Expanding from previous research into identifying contents in images and words in a conversation, the latest project has a greater understanding and can manage larger chunks of text. “We’re trying to develop what we call a literate machine: A machine that can read text, understand text and then learn how to communicate, whether it’s written or orally,” said Kaheer Suleman, the co-founder of Maluuba. Maluuba is a Canada-based startup that Microsoft acquired earlier in the year. Researchers working together are using Wikipedia entries to test AI systems. The machines can read and understand that text. To put the learning capabilities to the test, the researchers ask the system to answer questions on the read text. Microsoft created its own MS MARCO dataset that was released to other industry and academic researchers this year. The dataset uses anonymized data from Bing search queries to test a system’s ability to answer questions. “We’re not just going to build a bunch of algorithms to solve theoretical problems. We’re using them to solve real problems and testing them on real data,” said Rangan Majumder, a partner group program manager within Microsoft’s Bing division. Machine reading is among the most difficult tasks for artificial intelligence. Researchers say AI is more adept at image recognition. The cognitive nature of machine reading challenges AI to see a broader meaning behind words and to understand context. “Some words might mean different things, and the same things might be mentioned in different ways,” says Ming Zhou, assistant managing director of Microsoft Research Asia.
Asking Questions
The company says the best way of knowing if AI has understood something is to simply ask it questions. Mirroring how people use questions to gain knowledge, the team drew inspiration and sought to develop a new system. Researchers from Microsoft are now working on a system that can read text and form its own questions, and not just answer questions. “What’s interesting is you actually need a much deeper understanding of the text to generate a question rather than an answer,” Suleman said.
