After Winning at Chess, This Computer May Help Decide Loans

Human v Computer Japanese Chess Matches 2015 - Fourth Round

Professional shogi, or Japanese chess player, versus a computer.

Photographer: The Asahi Shimbun via Getty Images
  • Heroz programmers developed system that beat shogi expert
  • Adapting computers for consumer credit, financial markets

After defeating an expert in Japanese chess, this computer program’s next task is to figure out if you can make the payments on a new mortgage.

Heroz Inc., a small Japanese startup whose engineers designed computer systems that were the first to defeat an active professional player of shogi, the Japanese version of chess, says it’s now working on adapting its applications for the financial industry.

While Deep Blue became the first computer to beat a top chess player in 1997, the computer programming required to defeat a shogi master is even more complex than that required for conventional chess. That’s mainly because the number of potential moves in Japanese chess is much higher, as shogi’s rules on the reuse of captured pieces are more complicated.

Heroz is hoping that the lessons it has learned about how to recreate human judgment to help its computers win at shogi can be applied to the crunching of data for banks when they decide whether customers are creditworthy, according to Heroz’s Chief Financial Officer Daisuke Asahara. “There are times that computers can see as correct what humans perceive to be wrong,” Asahara, a former investment banker with Goldman Sachs Group Inc., said in an interview last month.

He said tests show the Heroz computers can successfully crunch the huge amount of data on consumers’ deposit and withdrawal information, and from social networks, to assist banks when they are making decisions on giving out loans. The firm is also working with Nomura Holdings Inc. to research whether its technology can be applied to financial-market forecasting, Asahara said.

Deep Blue

The Deep Blue computer developed by International Business Machines Corp. famously became the first computer to beat a top chess player, when it defeated then world champion Garry Kasparov in 1997. The added complexity of shogi meant it took another 16 years before Heroz computer’s program was able to outsmart an active shogi professional, with its 2013 victory over Shinichi Sato.

Such advanced computers are able to apply a sort of judgment to which of the many trillions of potential chess moves they should make, through a machine-learning process based on programming and past game data, according to Ho-fung Leung, a professor at The Chinese University of Hong Kong who specializes in artificial intelligence. The process gets around the need to laboriously examine every possible chess move, according to Leung.

“The approach they’re taking is very different from the conventional tree-of-possibilities approach to computer chess,” Leung said.

Shogi’s Complexity

There are about 10 to the power of 120 possible moves in a game of conventional chess, but that is dwarfed by the 10 to the power of 220 possible moves -- the number one followed by 220 zeros -- in a game of shogi, according to Heroz’s website.

Asahara wouldn’t disclose the names of the banks that Heroz is working with in analyzing consumer credit data. But Japan’s Financial Services Agency is trying to get banks to become more active in the field, saying in December that it will recommend legal changes to streamline the use of advanced technology in finance.

Japan’s largest lenders including Mizuho Financial Group Inc. are already developing services that make use of artificial-intelligence technology such as IBM’s Watson computer, mainly in call-center automation, according to Mizuho spokesman Masako Shiono.

Heroz was founded in 2009 by Takahiro Hayashi and Tomohiro Takahashi, both former employees of Japan’s NEC Corp. The 70-employee company develops and markets shogi and chess games for smartphones, which are the source of data for development of its machine learning techniques.

Its move into financial analysis will be assisted by a 100 million yen ($837,000) sale of Heroz equity to Hifumi Incubation Fund on Dec. 28, Asahara said. He said Heroz will use the funding from Hifumi -- a Japanese startup fund -- to invest in facilities such as cloud servers and to hire more computer engineers.