Microsoft Created AI That Reads Just as Well as Humans

It's no secret that Microsoft wants to be a leader in artificial intelligence (AI). Since forming its AI and Research group in 2016, the company has shown over the last few years that it's serious about becoming a leader in the space. With their newest achievement, Redmond is wasting no time staking their claim.

While this latest endeavor isn't quite as cool as building software that can conquer Ms. Pac-Man, the company has accomplished another AI feat. This month, it announced that its Microsoft Research Asia team has written AI software that can read and answer questions about a document with the same proficiency of a human reader. The software, which is already being used in its Bing search engine, has many applications for use in its products and services.

Questions and Answers

The research team used an established framework known as the Stanford Question Answering Dataset (SQuAD). The dataset is composed of more than 100,000 question-answer pairs on more than 500 different Wikipedia articles. For example, a machine learning (ML) model is given the text of an article for Super Bowl 50. It will then be asked questions such as "Who represented the AFC at Super Bowl 50?" or "In what city did Super Bowl 50 take place?" Questions get more advanced from there: "How many first downs did the Broncos have in Super Bowl 50?" "With how many teams has Manning won the Super Bowl?" The s oftware is scored based on its ability to answer these questions, and that score is compared to the performance of a human being.

On Jan. 3, Microsoft's team submitted a model that achieved a score of 82.650 on the exact match portion of the test, beating out the human score of 82.304. They currently sit near the top of the SQuAD leaderboard, sharing the No. 2 spot with a model created by the Chinese e-commerce giant Alibaba.

Smarter Software

This project is all part of an effort to shape the way we interact with technology. A good place to see these features play out is with Bing, where the company has already implemented some of the technology they submitted to the SQuAD leaderboard. Imagine, for example, the current way you use a search engine. Type in a question or search term ("Is coffee good for you?"), receive a list of links. With Bing's recently launched intelligent search features, however, you may very well find you could get the information you seek without clicking a single link. In this example, the search results are returned along with two paragraphs detailing the health effects of coffee. This, according to Microsoft, is a much more intuitive and "human" way of receiving information than the link-clicking mentality of the past.

"Microsoft is working on ways that a computer can answer not just an original question but also a follow-up," writes company blogger Allison Linn. "For example, let's say you asked a system, 'What year was the prime minister of Germany born?' You might want it to also understand you were still talking about the same thing when you asked the follow-up question, 'What city was she born in?' These investments are a conscious effort in making the use of a computer more natural and conversational."

AI Integration With Popular Tools

Microsoft has likely made major investments in ML, especially in reading comprehension, in part because it has use cases in so many of its products and services. Besides Bing or Cortana, it's not at all hard to imagine this document-reading functionality in the Office Suite.

Besides document reading, we are already beginning to see ML in other offerings. At its Ignite 2017 conference, Microsoft announced that it was ramping up AI integration into its biggest products. Dynamics 365, its popular customer relationship management (CRM) tool, will feature more AI enhancements and modular apps in the near future, including virtual agents for customers and virtual assistants for internal support. As an added bonus, these bots can be used in Facebook Messenger, Kik, Slack, and other popular chat apps. This integration aims to automate repetitive tasks and increase productivity. These tools are already being piloted with customers including Hewlett-Packard and Macy's and are expected to roll out to more businesses soon.

The entire Microsoft 365 ecosystem at large will see more AI integration as well. ML will be leveraged to bring users more relevant and personalized search results. Microsoft 365 is already a powerful tool and, if the AI integration works as intended, it could cement the company's place in the market even further.

Stiff Competition

While Microsoft has a stronghold in the AI space, they are far from alone. Just a few days after posting the story, their model was kicked from the No. 1 spot on the SQuAD leaderboard. It was usurped by a reader model submitted by a collaboration between the Harbin Institute of Technology (HIT) and AI company iFLYTEK in China. While being second is still a remarkable achievement, its demonstrative of how hard it truly is to be a leader in ML research.

AI hype (and the inevitable hangover) seems to be everywhere these days, and if Amazon and Google's presence at CES was any indication, the space will only get more competitive in 2018. Redmond will continue to innovate and lead advanced, useful AI implementations.

This article originally appeared on PCMag.com.