List of things that artificial intelligence can do!

I’ve heard a lot about AI, but I think there are many people who don’t really understand what it can do specifically.

Therefore, this time, we will introduce in an easy-to-understand manner what AI can do and what it cannot do. Please use it as a reference.

AI (artificial intelligence) can be divided into two types: specialized and general-purpose

First of all, I would like to explain the types of AI. AI can be broadly divided into two types: specialized and general-purpose .

specialized artificial intelligence

Specialized artificial intelligence does not have the sensibility and thinking circuits of humans, and is also called ” weak AI “. Since it automatically learns and processes by specializing in issues in a limited area, it can only deal with specific issues.

For example, Go AI cannot play Shogi. Image recognition AI cannot recognize voice. Machine learning technology such as deep learning, which is currently being used, is also classified as specialized artificial intelligence.

general artificial intelligence

Artificial general intelligence ( AGI ) is also called ” strong AI ” because it has human-like self-awareness and can think and act on its own .

Not only can they deal with specific problems, they can deal with a wide range of problems just like humans. However, artificial general intelligence has not yet been realized.

Classify what AI can do into 4 categories: Recognition, prediction, etc.

This time, we will classify what AI can do into four categories and introduce them. We will introduce the following four.

1. voice recognition
2. image recognition
3. natural language processing
4. Prediction based on past data

I will explain each.

voice recognition

Speech recognition recognizes human voices and converts them into text. In recent years, devices that can be operated by voice, such as smart speakers, have become popular, and voice recognition technology is used behind the scenes.

In recent years, many services that make use of speech recognition have been created, such as tools that automatically generate meeting minutes and tools that automatically answer telephone calls.

image recognition

Traditionally, image recognition technology has not been highly accurate and impractical. However, with the development of deep learning technology, the accuracy of image recognition has improved significantly, and its use in industry has come to be considered.

With conventional technology, in order to recognize an image of a cat as a cat, it was necessary for humans to define detailed features such as the number of whiskers and the shape of the face, create rules, and program it. However, deep learning automatically learns the features of images from a huge amount of data, making it possible to recognize various images with high accuracy.

Image recognition is also attracting particular attention in the use of AI, and in addition to being used to detect abnormalities in factory lines, the use of AI-OCR, which converts handwritten characters into text, is also progressing in recent years.

natural language processing

AI can recognize text as well as voice and images. Natural language processing is a term that refers to the technology used by humans to process natural language on a computer.

In recent years, Google, the American non-profit organization OpenAI, and LINE in Japan have been focusing on research and development of natural language processing, and the accuracy has been greatly improved.

Our lives would not be possible without text communication. If natural language processing technology develops and it becomes possible to understand the meaning of human words, the range of applications for AI will expand further, but no AI that can process words like humans has yet been created.

However, in recent years, the use of chatbots that can automate customer service using natural language processing technology has increased, and opportunities for natural language processing technology to play an active role are increasing. There are many characters that you can talk to on SNS.

Prediction based on past data

By learning time series data (data recorded over time), we can predict certain future numbers.

For example, in store sales forecasting, by learning the passage of time (days), weather and temperature, the number of passers-by, and past sales data, it is possible to determine how each element, such as weather and temperature, will affect the results (sales). Learn what influences This makes future sales predictable.

Similar forecasting technology is being used in many fields, but demand forecasting is particularly popular. For companies, if they produce too many products, they will have production costs and inventory costs, which will lead to lost opportunities.

Therefore, by using AI to predict demand, it becomes possible to produce products efficiently.

Two things that AI cannot do | Areas where it is not good at creating and reading emotions

From here, I will introduce two things that AI cannot do: ” creation ” and ” reading human emotions .”

to create

Humans are capable of creating things from 0 to 1. Machine learning technology, which is currently being used, enables image and voice recognition and prediction, but it is said that it is not suitable for generating 1 from 0.

On the other hand, even if we say that we create something, most of the time we abstract past experiences and divert them to come up with ideas. Human beings can create new value by having various experiences in their daily lives.

In terms of abstracting and creating the results of learning, AI is also good at fields. Especially in recent years, a technology called GAN that generates images of human faces and whole bodies has been attracting attention. Technologies such as DALL-E, which creates pictures from text, are also attracting attention. AINOW publishes an article explaining DALL-E .

reading people’s emotions

The second thing that AI is not good at is reading people’s emotions.

When we read people’s emotions, we judge not only words but also various information such as tone of voice, facial expressions, and context. Making decisions based on multiple pieces of information is called multimodal.

Currently, AI can partially read human emotions, such as recognizing people’s faces and reading their emotions. However, not all our emotions are expressed only on our faces.

If AI can perform multimodal processing in the future, it may become possible to read human emotions.

What AI can do based on four use cases

Here are four use cases to learn more about what AI can do. The following four use cases are introduced here.

1. Defective product detection
2. Catch forecast
3. Nursing care robot
4. Road cavity detection

I will explain each.

Defective product detection

AI is often used for defective product detection. This is an AI that automatically detects defective products by learning product images. For this, image recognition and anomaly detection technologies are used.

Catch forecast

AI is also used to predict catches. By analyzing the underwater conditions with AI and learning the past acoustic data and landing data from the fish finder, it is possible to predict the landing data by inputting the current acoustic data.

Nursing care robot

AI can measure the distance between the robot and the object using sensors, measure the weight of the object, and sense the fragility of the object . In addition, because it can easily lift heavy objects, it is also used for heavy work such as nursing care.

Road cavity detection

AI can detect cavities from radar images. By using AI, it is now possible to detect anomalies that cannot be detected by humans.

Current AI challenges

There are three main AI challenges currently being considered:

1. inability to respond flexibly
2. need accurate data
3. can’t relate to people’s feelings

I will explain each.

inability to respond flexibly

As we have introduced so far, AI learns from given data and performs specific processing. As a result, it cannot do anything other than what it was designed to do when it was programmed, and it cannot flexibly respond to various situations.

can’t relate to people’s feelings

Another problem with AI is its inability to empathize with people’s feelings. Since AI is only a substitute for a part of human intelligence, it cannot be said to be close to people and do something like humans.

need accurate data

And we need accurate and rich data. Since AI learns from data, it cannot learn or process without data.

What will AI look like in the future? What is the future of humans?

With the evolution of AI, some people may be concerned that their jobs will be taken over by robots.

However, isn’t it possible that productivity will improve as technology evolves, and wages will increase as corporate profits increase? General-purpose artificial intelligence that learns and understands like humans is the goal of current artificial intelligence research.

If this artificial intelligence is developed, human life may be further improved.

what we should do

If AI will do the work that humans have done so far, what kind of work should humans do? One of the answers to this problem is to turn to the side that uses AI.

As we have seen above, there are things AI cannot do. Humans have the ability to compensate for this part. AI should be viewed as something that expands the limits of human perception.

Jobs that fill this role include data scientists , AI engineers, AI planners, and AI producers. It is important not only to fear that jobs will be taken over by AI, but also to think about what role we can play in that situation.

summary

I think we now know that while AI has the ability to do many things that humans cannot do, there are also things that only humans can do.

Now that AI has come to be widely used, it is important to correctly understand what AI can do and think about what humans should do.