AI becomes a business infrastructure

April 24, 2019

AI becomes a business infrastructure

Yoshitaka Sakurai 
Associate Professor, School of Interdisciplinary Mathematical Sciences,
Meiji University

In recent years, the business application of AI (artificial intelligence), especially machine learning, has been progressing at a steady pace. However, Japanese companies are still lagging behind overseas companies in developing effective AI strategies. Associate Professor Sakurai talks about the keys to the utilization of AI and its future development.

Advancing AI-driven automation

Efforts to apply the effects of computers, machines and AI to business is nothing new. In the late 1970s for example, OA (office automation) created a big boom, and in the 1990s, IT (information technology) spread rapidly.

These technologies enable automation in the workplace and help to reduce the need for manpower, thus improving efficiency and reducing costs.

In recent years, big data, data scientists and AI have all gained attention, and the business application of these technologies in a variety of areas has been progressing steadily. Among such technologies, RPA (robotic process automation), which automates business processes that were once accomplished by people, has garnered increased attention, and more and more companies are considering the introduction of this technology in their business operations.

There have been cases of large-scale automation based on AI reported overseas. In the financial industry in particular, some companies have laid off hundreds of traders and have simply switched their operations to AI and a few engineers.

We are currently experiencing the “third AI boom.” A variety of factors have contributed to the expanding use of AI.

In the academic world, in 2012, a team led by Geoffrey Hinton achieved an overwhelming victory at the ILSVRC (ImageNet Large Scale Visual Recognition Challenge), a renowned object detection and image classification competition, by utilizing deep learning. Subsequently, machine learning, deep learning, and research on applications using machine and deep learning methods have progressed at a notable rate. Outside the academic world, IBM Watson defeated the reigning champions in the popular US quiz show, Jeopardy, in 2011, and AlphaGo, which was developed by Google, beat a professional Go player in 2015, both of which were widely covered by the media. These episodes were convincing enough for many people to believe that AI could learn enormous amounts of data at breakneck speed and surpass human capabilities.

In addition, many practical applications of these technologies have accelerated the current AI boom. The background behind this is the development of big data, the high-speed computation environment and tool library necessary for machine learning, which is a core technology of the current AI boom. In terms of human resources, substantial numbers of AI specialists – the number of which has been growing since the second AI boom – are also an important factor.

As described above, the current AI boom arrived after the environment to support such a boom had come into place, and it was apparently necessary. Just as the use of IT has become more common now, I think as we continue utilizing AI, it will eventually play a role in the infrastructure in our daily activities, instead of just creating a boom.

However, not everything has been smooth sailing. The business application of AI has come with a variety of challenges that I would like to highlight.
Challenges for the business application of AI

In the midst of the AI boom, more and more companies are considering introducing AI with the anticipation of further advancements in machine learning. However, there is so much to be dealt with, and many of the current challenges that we see arise from a lack of understanding of AI.

One such challenge is the environment in which AI is applied.

Machine learning requires data. However, for many companies even in-house data is owned by each department and there is no interdivisional cooperation. Because of this, these companies cannot even build a company-wide unified database.

Many overseas companies have engineers who are in charge of data and systems, and establish technical divisions assigned to secure the consistency of data management and systems investments. Importantly, they appoint the top engineer as both a CIO (Chief Information Officer) and a member of the management team.

This is in comparison to most Japanese companies, where the Information Systems Department Manager, the person responsible for the management of information systems, is probably the highest position, and the importance of managing and organizing data and systems on a company-wide level is not yet sufficiently recognized.

Whether AI is applied or not, data is still a corporate asset and systems provide a platform for corporate business activities. However, both data and systems are indispensable for utilizing AI.

Another challenge is how AI should be applied.

RPA, as mentioned earlier, replaces human tasks with AI. Recently, I have been hearing quite frequently that departments within some Japanese companies are considering adopting RPA to automate individual staff jobs in an effort to reduce costs and increase efficiency.

But there are some problems here: replacing operations, which are heavily dependent on personal skills, with AI will further create black boxes; and building a system by assembling separate parts without regard to the overall design may jeopardize system integrity.

Yet another challenge is how AI should be evaluated.

