Ever Evolving Artificial Intelligence--A Future Where Humans and AI Coexist--

Key to the current AI boom: Deep learning (machine learning)

Self-driving cars, smart devices that execute verbal commands, and computers that defeat chess masters and professional shogi players-Today artificial intelligence, or AI, is becoming increasingly more familiar to us in our lives.

Although there are various different definitions of AI, Fujitsu considers it to be the technology to "make computers perform what humans do by using human intellect."
For example, humans can adapt to circumstances by making decisions and solving problems based on the knowledge and wisdom they have acquired through experience. On the other hand, although ordinary computers excel at execution according to set programs, they are unable to perform processing beyond the predetermined scope.
Lately, however, using artificial intelligence, some computers have been able to go beyond the limits of programs and learn and make decisions like humans do.

Figure 1: The evolution of artificial intelligence

Let's take a look back at the history of artificial intelligence.
AI has a surprisingly long history, and its first boom dates back to the second half of the 1950s to the first half of the 1970s. The word "artificial intelligence" was coined around that time. The first boom had passed soon thereafter, and the second AI boom arrived in the 1980s. The artificial intelligence during this time mainly acquired knowledge from humans, not yet the kind with which computers learn and make decisions autonomously. In the 1990s, a computer defeating the world chess champion made big news. However, it was still extremely difficult to teach computers human knowledge and manage it back then, and an AI winter arrived once again.

Figure 2: Neural network

In the 2010s, a new technology called deep learning (machine learning) was thrust into the limelight, and the third AI boom began. Deep learning is the latest artificial neural network technology, which simulates the mechanisms of the human brain. The human brain consists of neurons and synapses that connect neurons to transmit information. Artificial neural networks (or simply called neural networks) modeled these neurons and synapses. This model was studied actively during the second AI boom, but, for example, the neural network used in a mobile robot Fujitsu Laboratories developed back then had only three layers with 29 neurons and 232 synapses. Technological advances have expanded it to an enormous network, enabling computers to execute deeper learning. The object recognition network Fujitsu Laboratories developed in 2015 was expanded to have seven layers with 1.1 million neurons and 730 million synapses.

Figure 3: Changes of neural network associated with changing AI booms

Let's take a closer look.

Let's say we want to develop a system that tells if a face is that of a human, a monkey or a dog. Until now, an engineer needed to write a program to provide conditions of a human face. That is, a human was educating a computer. However, for a system like this one to work, we need to teach a computer every possible condition of being a human, a monkey or a dog. It would take an enormous amount of time to just write such a program.

With deep learning, simply by reading a large amount of image data, a computer can autonomously acquire the conditions of being a human, a monkey or a dog.
This progress was made possible through improvements in the computer's processing power, advancements in machine learning algorithms and the spread of the Internet and big data, which has made it easier to obtain large amounts of information.

Will AI pose a threat to humanity? What AI needs to collaborate with humans

This ever-evolving AI is not always received favorably. People who oppose AI argue that "it will take over human jobs" and "AI will eventually pose a threat to humanity as in sci-fi movies."

Against this backdrop, Fujitsu has been pursuing R&D into AI for over 30 years to create human-centric AI systems designed for coexisting and collaborating with humans.
What Fujitsu keeps in mind with AI research is the idea that "the reason for AI's existence is to support humans in the areas they are not able to do by themselves or carry out efficiently." AI may seem very smart, but often it is not good at handling things humans can do easily or does not understand common sense. On the other hand, we humans are unable to process a vast quantity of data as quickly as AI can.
Because humans and AI have different areas of specialty, we believe humans and AI can thrive together by humans appropriately specifying the range of discretion, such as "these matters will be left up to AI's judgment" and "human operators will make final decisions."

Another thing Fujitsu keeps in mind is to create "AI that continuously develops." Humans grow by learning from their failures and gaining experience. However, today's AI is not sufficiently able to do so yet. Suppose AI acquires certain intelligence. When AI receives a new massive amount of data and acquires more intelligence, the previously acquired intelligence may not be retained and lost.
In order to develop continuously, AI needs to acquire new abilities while making use of what it has learned.

Fujitsu is also studying "social receptivity" for AI to be accepted in society.
One example is the efforts of Fujitsu Social Mathematics Joint Research Unit with Kyushu University. The Research Unit is a place for interdisciplinary academic research activities that combine areas such as economics, psychology and mathematics. As AI and ICT permeate into society, the Research Unit studies their impact on ethics or human's psychology and designs social systems and policies that will optimize AI to support our lives and society.

Zinrai, the product of Fujitsu's extensive technical expertise and experience

Figure 4: Human-centric AI Zinrai

In November 2015, Fujitsu announced human-centric AI system Zinrai, the industry's first product of extensive technical expertise in the AI field.
Zinrai features the following core technologies: Affective Media Processing Technology, which uses the five human sensory inputs to understand and interpret human emotion; Knowledge Processing Technology, which generates knowledge that can be processed by machines; Mathematical Technology, which solve key challenges even utilizing supercomputers; and Learning Technology to support the above three technologies. Zinrai's Learning System enables it to grow and develop by extracting valuable knowledge from continuing interactions. In addition, we provide Zinrai technology on FUJITSU Digital Business Platform MetaArc as a service to enhance the value we offer our customers.

Collaboration between humans and AI creates a prosperous future

In the future, AI will be used in various areas of our lives, such as bringing comfort to humans through communication, helping advance medicine and creating new business opportunities.
Fujitsu will continue researching human-centric AI systems that continuously develop and are designed for coexisting and collaborating with humans to bring a better tomorrow.