Main menu




What is the difference between AI, machine learning and deep learning?

 1. Introduction

 Recently, I often see and hear the word AI in news and books. It is treated like a savior with a shortage of manpower, or written like a devil who robs people of their jobs, but its substance is not easy to understand. Moreover, I often hear people say that they don't know exactly what they will do for the work they are involved in.

 Another confusing thing is its name. AI, machine learning, and deep learning are all unclear what they are and what they are related to, and I can't help but ask people. In this article, I would like to explain the substance, focusing on machine learning.

2. What is AI?

 AI is an abbreviation for Artificial Intelligence, which is an old and new word for artificial intelligence in Japanese. The dream of building intelligent intelligence like humans on a computer has disappeared in the past several times.

 In fact, AI itself does not have a formal definition. AI is indebted when using some nifty applications and services. The Japanese input that you use to input sentences is also one of the AI ​​applications that convert characters by considering the surrounding words and frequency of use. A long time ago, it was sometimes called AI conversion. Another type of AI is an application that plays an opponent in computer games such as shogi, chess, and go.

 In this way, AI is just the name of an application that "looks intelligent" like a human being. It doesn't matter what the logic in it is. Maybe it's just fetching from a huge database, or maybe it's using machine learning, which I'll explain later.

 3. Why AI is booming now

 Research to realize AI has been underway for a long time, but there are four reasons why it has been booming recently.

 The first is a dramatic increase in computer processing power and recording capacity, as well as a dramatic reduction in its utilization costs. Second, the spread of Web services and smartphones has dramatically increased the amount of digitized data required to learn and use AI. Third, it has had a great impact on the business in marketing, advertising and mail-order sales. AI is now taking on important functions that are indispensable for recent web services and services targeting smartphones, such as product recommendations, appropriate ad matching, and categorization of customer behavior. .. And fourth, with real-world feedback and financial back-up from the business world, talented researchers and engineers are working on AI to improve algorithms and develop great software one after another. .. Due to these virtuous circles, AI has become unprecedentedly successful.

 The excitement of AI has spread to the industrial world. Until now, the main use of AI has been IT services for consumers centered on smartphones. Services with many users such as Google (Google), Amazon (Amazon) and Facebook are supported by analyzing digital data such as personal behavior and information collected in large quantities and utilizing it for marketing and advertising. ing.

 However, the wave of AI is rushing to industries that have not benefited much from IT, especially manufacturing and service industries. This is because it has become possible to collect digitized data through the IoT (Internet of Things). Especially in Japan, due to the combination of rapid labor shortages and the aging of skilled workers, efforts to save labor by utilizing data have not been waited for. Tools for general engineers are appearing at the right time, and the hurdles are gradually decreasing.

4. What we can understand by machine learning is "correlation", not "causal effect"

 Machine learning is a technology that uses large-scale data for prediction using statistical algorithms. It is the most popular technology used in recent AI. By combining statistical analysis and simulation with a large amount of data, the computer can automatically analyze / predict the target thing.

 However, because it is a "statistical analysis", it does not understand things. To the last, it predicts the transition of the numerical value expressed in the data and classifies the character string or the image. There is no formal definition for this either, but statistical analysis emphasizes understanding of the subject, and machine learning emphasizes future prediction.

 The important point is that machine learning reveals "correlation", not "causal". Correlation can show that multiple variables are involved, but it is unclear if they are really related. It is human work to find meaning and prove "causality". Fortunately, human work is not lost and becomes more important. Here's an interesting example, so take a look.

 As shown in Figure 1 , there is a correlation between divorce rates in Maine, USA and per capita margarine consumption. But are there any consequences to reducing margarine consumption and divorce rates? It's a joke because it's a story that can be understood by thinking for a moment, but it may be difficult to make this distinction when considering the main business of a manufacturing engineer. For the time being, it's best to keep in mind that machine learning deals with correlations.

 It may sound negative to machine learning, but of course it makes a lot of sense if you only know the correlation. Machine learning comes into play because there are too many factors involved and the cause and effect are unknown in the first place. In addition, it is meaningful to be able to make comparative studies by making numerical values, or to have an idea of ​​analysis of a huge amount of data. Also, it takes a long time to understand the cause and effect. The reality is that you can't spend that much time in business, so even if it's not the correct answer, you can often use it enough if it's a good solution.