Fujitsu issued a press release on their AI in November 2015. The release said that they had been conducting research on AI for more than 30 years. They packaged these results as Zinrai. This set of slides gives more information on Fujitsu’s AI work and Zinrai.
The Zinrai architecture below covers many aspects of different AI areas. The three major areas are sensing and recognition, knowledge processing, and decision making and support based on AI technologies like learning.
The Zinrai architecture (Source: Fujitsu)
Three major areas
The following is based on the press release in November 2015.
1. Sensing and recognition
“Using human five senses, processes human emotions, and caring examples include scam preventions and improving customer relation services.”
2. Knowledge processing
“Technologies to create knowledge for machine processing as well as one for humans. Examples include medical diagnostics and overseeing financial regulations.”
3. Decision making and support
“Technological solutions to social and business problems mathematically with supercomputers. Examples include crowd control and a simulation of flooding by Tsunami.”
Accomplishments and business model
Participation in the program that tries to see whether a robot with AI can be admitted at the University of Tokyo. This is known as The impact of AI—can a robot get into the University of Tokyo (Todai)? project. Note that top universities like the U. of Tokyo are notoriously hard to be accepted by. Professor Noriko Arai of the National Institute of Infomatics is project director of this program. For your information, Dr. Arai did not go to Todai.
Fujitsu also experimented with character recognition using handwritten Chinese characters and reported a 96.7% recognition rate.
The business is promoted by its AI Application Consulting Department.
“Fujitsu will provide this technology in the form of a consulting service that utilizes AI to support dedicated consultants who will be able to use it to bring about business transformation and innovation.”
This is similar to NEC’s model, which utilizes system integrators. It appears that without help from some experts, it is hard to develop a system to incorporate AI technologies into each customer’s unique environment.
Its uses for now:
“[A] new service being developed that uses Zinrai’s machine-learning technology is FUJITSU Business Application Operational-Data Management & Analytics (ODMA), a solution that uses big data. Zinrai technology will also be made available as a service using the FUJITSU Digital Business Platform MetaArc.”
“… incorporates cutting-edge cloud, mobile, big data, and Internet of Things (IoT) technologies.”
Operational Data Management & Analytics
“ODMAuses big data, already implements the machine-learning parts of Zinrai. ODMA Predictive Supervision, which is currently under development, will accurately predict abnormalities in hardware and services. With this service, factories that use IoT technology will be able to take advantage of autonomous equipment maintenance to further improve operational continuity, supporting more efficient business overall. ODMA will be providing solutions that implement Zinrai on an ongoing basis, and Fujitsu plans to improve its products in order to expand the scope of AI applications.”
Fujitsu expects a total of JPY 50 billion for all Zinrai-related solutions by fiscal 2018. JPY50B is roughly US$500M. This is a good contrast to what IBM expects. IBM so far has spent $1B and expects to earn $100B in the future.
The information on Zinrai is very limited except for the press release.
From my skim through the Zinrai architecture, it appears that the architecture includes all the necessary AI components. As a native Japanese speaker, I am very interested in their natural language processing (NLP). I wonder if Zinrai can handle both Japanese and other languages, like English.
NLP is hard because natural languages do not follow rigorous rules. In a recent meetup, Niyati Parameswaran, a data scientist at the IBM Watson group, said that “language is ambiguous,” “ambiguity is ubiquitous,” and “ambiguity is explosive.” (originally mentioned by Professor Raymond Mooney.) You may think NLP in western languages, including English, is hard to implement. But think of two-byte languages like Japanese and Chinese. Japanese grammar and vocabularies are drastically different from western languages, and NLP developed for western languages may not be easily applied.
I am very much interested in finding out more on their NLP and that of other companies like NEC, Fujitsu and NTT. More on this issue in later blogs.