"Explainable AI" in the digital era
The driving force behind the explosion of AI-based solutions is deep learning, one of the core machine learning algorithms. There is, however, a fundamental problem with this rapid AI adoption – the black box nature of AI. This introduces major challenges with regard to applying AI to business decisions where reliability and responsibility are paramount considerations, because with deep learning there is no transparent basis for the actual decision-making. Fujitsu Research has been actively addressing this issue, developing the world's first "explainable AI" so that the reasoning behind any decisions is easy for humans to understand. With a clear explanation and verification of how AI has reached a decision, we can eliminate concerns, promote confidence in the AI process and achieve a level of AI that is genuinely trustworthy. Hence, we are working to ensure that as the use of AI becomes ever more ubiquitous, it is underpinned by a genuinely social and responsible role.
An example of Explainable AI’s value includes genomic medicine such as cancer treatment. In this instance, Fujitsu Research' proprietary AI technology, the Deep Tensor, learns from 350,000 pieces of data in order to estimate the relationship between a disease and a specific genetic mutation. In conjunction with the Knowledge Graph, which is a structured collection of more than 10 billion specialized knowledge from approximately 30 million medical articles, this can provide the medical basis for doctors to conduct much more rapid genetic analysis and tailor drug treatment to individual genetic abnormalities – reducing timescales from two weeks to just a single day.
Expanding the use of AI with autonomous machine learning
The potential of AI is enormous and growing all the time, with applications ranging from image and voice recognition to business decision-support. In addition, AI tools such as open source are widely available, and we are rapidly approaching a time of AI democratization, in which everyone can use AI in the workplace, without the need for dedicated expertise.
Another important technology that is contributing to the widespread adoption of AI is “wide learning", the world's first AI technology developed by Fujitsu Research, resulting from the evolution of AI technology in a different domain from deep learning. Wide learning brings an important new dimension for highly accurate decision-making, even when there is insufficient training data and deep learning is not well-suited.
Trustworthy AI for a better society
The behavior of AI changes depending on the data used for learning. If the training data is biased or erroneous, it becomes extremely difficult to maintain the reliability of AI decisions, as well as potentially causing ethical problems.
Fujitsu is firmly committed to pursuing the highest standards of ethical operation, publishing the "Fujitsu Group AI Commitment" in March 2019, and detailing the value-based items to be observed, including AI ethics. In Europe, Fujitsu Research of Europe became a founding partner of AI4People", an expert group that is working to create an AI Ethical Framework and recommendations around the Governance of AI, initially for adoption across Europe. In addition to advancing AI technology itself, we are actively engaged in activities to promote the acceptance of AI in various aspects of society, such as AI ethics and fairness.
Based on the concept of building a human centric society, Fujitsu Research is working to realize "AI that can be trusted by people and develop society" with the underlying objective of empowering people’s lives and enhancing their abilities.