Artificial intelligence is one of the key technologies advancing digital transformation across a range of industries. And more and more, AI has attracted the attention of the financial services industry because of its potential applications. However, for all the results AI can deliver, the technology faces a problem: “blackboxing.” That is, it's often hard to tell why or how AI reached a certain result, even if the findings are accurate. In the following article, we describe how Fujitsu solves this problem with "explainable AI," a technology that can be applied to the financial services industry.
Fast-evolving AI for financial services
AI technologies have rapidly developed in recent years. In games, such as Shogi (Japanese chess) and Go, both of which are known for their complexity, AI can beat professional players, as today's AI-enabled machines continue to explode the myth that machines cannot beat humans at games.
As voice and image recognition technologies along with natural language processing grow more sophisticated, deep leaning is finding a place in a variety of business fields.
In the area of financial services, AI shows great potential for applications, such as risk management for investment and loan products, detection of money laundering and other illegal transactions and overall improvement of the banking customer experience.
The Challenge of 'Blackboxing'
While companies have started applying AI to their business processes, a challenge has emerged called "blackboxing." The problem is that usually no one, not even the developers of the technology themselves, can explain how or why AI reached a certain result or outcome.
This problem could hinder deployment of AI in business areas that require high credibility, such as the financial services industry.
There is growing demand in the industry for explainable AI that allows banks and service providers to logically account for the reasons behind an AI result, instead of simply presenting it without explanation.
Fujitsu's unique technologies for explainable AI
Fujitsu Laboratories tackled this challenge and recently announced a solution: Deep Tensor® connected with Knowledge Graph. The former is a unique technology based on machine learning, while the latter is a knowledge base that presents graph-structured data gleaned from documents and databases. Fujitsu has combined the two technologies help to solve the problem of black-box AI.
A new development for AI that Explains Reasons for the Results
Deep Tensor® : learning from graph-structured data to deliver highly accurate inferences
Fujitsu's AI technology Deep Tensor® has achieved highly accurate inferences in various fields based on a method of deep learning that analyzes graph-structured data, which is often used to illustrate relationships between things or people.
Deep Tensor® converts graph-structured data to a form of mathematical expression called a tensor and performs deep learning to achieve the highly accurate findings.
The technology is also able to run a reverse search of the deep-learning output to identify factors that had a significant impact on the results.
Knowledge Graph: a graph-structured knowledge base
Knowledge Graph is a dataset of knowledge collected from a variety of information sources. Connecting the inferences derived by Deep Tensor® to Knowledge Graph enables the system to show the reasons behind the AI-generated findings and to make them explainable.
Applying AI technology to credit risk assessments
Fujitsu's explainable AI technology has already been applied to financial services.
Accurate credit risk assessments for proposed corporate investments and loans is in high demand in the financial services industry. Traditionally, the task requires examination of a balance sheet and other performance data provided by a company. The industry was expecting AI to greatly improve the risk assessments and make the process more efficient.
But financial services providers find that some companies seeking loans or investments are not even able to provide balance sheets for their companies. In other cases, their balance sheets lack credibility, especially small to midsized companies. To accurately assess the companies' credit risk, financial services providers needed to collect other types of data.
Fujitsu developed a new approach to solve the problem, in which AI helps the service providers to assess credit risk based on bank-transaction data, instead of balance sheets.
In this case, Deep Tensor® analyzes such graph-structured data as money-transfer records to help assess credit risk. The technology identifies the factors that influenced the inferences drawn by the AI technology. It links these factors with related knowledge in the knowledge graph to make AI-generated findings explainable. The knowledge graph itself collects a wide range of company data, including information on parent and group companies, subsidiaries, stock holdings and board members.
Potential to address a wider range of financial services
Fujitsu has started applying the explainable AI technologies as a proof of concept to enhance credit risk assessments for investment and loan products.
Fujitsu expects that application of the technologies will grow in coming years to serve a wider range of tasks in the financial services industry, including credit risk assessment for retail finance, prevention of money laundering and marketing initiatives.