Skip to main content
There has been a growing expectation over recent years for the application of artificial intelligence to solve business issues. Hitachi has developed an artificial intelligence system that understands the fluctuation of demand, on-site arrangements and other factors as it issues work instructions. The Company has incorporated the new technology into a logistics system and has verified its effectiveness, confirming an 8% reduction in the on-site work time. Here is an interview article with participating engineers, asking them what episodes or hardship they encountered regarding Hitachi AI Technology/H, a system that can improve by cooperating with humans.
(Publication: June 7, 2016)
HIRAYAMAIt is a program that uses work data and past results to understand demand fluctuation and on-site arrangements, and issue work instructions that are the most suitable for on-site operations. We have recently incorporated the Hitachi AI Technology/H into a logistics system and verified that it can effectively improve on-site operations.
AKITOMIConventionally, it was not easy to make changes to a work system once it was established. Even if on-site workers tried to reflect their ideas for improving their work onto the system, it was difficult to achieve because such a process had to be done through humans like system engineers.
However, once the Hitachi AI Technology/H is incorporated, it allows the work system to issue work instructions, which has accumulated ideas for betterment, to the on-site workers without modifying the work system itself. On the work floor, workers operate in accordance with the work instructions issued by the Hitachi AI Technology/H, or by making their own arrangements for improvement. Then, the work results are accumulated as actual performance data. Based on such data, the Hitachi AI Technology/H will be able to issue work instructions that reflect such new arrangements. In other words, the Hitachi AI Technology/H becomes smarter by cooperating with humans.
Figure 1: Work improvement cycle through the Hitachi AI Technology/H
AKITOMII agree. There are some interesting cases. For example, it is already known that machines are more adept at chess than humans. However, machines play chess the best when they are in collaboration with humans. Indeed, there are championship games in which machines and humans form a partnership to play chess. As this suggests, it has become a common understanding that artificial intelligence can exert its capabilities more powerfully when it works together with humans, rather than when humans let it do everything.
AKITOMIEssentially, I was engaged in the research of analyzing data and applying the results to work improvement. That is what was called "Business Microscope," and it was the starting point of the latest development.
In the research of Business Microscope, we collected a variety of data concerning behaviors by attaching name card sensors to people. Such data included whom they met during work and who nodded affirmatively to them while they were talking. The collected data was analyzed and utilized for solving on-site problems. However, the collected data had to be analyzed for each business type from scratch, and that was too cost inefficient to justify. To solve this problem, we wished to create a general-purpose model applicable to a variety of business types, not limited to certain business types. This was one of the factors that triggered our research.
HIRAYAMAOn the other hand, the frontline of the logistics section faced a challenge of enhancing the operational productivity to reduce personnel costs. People at the frontline discussed how to improve the situation and made various attempts, but they couldn't succeed. They sought to take advantage of performance data in the past. Such a need came to our attention just when we were about to start development of the Hitachi AI Technology/H, and it was decided to conduct on-site verification of the technology in the logistics system.
HIRAYAMATo tell you the truth, I was half doubtful if artificial intelligence can be realized in any form. Although artificial intelligence had long been studied around the world, I was not sure if the Company could commercialize the product with the name of "artificial intelligence" affixed to it. Still, I also thought it might be interesting.
AKITOMII thought that I got involved in a tough subject. Because the research was a subject that drew attention, I was under pressure to produce good results. It was such a challenging theme that I felt people were watching us with interest.
Figure 2: Functions of the Hitachi AI Technology/H
HIRAYAMAThere are three phases of processing before the Hitachi AI Technology/H issues work instructions to the frontline based on input data.
First, it automatically determines, to a certain degree, what types of data have been inputted to the work system (automatic analysis of data).
Next, it combines pieces of data into numerous combinations to create items to check the correlations among the data (generation of feature values).
Then, it narrows down the feature values that should lead to improved operations, and make hypotheses (formulation of equations).
In accordance with the equations created by the Hitachi AI Technology/H, the work system issues work instructions, which reflect ideas for improvement, to the workplace.
AKITOMIWe conducted verification at the work floor for item collection, or what is called "picking," at a warehouse. Picking is an operation to pick up ordered products from warehouse for delivery. When orders for products are received, the logistics system prepares written instructions, and workers collect items in accordance with the lists stated in the instructions.
