2020/11/25 07:04:52

How to develop AI?

How to develop AI?

When pursuing the efficiency of a company, productivity will be dramatically and effectively improved by analyzing data such as forecasts by processing data. Therefore, attention is now being paid to efficiency improvement using AI. Introducing an article about AI development.

Is it possible to develop AI by self-education? Introducing the mechanism of artificial intelligence and learning methods

AI (artificial intelligence) is now well known no matter what field you work in, and even people who are not familiar with engineering have more opportunities to hear it.

From now on, in the age of AI, not only knowledge but also skills that can be actually operated are required.

In addition, it is said that the mechanism and development of AI is much easier than in the past, according to its familiarity.

This time, we will introduce the mechanism of AI that people living in such a modern age should remember, the knowledge necessary for developing artificial intelligence, and teaching materials that are useful for learning.

Has AI development become familiar?

It may seem like only recently that the keyword AI has begun to take on reality, but AI development has continued even before that.

However, it is thought that the reason why AI has become so popular is that AI of major IT companies has been released to the public as open source one after another.

AI is becoming more open source
Open source means that you can open the code of a certain programming to an unspecified number of people, and anyone can freely copy and paste the code or modify it independently and rearrange it into the original program.

In general, programs that companies develop independently are confidential and are often kept private unless otherwise instructed.

Publishing the program increases the risk of exposing security vulnerabilities and creating rights issues such as patent infringement.

On the other hand, the merit of open source is that the program can be developed not only by companies but also by anyone who is interested.

The scale and speed of AI development in companies is different from that of individuals, but by making it publicly available, Google and Microsoft provided opportunities to further improve the accuracy of AI and create new services. That's why.

Creating AI from scratch is costly, but if you customize the existing foundation and assemble it, you can dramatically reduce development costs.

This is probably the reason why AI has been installed in various apps and hardware.

A service equipped with AI that even beginners can make
This greatly simplification of the difficulty of AI development has made it easier for people who have never been involved in programming, let alone AI, to develop services equipped with AI.

For example, various models are active in various fields, such as sales prediction AI that predicts the next sales from past sales, and AI chatbot that can give flexible answers closer to human beings while being a text chatbot. ..

AI is not a futuristic robot, but a useful tool to further simplify your work. Furthermore, most AI is an extension of statistical analysis software, so if you think of AI as a new statistical software, the hurdle may seem low.

Knowledge required to develop AI

Next, I would like to introduce the basic knowledge for developing AI.

AI development seems difficult, but if you look at each process in several stages, it may be easier to get a feel for it.

How AI works
Like other systems, AI is basically just one of the programs that is a collection of code.

AI, which is generally popular, is called specialized AI, and it demonstrates excellent ability in one task.

The roles of AI, such as image recognition, statistical prediction, and text communication, vary depending on AI, but it is still difficult to combine them in a complex manner like humans.

And now, it is a kind of machine learning called deep learning that dramatically increases the processing power of many AIs.

Machine learning gives AI a large amount of data with a specific purpose and process and encourages it to process data voluntarily, while deep learning gives only purpose and data and leaves the data processing method to AI. It's about to end up.

The method of processing a given data is derived by a unique process, sometimes giving a perspective that humans could not think of.

Deep learning is also an advanced technology and may not be available to beginners overnight.

However, if we know that we will be able to handle such futuristic technology in the near future, it will lead to motivation.

AI development process
The AI ​​development process is generally divided into three stages.

The first is data collection. It is necessary for humans to collect and accumulate a large amount of useful data according to the AI ​​to be implemented.

The second is data learning. We will load the collected data into AI, specify the process and purpose in advance in AI so that the expected data processing can be performed, and let AI learn correctly.

If learning progresses with AI to some extent at this stage, the basis of the system will be completed.

The third is the implementation of the system. We are at the stage of expanding the UI and preparing it so that it can be used as a service so that anyone can use the built AI correctly.

In this way, the system using AI is completed, but the strong point of AI is that it can proceed with learning autonomously even after completion.

The more users use the service, the more new data samples they get, so the advantage of AI is that it has the potential to improve accuracy the more it is used.

Programming language required for AI development

Next, let's take a look at programming languages ​​that are often used in AI development sites.

AI development, like normal programming, does not use any special language, but the popular ones are becoming limited to a few languages.


Python is by far the most popular programming language in the field of AI development.

Originally, Python has gained a lot of support in the field of statistical analysis, and it is a language that is also very active in the field of AI.

It also has a reputation for being easy to read and easy to write, as well as being easy to read and write by entering a small number of lines of code.

It's also a great language for beginners to learn programming.


Like Python, R is a programming language that plays an active role in the field of statistical analysis.

Due to the thickness of the community and the large number of users, there is a lot of information on the Internet, and there is no shortage of troubleshooting.


Julia is one of the fairly new programming languages ​​released in 2012.

Julia, which is based on C language, is characterized by its speed.

As coding becomes more complex, optimization steps must be taken to make it work smoothly, and Julia is a language that has the potential to significantly reduce the burden of such work. It has become.

5 Teaching Materials to Help You Develop AI on Your Own
Finally, I would like to introduce some of the best learning materials for self-taught AI.

Learn by running with Python! A new machine learning textbook (Shoeisha)

This is a must read for anyone who wants to learn AI using Python.

Since you can learn the machine learning mechanism that is the basis of AI while using Python, it is a great book for those who want to know the mechanism of AI and how to use Python.

Introductory Python 3 (O'Reilly Japan)

This is a volumey study book that comprehensively describes how to use Python.

Not only beginners but also intermediate users can refer to the necessary information in a reverse lookup, so it is a book that can be kept on the bookshelf for a long time.

Let Python do the boring things-automated programming that even non-programmers can do (O'Reilly Japan)

This is not only a rudimentary way to use Python, but also a teaching material that describes know-how for automating daily miscellaneous tasks using Python.

The basis of AI is automation of work, but you can learn the essence of AI from this book.

First R: From very rudimentary operation to introduction of statistical analysis (Kitaooji Shobo)

For those who want to focus on statistical analysis, we recommend learning using R.

As with Python, there are many types of R language-related books on sale, so it's a good idea to pick them up at a bookstore.

What AI can and cannot do, and the four powers to live in a business society (Nihon Hyoronsha)

Although this is not a guidebook for implementing something directly using programming, it is a useful book for setting an index on how to use AI.

It describes how to use AI from a business perspective, which helps you understand AI with a different approach than engineering.

At the end
AI has become a technology that we can learn familiarly with the spread of programming languages ​​and the open source of AI by companies.

AI development is also composed of programming when you open the lid, and the rough development process is not so different from other systems.

It's a good idea to choose a language that suits your interests and interests, such as Python and R, and start learning from reading practical books.