The effectiveness of AI’s machine learning does not necessarily immediately manifest itself in ways that are tangible as is common with OA and IT. Because of this, there are many cases that fall through before results are available. Data is especially important for machine learning. However, the initial set of data is insufficient for some systems, and data may be collected after the system is introduced to increase prediction and classification accuracy. Systems based on machine learning need to be assessed from an “investment” perspective.

The full-fledged introduction of AI will not function effectively by simply assembling a system of individual parts. The system requires a comprehensive design, including organizational reform of the company. This is one of the challenges that Japanese companies have to address most urgently.

“Decision support system” with AI as a competent secretary

Lastly, let me share the applicability of AI.

The applications of AI I have talked about earlier are limited to specific operations such as playing Go and image classification. In other words, by narrowing down the data AI is tasked to process and learn, specific operations can be tremendously efficient. However, narrowing down data is often not easy. Current AI technology is still far from realizing a solution for complete automation, prediction or classification with practical accuracy.

On the other hand, there is an AI study on a decision support system. This study does not aim for automation or system efficiency without human involvement. Rather, the study is about using AI to support human decision making. Humans and machines are different when it comes to what they are capable of. This is why the cooperation between them can achieve something even more sophisticated.

I am also involved in this study. In fact, this type of system is already being utilized around us.

For example, when you search the Internet for news, images, videos or products, information relating to your searches starts appearing as so-called “recommendations.” I think many of you have probably had this experience. This is called a recommender system.

AI compares the data on an individual who shows interest in a certain item with the data collected from all Internet users, calculates the rating the individual would give to the item, and predicts the user’s next selection.

In other words, the system produces “recommendations” with the help of AI in order to help the individual make a decision on the next selection.

Although this recommender system has been mainly used in marketing, it will soon be used in departments relating to human resources and corporate strategies.

ATS (applicant tracking system) for example, is a recruitment management system that handles all recruitment needs including: Recruitment, applicants, the evaluation of applicants, and the applicants hired.

The system filters this down even further to the departments to which hired individuals are assigned, job responsibilities, performance and assessment.

The system then tracks and analyzes consistent data created for each employee, which includes records of their application information, business activities and every department with which they have been associated.

ATS enables companies to easily manage human resources requirements by analyzing what types of employees are currently working, or to get a grasp on the types of human resources that are missing in their companies. This contributes to accurate and quick decision making concerning human resources.

This system has not yet become common in Japan, and our research team has been devoted to this study with a view to the introduction of a full-fledged system in collaboration with companies. But I believe it should not be too long before AI-driven personnel systems will be a common part of the business infrastructure.

* The information contained herein is current as of April 2019.
* The contents of articles on Meiji.net are based on the personal ideas and opinions of the author and do not indicate the official opinion of Meiji University.

Profile

Yoshitaka Sakurai
Associate Professor, School of Interdisciplinary Mathematical Sciences

Research fields:
Machine learning, data mining, evolutionary computation

Research themes:
Decision support system based on machine learning
[Keywords] Marketing research, recommender system, Kansei retrieval

Main books and papers:
◆“VICA, a Visual Counseling Agent for Emotional Distress” Journal of Ambient Intelligence and Humanized Computing, DOI 10.1007/s12652-019-01180-x, pp.1-13, Springer, January 2019
◆“An Efficient Language Pipeline for Flexible Rule-Based Context Representation” Journal of Ambient Intelligence and Humanized Computing, Vol.4, Issue 4, pp.439-450, Springer, August 2013
◆“A Retrieval Method Adaptively Reducing User’s Subjective Impression Gap” Multimedia Tools and Applications, Vol. 59, Issue 1, pp.25-40, Springer, July 2012
◆“Enriched Cyberspace through Adaptive Multimedia Utilization for Dependable Remote Collaboration” IEEE Transactions on Systems, Man and Cybernetics – Part A, Vol.42, Issue 5, pp.1026-1039, IEEE, September 2012
◆“Toward Sensor-Based Context Aware Systems” Sensors, Vol.12, No. 1, pp. 632-649, MDPI, January 2012

Page Top

Meiji University

Copyright © 2015 Meiji University. All Rights Reserved.