Even though Microsoft's digital assistant, Cortana, was mysteriously missing from CES, the Redmond giant has been reiterating that 2018 is the "year of AI". It recently announced partnerships with Fujitsu and Nordcloud to boost the spread of artificial intelligence, and also stated that the firm and Adaptive Biotechnologies will both use AI to decode the immune system.
Now, Microsoft has announced that it has developed an AI that read and answer questions about a document with human-level accuracy.
In a blog post, the company explained that its Microsoft Research Asia wing has developed an AI which has achieved human parity in the Stanford Question Answering Dataset (SQuAD), a dataset which consists of questions regarding Wikipedia articles. According to the firm, humans have a score of 82.304 on the ExactMatch test based on the metric. However, Microsoft's AI was able to slightly surpass that with a score of 82.650.
It is important to note that Alibaba also submitted a score of 82.440 two days after Microsoft, and both the firms are currently tied for first place on the SQuAD leaderboard.
However, Microsoft says that it is not resting on these laurels and will continue to improve AI's natural language processing capabilities to the extent that it can easily answer follow-up questions. The company went on to say that:
For example, Microsoft is working on ways that a computer can answer not just an original question but also a follow-up. 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?”
It’s also looking at ways that computers can generate natural answers when that requires information from several sentences. For example, if the computer is asked, “Is John Smith a U.S. citizen?,” that information may be based on a paragraph such as, “John Smith was born in Hawaii. That state is in the U.S.”
Microsoft's achievements in the field of natural language processing and AI will lead to relevant information being easily extracted from a dataset. Scenarios include lawyers searching for a rare legal precedent, doctors locating medical findings from hefty journals, and drivers finding the answer to a particular question in a complicated car instruction manual.
While the company says that there are still challenges to overcome in natural language processing, it is already integrating earlier versions of the AI model to its Bing search engine.