Global Site
Breadcrumb navigation
Responsible AI coexisting with humans in the AI society: Why is white-box AI drawing more attention?
What is reliable AI that NEC aims for?
NEC works toward providing reliable AI (e.g., white-box AI, which makes reason for its prediction result explainable), for which we take account of quality and respect for human rights such as privacy. Our reliable AI supports customers who use responsible AI.
See also: An unavoidable issue to know in using AI: how to consider AI's responsibility
White-box AI
Guidelines to Quality Assurance for Machine Learning-based Artificial Intelligence
Why is white-box AI drawing more attention? What made it make such a judgment or classification?
AI utilization may be at a loss on the way to proceeding in full swing and enlarging the range.
One of the difficulties is unknown reason why AI made such a judgment or analysis.
You can find an example in deep learning, which is now becoming popular. It has improved efficiency in image-based product inspection or other operations by leveraging the feature of automatically finding out characteristics from enormous amount of data. However, deep learning is not good at clarifying reason for the result, making them a "black box." This is a question in some operations.
That is why people are paying attention to explainable AI, also known as XAI or white-box AI, which gives reason for the result of processing.
Issues of black-box AI
Unknown improvements prevent us from increasing the accuracy. The lack of feeling persuaded or confident prevents us from making decisions. We are forced to lack accountability.
White-box AI has interpretability which is required to perceive how the AI brought the result of processing. This interpretability is necessary for decision-makers in business management, employment, financial screening, or healthcare; and for cases in which the result of AI's judgement affects the user.
Choosing an AI type based on the purpose: for better efficiency or suggestion
Black-box AI is strong in areas where improving efficiency (e.g., performance or speed) needs to be prioritized with clear AI tasks--such as product inspection (as above), real-or-fake determination, and deterioration diagnosis.
On the other hand, white-box AI is suitable for areas where reason for analysis needs to be explained for an issue with more than one possible goal. For example, white-box AI has been much used for demand forecasting in the retailing and manufacturing industry.
With the diversification of consumer needs and a shorter product life cycle, demand forecasting requires more advanced analysis know-how. In addition, there is only a little time to acquire the analysis know-how for quick decision-making based on the market changing from moment to moment. White-box AI allows you to combine its prediction result with human knowledge and then increase the accuracy of prediction analysis. Moreover, white-box AI's interpretability makes the user feel persuaded of the analysis result, which leads to smooth consensus-building and decision-making.
Thus, white-box AI is good at cooperating with humans to offer better hints. Based on the result and its reason, you can have consideration, make decisions, improve the accuracy, discover something new, and receive suggestions.
NEC, leveraging black-box AI, was quick to start researching and developing white-box AI, commercializing it in 2012. We had as many as 360 business talks about white-box AI, 120 of which led to its verification and operation, only from fiscal 2017 to the first half of fiscal 2018.
NEC offers cutting-edge AI technologies, including black-box AI such as RAPID machine learning, and white-box AI such as heterogeneous mixture learning and invariant analysis.
White-box AI being adopted in society
White-box AI has already operated in various scenes of society:
Chubu Electric Power Co., Inc.
For detecting signs of failure
The company uses white-box AI for early anomaly detection, performance management, and anomaly diagnosis.
NEC Solution Innovators, Ltd.
The company uses white-box AI to make the health guidance more persuasive, and to prompt the employees to improve their lifestyle.
NEC's AI technologies uses white-box AI
Our AI technologies below allow analysis with interpretability which gives clear reason for what process made the prediction:
Causal Analysis
Visualizes a cause-and-effect relation in data.
Invariant Analysis
Visualizes correlation among sensors to early find unusual behavior.
Heterogeneous Mixture Learning
Automatically detects multiple rules in various data.
Download