We aim to make AI that actually "thinks" like a human.
Here we introduce the AI research and development currently underway at NTT.
Artificial Intelligence (AI) has been rapidly pervading society through technological innovations such as deep learning, and it will soon be solving everyday problems of various kinds and scales without human awareness or understanding. The NTT Group has been promoting the introduction of AI technologies into society in collaboration with its partners under the unified brand of corevo®, which was founded in 2016.
However, there have been fears that the use of AI could lead to unintended discrimination and unjustified restrictions or inducements to action. In addition, the long-term consequences and impact of AI behaviors are still uncertain, thus raising both expectations and anxiety. In order to answer these concerns and to embed AI more deeply and successfully into society, the NTT Group, as well as its employees and engineers (hereafter 'we'), need basic principles in our approach to AI, which we should constantly be aware of and keep in mind as we get more involved in the use and R&D of AI.
Accordingly, we put forward six principles for dealing with AI and explain our intentions behind them.
We would like to express our sincere gratitude to the following people for their valuable input and support during the three expert meetings held in 2020:
|1. Enabling Sustainable Development||Commentary|
|AI is a beneficial technology that enables holistic development of lives and society. We ensure that our active R&D and our fair social implementation of AI will continue to benefit people, society and the planet into the future, with full respect for the norms of various countries, regions and communities. Establishing AI technologies that enable people to pursue sustainable development fulfills our mission, from both a public and a corporate viewpoint.|
|2. Human Autonomy||Commentary|
|AI is not just a tool for people to solve social issues but is evolving into an entity that brings about changes in people’s behavior, living environment, and even the consciousness of individuals and society. Human autonomy should always be respected when such AI is designed and adopted. We will continue to extend our knowledge of AI and our understanding of its characteristics so that we can design and apply AI appropriately in accordance with social norms such as human rights and diversity.|
|3. Ensuring Fairness and Openness||Commentary|
|The operational results of AI systems may contain bias. We will ensure that fairness is always maintained by understanding the characteristics of data and algorithms in AI so that users will not feel compelled to make unfair decisions. The mechanisms of AI will become more and more complicated. We will ensure openness by presenting AI’s operational scope and limitations as well as by improving its transparency and explainability while always listening to public voices.|
|We will build AI systems to be secure throughout their operational lifetime and respond quickly in cases of error or abuse. Automated AI systems will bring more benefits but even more risks. We will ensure their safety by understanding the details of their operations and by preparing for anticipated risks.|
|We will ensure privacy in AI to improve its trustworthiness by observing the secrecy of communications and by establishing an appropriate flow of personal data for AI.|
|6. Communication & Co-Creation with Society||Commentary|
|AI and its surrounding environments are changing dynamically. We will constantly share information and communicate with society. We need to collaborate with stakeholders in conducting R&D and creating new AI businesses. Through communication and co-creation with society, we are contributing to the development and spread of an AI that can be trusted by everyone.|
Through its appropriate use, AI is a beneficial technology that permits a holistic development of lives and society into the future. It is important that all people around the world can equally enjoy the benefits of AI.
We have been developing Information and Communication Technology (ICT) infrastructure on a global scale based on our mission from both a public and a corporate viewpoint. Particularly in Japan, we have been responsible for delivering ICT infrastructure nationwide as universal services. We would like to continue this approach as well in the field of AI.
We need to accomplish two tasks so that all people throughout the world enjoy the benefits of AI: the first is to promote active R&D as well as fair and equitable social implementation while respecting the norms of various countries, regions and communities; the second task is to continuously provide AI throughout the world, not just to people today but also to future generations.
AI is not just a tool for people to solve social issues but is evolving into an entity that brings about changes in people’s behavior, living environment, and even the consciousness of individuals and society. Countless AIs will be networked and used to solve a variety of large and small problems without users being aware of them. AI is a technology that is beneficial to people and society, but at the same time, it has new aspects that are not found in conventional technologies. For example, AI may be able to propose highly sophisticated solutions to complex problems that are far beyond the imaginations of humans and that humans cannot even comprehend. This means that humans may no longer be able to judge the validity of conclusions made by AI. It is also possible that human users may not always realize when and how AI is being used.
Considering these aspects, we foresee that AI will go beyond being a mere means and tool for humans, and that it will permeate society and bring about changes in the consciousness of individuals and society or engage humans in confrontations. Even under such circumstances, human autonomy should be respected in how AI is designed and utilized.
In order to achieve this, it is imperative that we, as researchers and developers providing AI to the world, share an awareness and responsibility as experts, and that we continuously extend our knowledge of AI and our understanding of its characteristics, including its potential and limitations. At the same time, it is essential for us to promote the use of AI in accordance with the social norms of the times, such as human rights and diversity.
The operational results of AI research may contain bias depending on the input data and the nature of the algorithms used. If the data input to AI is based on human activities in the real world, it may reflect customs and prejudices inherent in the real world, which may directly or indirectly influence AI output. In addition, depending on the nature of the algorithms used in AI, such effects may vary in different ways. With the ideal of always maintaining fairness when using AI, we will understand the characteristics of data and algorithms so that the output of AI does not misguide users toward making unfair decisions.
However, it is difficult to fully explain the reasoning behind AI output. This is due to both the quantitative aspect in which the amount of data processed by AI is so huge that humans cannot handle it and, on the other hand, the qualitative aspects such as the sophistication of algorithms, the black-box nature of processing, and the complexity of system implementation. AI is not omnipotent but works within a limited range of operation and accuracy. Even if the operational mechanisms of AI become more complicated, we will ensure openness by presenting its operational scope and limitations, by improving its transparency and explainability and by listening to public voices at large so that AI will be widely accepted in society.
AI has been pervading society and is now essential to the pursuit of sustainable development. We believe that AI systems should be operated securely at all times, and we will work to achieve this through R&D as well as business operations. On the other hand, even if we have built a perfect AI system at the time, it is not easy to keep AI running securely all the time. We will ensure AI security by responding quickly in case of error or abuse.
Furthermore, the impact of AI behavior varies depending on how it is utilized. When AI's operational results are directly used in automated systems, the benefits are assumed to become greater but so are the risks. In particular, when AI is used in a cyber-physical system through actuators, etc., it may affect human safety. We will ensure the safety of those systems, preparing for worst-case scenarios by fully understanding the behaviors and risks of AI.
When solving social issues with AI, the use of data containing personal information is indispensable. Processing personal data in a fair and transparent way with appropriate security is thus critical in the sound operation of such AI. Since AI is networked, we will ensure privacy in AI by observing the secrecy of communications and by establishing an appropriate and secure flow of data in communication networks while improving the trustworthiness of this flow.
The application domains of AI have been expanding steadily. AI is becoming ubiquitous and fully networked without users being aware of it. There may be completely new and unexpected use cases of AI anytime in the very near future, whether good or bad for society or for the global environment. Since AI and its associated environments are changing dynamically, we need to share information and engage in constant communication with society. We are determined to inform people about the beneficial uses of AI while also providing clear cautionary information about its potentially improper uses.
Finally, we declare that collaboration with our partners is essential when conducting R&D and business development of AI, which is evolving and diversifying day by day. We would like to contribute to the spread and development of the kind of AI that can be trusted through dialogue and co-creation with society.