Figure 3: Difference in congestion situations in a warehouse
However, because multiple people are involved in item collection at the warehouse, workers are concentrated at certain sections of the warehouse when orders are centered on products of certain shelves, and the congestion has caused the problem of lowered productivity.
Once the Hitachi AI Technology/H was incorporated into the logistics system, the Hitachi AI Technology/H made a hypothesis that, by changing the order of the written instructions, the congestion within the warehouse is controlled, which should lead to higher productivity.
So the logistics system changed the order of the written instructions in accordance with the hypothesis. And the mere fact that the workers followed the instructions to conduct item collection proved effective, as it reduced the work time for item collection by 8%.
HIRAYAMAThe most difficult part was the second phase of processing, or "generation of feature values." In this phase, we implemented the rules for how to generate feature values onto the Hitachi AI Technology/H, but determining the rules was quite tough. Our first idea was to generate the feature values by combining pieces of data to the best of our knowledge. However, when we loaded all the possible combinations of data, the result was the feature values contained many arithmetic operations. That might be good for the feature values themselves, but it was difficult to tell at a glance what their intent could be. To "decode" them, researchers had to have many discussions. It was not right, we thought, because it was like we were putting the cart before the horse.
Later, Mr. Akitomi took the lead in experimenting and through such endeavors we were able to narrow down the feature values that could be used more regularly and were applicable to a variety of business types. In the end, we arrived at what we have now.
AKITOMIIn order to narrow down the feature values for general purposes, we tried as much as possible to not to think of applying the system to a specific business type, like logistics. Rather, we forced ourselves to have a macroscopic view, hoping to obtain adequate phenomena from the feature values. Such a way of thinking is contrary to what researchers generally employ. Researchers will often think by being specific and digging into specific matters. However, we consciously attempted to think along a broad and shallow scope, looking for what to do to obtain phenomena that could be applied to a variety of business types.
HIRAYAMATo avoid that, we are engaged in the research of creating the feature values systematically by first conducting the processing of "automatic analysis of data." For example, the work system cannot determine whether the 9-digit number, "123456789," represents a volume or an ID. The Hitachi AI Technology/H checks about 20 attributes, such as data distribution and differences in phraseology, and analyzes whether the figure represents a volume or an ID. This function of automatically analyzing data makes it possible to generate necessary feature values without conducting useless calculations.
It is quite a new achievement that the data inputted into the work system is analyzed automatically to such an extent, although people might think it should already have been realized.
AKITOMIAt an exhibition where the company's research results are demonstrated to outside customers, the Hitachi AI Technology/H received the customers' highest acknowledgement in the Internet of Things (IoT). I think this suggests that an automatic solution for problems in conducting business is a subject that appeals to people of any company.
HIRAYAMAAt the exhibition, we were asked for consultations by many customers, who said "we currently face so-and-so problems" and "we hope to achieve so-and-so results with AI, but do you think it possible?" I really felt that the Hitachi AI Technology/H is needed by so many people.
AKITOMIAt present, the Hitachi AI Technology/H conducts automatic analysis of data through generation of equations by using very simple logic. Going forward, we are going to let the Hitachi AI Technology/H evolve steadily so that it can discover and ameliorate more complicated factors.
To do so, the key is in how to generate the feature values. I hope to make the Hitachi AI Technology/H capable of generating equations from the feature values that are more generally applicable, in spite of the underlying phenomena.
HIRAYAMAI wish to apply the Hitachi AI Technology/H to a variety of business types. We have achieved results with logistics, and we should be able to verify that it produces results for other business types like retail and plant construction. I want to demonstrate its functions both internally and externally as much as possible.
AKITOMIOriginally, my motive for joining Hitachi was to understand human behaviors. Currently, I'm studying artificial intelligence as my assignment. Eventually, however, I hope to further clarify human behaviors by using artificial intelligence. That's my pure interest as a researcher. It would be great if artificial intelligence is utilized to make it easier for people to work or to improve their overall happiness.
HIRAYAMAI like to study latest technologies, and I hope to continue pursuing the latest technological achievements.
Furthermore, I wish to make artificial intelligence succeed as a business. I hope people outside Hitachi learn that the company is working on such an enigmatic but intriguing subject as artificial intelligence. It would be my pleasure if it reaches people's attention that Hitachi is doing something interesting.