Can we teach robots to think like humans? How about teaching them to feel human emotions and even dream? These are all subplots within the seemingly unreal potential that artificial intelligence (AI) research holds for the future.
Perhaps we can accomplish those goals one day, but likely not soon. In the meantime, one Montreal startup is teaming up with McGill University’s Reasoning and Learning Lab to attack another goal: teaching machines to understand common sense.
Deep-learning startup Maluuba announced the new partnership. It wants to help research a complex and challenging aspect of natural language understanding. Generally, Maluuba’s mission is to help machines “think, reason and communicate with human-like intelligence.”
Maluuba will collaborate with McGill assistant professor Dr. Jackie CK Cheung, whose research seeks to develop computational methods for understanding text and speech. His goal is to enable a machine to generate language that is fluent and appropriate within a given context.
“Common sense is actually potentially the hardest thing we could teach a robot,” Maluuba cofounder Kaheer Suleman told MTLinTECH. “What we call common sense is stuff we learn implicitly. We know when you throw a ball in the air it’s going to come down, without being explicitly taught. The problem is machines don’t have any ability right now to experience anything since they don’t interact with the world. We have the teach machines these things.”
Other popular areas of AI like image and speech recognition are easier for machines, said Suleman. We can easily define those as teachable tracts. Not so with common sense.
Teaching machines to become literate is challenging due to the many subtle nuances of language. People learn language through exposure and experience over time.
The startup urged us to consider the statement, “The bee landed on the flower because it had pollen.” Humans know that “it” in this context refers to the flower. Common sense, while natural for people, is very difficult for machines because they don’t have a groundedness in the world around us, nor do they possess experiences living and interacting in that environment.
“Human-machine interaction using natural language is moving from the realm of science fiction to becoming reality. We can interact with chatbots or even talk to devices in our homes. Conversational interactions with current systems have been constrained, as it can be very difficult for machines to understand the subtleties of language and the broad range of words, terms and phrases that humans use,” said Dr. Cheung. “Partnering with Maluuba provides our students with the data and resources needed. Most importantly, we’re able to demonstrate how our research can be put into practice to drive even more AI innovation.”
Since opening its deep learning lab in Montreal in January, Maluuba has been building what it calls “the largest deep learning lab focused on natural language understanding in North America.”
The new collaboration with McGill was recently awarded an NSERC Engage grant from the Canadian Government, meant to support R&D collaborations between university researchers and industrial partners. They’re expected to publish an academic paper in Spring 2017.
Maluuba also collaborates with the Google-funded Montreal Institute for Learning Algorithms lab (MILA) led by Yoshua Bengio at Université de Montréal.
On Google and Microsoft’s recent investments in MILA and Montreal AI heavyweight Element AI (also cofounded by Bengio), Suleman commented that the city is “really becoming the hub of AI.” He credits Bengio and the Université de Montreal as those who paved the way for the city’s renaissance.
Suleman and Maluuba actually moved to Montreal from Waterloo specifically so they could operate their company in this environment.
“At the time nobody was thinking of Montreal. But U de M already had 100 students in AI and we knew Montreal would be the place for deep learning and AI. Now it’s starting to take off,” he said.
Maluuba could have settled in Toronto, but “after Google sort of came in and raided U of T, there was nobody left. It made more sense for us to go to Montreal.”