Artificial Intelligence and Neural Networks: What Are They and How Do They Work?
August 9th, 2016
Artificial intelligence is one of the most important IT trends in recent years. It is a trend rich people invest in and, at the same time, it’s the trend that frightens futurists and journalists. So what is it, how does it work, and should we be afraid of it?
What is artificial intelligence?
Artificial intelligence is both a rather young science as well as application and device development technology that uses special algorithms. The essence of AI is to get the computer to be reasonable, or, in other words, simulate artificial mind. The main challenge in this is to define the “intelligence” and as a result, it is hard to estimate what would it take to create an intelligent algorithm.
On one hand, the computer can remember a lot more information than an average person, the computer thinks quickly, but a person can learn and, based on the information s/he is given, draw conclusions and make decisions. In addition, a person has an ability to create - write music and poems, paint, make movies.
One of the goals of the artificial intelligence is to teach the computer the skills that are hard to measure - for example, learning abilities and decision-making based on their own knowledge. Another challenge is to model and reproduce creative processes. Therefore, the news that artificial intelligence has learned to write poetry or draw pictures are incredibly important for the development of the field as a whole.
By and large, AI’s abilities can be summarized as the ability to solve problems of non-computing nature and to perform actions that require data analysis and processing as well as subsequent decision making. The global goal of AI is to create intelligent self-learning robots that could perform the tasks formulated in natural language with a minimum of human intervention and adjust their performance. For example, you can ask such robot to make you a cup of coffee and specify that you’d like a small amount of sugar in it. The next time, the robot will take into account your wishes without you having to repeat the conditions again (as well as wishes of other people who ask it for something).
Artificial Intelligence and Neural Networks
There are trends within the AI field as well. The neural network is one such trend. Neural networks’ models actually are an attempt to simulate the operation of the human nervous system. Artificial neural networks, just as the neural networks in the human brain, are not only able to store and reproduce information, but also study their own mistakes.
The main feature of the neural network is the ability for self-learning and acting on the basis of previous experience. A properly designed neural network should make fewer and fewer mistakes every time it performs a standard action since it’s learning from the feedback (for example, comments from people).
A neural network consists of a set number of neurons - the computational elements, each of which relates to a single network layer. Data is processed on all layers, where each layer gets information that was changed and is ready for the layer to process it. For example, the neural network first finds out the common characteristics of the image and sees that the picture contains an image of an animal of a human. The next layer then complements the previously learned information by studying other characteristics and defines whether the photo contains a picture of a wolf or a five-year-old girl.
All the tasks neural network is capable of dealing with are associated with learning. It can be anything from forecasting, decision making, and data analysis to pattern recognition and optimization.
For example, neural networks are used for speech or image recognition. They can be used to construct algorithms for unmanned vehicles’ operation so that they can know what to do in an unusual situation. Another area of neural networks application is recognition of illegal content, even if it has been masked as legitimate.
Best-known Systems of Artificial Intelligence
Every day there are more and more news about the use of AI in various spheres of life. Large companies invest in AI and neural networks to use them in mobile applications, games, and entertainment, as well as a ton of other cases.
Chess supercomputer Deep Blue defeated the world chess champion Gary Kasparov in 1997, winning with the score of 4:2. The supercomputer was developed by IBM.
IBM Watson is another IBM’s creation that played against other players in the American game “Jeopardy!” In 2011, the supercomputer has defeated the two most famous players.
Today, various applications that use artificial intelligence and data analysis are built on the basis of IBM Watson. Early summer this year, IBM and X Prize Foundation has announced a contest called IBM Watson AI X Prize, where developers have to create an app for IBM Watson. At the moment, Watson is particularly widely used in the field of digital health - systems for disease distribution forecasts and diagnosis is built using its algorithms.
Google DeepMind is a division of Google engaged in research in the field of artificial intelligence. Its most famous project - AlphaGo program for the Go game, which won the championship in the Go game in March 2016 with a score of 4:1. In addition, DeepMind developed an artificial intelligence system that can learn how to play and win popular video games. The department plans to create algorithms for games of complex strategies and racing simulators. DeepMind is also working to develop algorithms for the understanding of natural language, image generation patterns, and its division DeepMind Health is working in the field of digital health.
Microsoft also works on AI development. This spring the company launched a new project called CaptionBot - an image tool that allows users to upload photos to the site and CaptionBot will come up with captions for that picture based on what is in the picture. The website uses Computer Vision API, Emotion API, and Bing Image API. For more entertaining projects of the corporation, check out WhatDog, a service which identifies dog breed based on the picture and HowOld, a service that determines the age of a person in the photo.
Neural networks are used not only in complex algorithms designed to solve serious problems. There are quite a few entertainment applications that became quite popular due to their features, which use neural networks. We’re talking here about MSQRD video service and Prisma photo processing app. These tools demonstrate the abilities of neural networks and allow them to learn and improve.
AI research is done by other large companies such as Facebook’s AI Research and Chinese Baidu Institute of Deep Learning.
Artificial intelligence is used by banks and financial companies to determine various factors like future solvency of a client and predict the value of securities in the stock market game. Pattern recognition, one of the AI applications, is used in medical diagnostics, spam filters, and military affairs.
The future of artificial intelligence and neural networks lies in the technology gradually becoming more and more ordinary for people to use in their everyday life and in completely different areas of their lives - from e-commerce system recommendations to predicting traffic jams in large cities. However, time may come when mankind will want to give up the active use of AI, especially since a number of scientists and businessmen, led by none other but Stephen Hawking himself, demanded to limit the development of weapons with AI, saying that such weapons could one day threaten the existence of mankind.
Let’s just not create something along the scenario of Ex Machina.