FinTech Firm’s AI System Nears Human-Level Reading Comprehension in Stanford Test

The FinTech subsidiary of a major Chinese insurance and financial services company has developed an AI system that can read text almost as well as a human can, according to recent test results.

The AI system, GammaLab, was developed by OneConnect, a subsidiary of Ping An Insurance. The test in question, meanwhile, is the Stanford Question Answering Dataset 2.0, or “SQuAD2.0”, which presents AI systems with over 100,000 questions based on text paragraphs. It’s a reading comprehension test, in other words, and it includes 50,000 questions that cannot be answered based on their associated paragraphs – the point being that AI systems also need to be able to determine when they don’t have sufficient information to answer a question.

In a statement, Ping An announced that GammaLab took the top ranking in the latest SQuAD2.0 challenge, with a score of 83.435, coming very close to the human average of 86.831, and going “way ahead of other companies in the challenge,” Ping An said.

This isn’t Ping An’s first foray into sophisticated AI: back in early 2016, the company announced that its internal Science and Technology Artificial Intelligence Laboratory had developed a facial recognition system that was 99 percent accurate. Now, with the success of its OneConnect subsidiary in the SQuAD2.0 test, it’s clear that the growing demand for AI-driven technologies in financial services and adjacent areas is helping to drive some serious FinTech innovation, even on a more or less in-house basis from more traditional financial services companies, and not just tech startups.