TOKYO -- Among the various products offered by Mitsubishi UFJ Kokusai Asset Management is AI Japan Equity Open, an investment trust launched at the beginning of February.
As of March 31 it had attracted total investor assets of 10.7 billion yen ($98 million).
Asked why it is selling so well, director Hideo Shirota said, "Many investors want their assets handled based on objective judgment rather than the professional sense of a fund manager."
The company receives advice about the selection and timing of equity trades from Mitsubishi UFJ Trust Banking, where chief fund manager Noriyuki Okamoto, a man of over 20 years experience, and moreover, an artificial intelligence system are the brains behind the product.
Arriving at the company's Tokyo head office at 8 every morning, he turns on his computer to read an AI-generated report on which equity prices will likely rise and the best timing to sell index futures. "The volume of the work is far beyond what humans can handle," he said.
The AI tool has four analytical functions. Two employ text mining, specifically analyzing text data. The AI looks at earnings summaries and securities reports from roughly 2,000 listed Japanese firms to choose equities that reliably provide high-yield dividends. To choose the best equities for short-term investment, it analyzes data from Bloomberg News and the profit predictions of stock analysts.
The other two employ a deep-learning method, simulating a human brain, to study the characteristics of a given data set. By finding relationships between the prices of stock futures and 300 relevant factors, including investor psychology, exchange rates and interest rates, the system offers daily reports on price trends, predicting both the coming day and a month ahead.
Read the world
It's important to note that the AI doesn't make all the decisions. "It's up to humans to spot the major turning points in the price trend," said Shirota. The big-picture decisions, he said, are down to us.
So Okamoto looks ahead for major shifts, employing chart analysis based on stock index trading data.
Humans cannot take in every detail of the information available on listed firms. AI can, but the kinds of information it sees are preset. It is like a marine radar, dedicated to monitoring the specific areas assigned to it.
The decisions that AI is incapable of taking are made, for example, by approaching the same problem from different angles or with different methods. People can look at AI-generated answers from viewpoints outside the realm of the program, such as what books they have read or opinions they have heard.
AI also has difficulty grasping general trends, such as how the world changes. When a person feels something is happening but cannot put their finger on what it is, they naturally try to figure it out. AI, by contrast, is incapable of having a purpose of its own; it can only do what it is told to do.
Keeping an eye on AI
The details of the Mitsubishi UFJ Trust Banking show it is designed for humans to handle up to 24% of total investor assets without regard for AI advice. This portion represents the value of the job assigned to humans from a division-of-labor standpoint.
That said, the group is considering development of AI investment products that exclude all human involvement. "We want that if we can get it, as it will cut costs," said Shirota. Judging from the rapid progress in the technology, it will be no surprise to see it happen soon. "At this point, however, an investment trust without human involvement is unthinkable, even with the best AI capabilities," he continued.
With the AI system that Nippon Life Insurance began using in March, people play the role of monitors.
The system provides real-time suggestions for insurance products to 50,000 salespeople meeting consumers. Based on the input information like family size and annual income, the system automatically advises customers, replacing the manual work that the product-development team formerly did behind the scenes.
Kazuya Tajima, who till March 31 was an executive in charge of product development, said the division's responsibility has changed to spotting and correcting erroneous instructions from the system. With humans monitoring the AI and applying the brakes as needed, a human-AI hybrid model is possible.
Rather than stand in awe of the brilliance of AI-assisted automation, corporations must understand what it is incapable of doing -- big-picture thinking, goal setting and leadership.
As the use of AI spreads, those areas of human responsibility will be in greater demand. Well-designed division of labor balancing respective strengths and weaknesses, will be of huge importance for corporate management.