A Canadian company has come up with an algorithm that can read texts, and then accurately answer questions about them. The software is meant to help people by scanning and responding to their questions about boring technical texts — but there could be so many other great ways to use it.
The company, Maluuba, is interested in “building systems that replicate how human beings learn to read, understand, and reason using state-of-the-art deep learning techniques”. If that summary leaves you with some questions, Maluuba’s software might be just the thing to get the answers.
The software is meant to read through the long, boring texts that we don’t want read — like the instructions for a new household appliance. We’d love a computer program that could read boring texts and answer our questions and it’s possible that one day the software could read your notes and schedules, understand them, and make suggestions to keep your prepared for the next day.
The software is fairly proficient at reading stories as well. Specifically, the company had their Machine Comprehension System read Harry Potter and the Philosopher’s Stone, and answer some basic questions about it. The machine managed to answer the questions with 70 per cent accuracy. This video shows the testing process:
The machine has to integrate information from different sentences. It also has to discount certain sentences, such as those describing the scene where the sorting hat is deciding whether to put Harry into Slytherin or Gryffindor, because the narrative twists and moves on. Granted, the computer is only completing a multiple choice test, but it’s a good start.
But there’s so much more that could be done. We’d like a computer that could read its own manual and fix itself. How about computer that could scan the comments section of any website and alert you as to which commenters you could avoid and/or fight with, depending on your mood? And what about something that could read ahead and blank out spoilers for films we haven’t seen yet? This could be the start of a glorious age of non-reading. [MIT Technology Review, Maluuba ]