entrenched in our lives. During testing in 2012, the chatbot “Evgeny Gustman”, imitating a boy from Odessa, managed to convince more than 20% of experts that he is a real person. After 3 years, “Sonya Guseva” deceived 47% of the judges. The effectiveness of such a neural network in Internet marketing, for example, is difficult to overestimate.

Remember in the movie I, Robot, where Will Smith’s character asked the machine if it could compose a symphony or create a masterpiece of art? In 2016, scientists at Hakodate University in Japan created a program that independently wrote the book The Day the Computer Writes a Novel. This book entered the final of the literary competition, surpassing 1,450 human authors.


Near future

Futurist Ray Kurzweil said that full-fledged artificial intelligence will be created by humans in 2029. Many consider this assessment too pessimistic. Already in 2022, IBM and the Swiss Polytechnic from Lausanne promise to show a functioning model of the human brain. To understand the scale, the following figures can be cited:

  • The average brain is 8.6 billion neurons;
  • Each neuron can have an average of about 2 thousand processes;
  • In total, this is about 150 trillion synapses;
  • Conventionally, each synapse is thousands of molecular triggers;
  • Roughly speaking, the human brain consists of 150 quadrillion transistors, while in the most powerful artificial processors, the number of transistors barely exceeds 10 billion.

Of course, in this case, the operating frequency of an organic brain is much inferior to that of a machine.

artificial brain

However, the human brain has natural limitations. It requires a lot of energy to work (up to 20% of all calories burned by the body), and the size cannot be increased significantly. In the case of a man-made device, these limits can be neglected and a much larger analogue with any power consumption can be created. Actually, a person, with all his might, will not be able to realize the scale of a network with a million neurons (this is how to imagine the size of a galaxy), and if their number increases thousands of times, then it will be rather difficult to predict the capabilities of such a system.


ProApk on Google News
ProApk on Telegram
ProApk on Twitter
ProApk on Facebook