Microsoft has been a leader in the technology industry for some time, even creating systems that can recognize words in a conversation and identify images with similar accuracy to that of human beings. Now, the company is taking the next step, aiming to create AI solutions that can review written passages and answer questions relating to content. The goal is to build a literate machine that can understand text-based information and communicate with people about what is discovered.
Microsoft recently acquired Maluuba, a deep learning startup company based in Quebec. They are one of the several teams working on the challenges associated with machine reading. Other groups located in Redmond, Washington; and Beijing, China, are also organizing a competition with help from Stanford University that intends to use data on Wikipedia to test responses from AI systems about the content.
At this time, the SQuAD data set is considered the benchmark for machine reading, and the hope is the competition helps improve the quality of AI answers. Microsoft’s MS MARCO, a system that uses data from Bing to test the accuracy of AI responses, is also in play.
Tests are being conducted based on real-world questions, a point the teams say makes the venture especially challenging.
Where the Difficulty Lies
Many experts in the field of AI suggest machine reading is particularly difficult to accomplish because there is an increased level of ambiguity than with prior projects, like image recognition. Instead of being focused on perception, such as the ability to view an object and identify it based on previous experience, machine reading uses cognition. The system has to know more than just the definition of words as it must also review context and associate it to its knowledge on the given subject.
For example, words often carry with them numerous potential meanings, and the AI must determine which option is correct based on the entirety of the question. Similarly, multiple words may be used to describe the same thing, adding a level of complexity.
To make matters more complicated, the AI may be tasked with drawing conclusions based on available information, even when the words in the question aren’t present within the text. This requires the systems to infer possible answers once the question is understood.
An End to Traditional Search Engines
Machine reading will have a significant impact on how search engines operate. If a person types a question into the search box, AI with machine reading capabilities could simply formulate an answer instead of just presenting a list of potentially relevant links. This would go beyond what is available today as only the most common, basic questions asked within a search engine can access such a response.
If you are interested in learning more about what’s new with Microsoft or need resources with Microsoft product expertise, the professionals at Solving IT can assist. Contact us to discuss your needs and let one of our skilled team members guide you through your options.