There are no Ping-Pong tables or beer kegs at Maluuba’s Peel street office. There’s no receptionist to enthusiastically tell you to wait. People aren’t dressed as walking Frank & Oak models and no one is sitting around in beanbag chairs sipping expensive coffee.
Instead it’s just people, mostly guys wearing hoodies, sitting in front of their screens in a quiet room. I even have to break someone from concentration to let them know I’m here to interview, which I admire.
My two subjects Wednesday afternoon were Kaheer Suleman, cofounder and Rahul Mehrotra, program manager. They’re with the Montreal-based Maluuba, a deep learning research lab that was acquired by Microsoft in January. They teach machines how to think, reason and communicate.
Mehrotra greets me in an extremely soft tone that makes one think he probably treats his elders nicely.
As we take the elevator down two floors to Maluuba’s extra space, the cofounder Suleman paces around in half-circles like some kind of savant, as if performing calculations in his head.
He also looks really tired.
“How much sleep did you get last night?” I ask.
“Not as much as I wish,” responds Suleman. But within a few minutes of fun conversation his face lights up as we chat about different concepts that the company is working on. His smile is strikingly genuine.
My first impression of the company is a humble, unassuming group of people. There doesn’t seem to be too much passive-aggressive bulls—. People are just working and being nice to each other.
“We always say that our goal has never been to chase money,” said Mehrotra. “It’s to build cool stuff and solve hard problems.”
This is the same company that started out as a school project while Suleman and cofounder Sam Pasupalak studied at the University of Waterloo. Suleman’s mom would bring them food everyday as the line blurred between fun project and real company.
Their origins are varied as well. Suleman grew up in Cambridge, Ontario, his mom from Uganda and his dad from Tanzania. Mom came in 1972 when Ugandan dictator-president Idi Amin ordered the expulsion of his country’s Asian minority. Dad studied in Uganda, but left for Australia and eventually Canada (Toronto) and the United States to study radiology.
Mehrotra is from New Delhi, India’s capital city, and grew up in lovely Dartmouth, Nova Scotia. He later went to school at Waterloo.
“Sam says, ‘humble beginnings,’” said Mehrotra. “And I think everyone here possesses that. I still remember what it was like to not have a lot.”
The guys say they’ve kept up this humble attitude even after being acquired by one of the world’s tech behemoths in Microsoft, and after having previously raised over $10 million in venture capital.
The company is staking out research that will likely change the way we interact with computers. It’s work that maybe one of two other groups in the entire world are working on within deep learning, including Google and possibly Facebook.
Right now, Maluuba wants to solve the problem of text understanding, building machines that can read and write. This isn’t just building a rudimentary chat bot, but addressing core, fundamental issues within text understanding. It’s building machines that can reason with text and effectively communicate with people.
“If you look at chatbots today their session memory is very small. Once you say something they maybe remember it for one turn. We’re working towards systems that can have much, much longer memory,” said Suleman.
Last year Maluuba achieved state of the art results within reading comprehension benchmarks and also released its own benchmark that targets more sophisticated reasoning in language understanding.” In April it taught leadership to robots.
The company isn’t working on anything that can be sold to people. They do research. But it’s this research that may eventually be seen in consumer products, maybe changing the way we live.
Unlike a startup building and releasing an AI product, Maluuba will spend significant time on its research and come up with technology that’s “significantly better than anything on the market,” according to Mehrotra.
“We don’t think the fundamental problems have been solved yet,” said Mehrotra.
The big acquisition made sense for Microsoft, given that the giant corporation is building several products that Maluuba’s technology fit into, containing text at their core. Microsoft’s enormous resources will also aid Maluuba in tackling large problems that confront AI research.
“Their vision and our vision were essentially perfectly aligned,” said Mehrotra.
The marriage seems to be working well so far. Now that they’ve been acquired, the guys said they no longer have to go out and raise money. They can just focus on solving the problems they want to solve.
From an application standpoint, Maluuba recently focused on building technology that read and understood news articles in real time on CNN.com or Techcrunch.com. It could reason and answer questions from people about the articles. This technology could theoretically be applied to any text-based product like emails, legal documents, product manuals and more.
Imagine the productivity it could lend to a paralegal scouring over documents, said the pair.
“We wanted to change that way we interacted with our computers,” said Suleman. “Even today, the standard point-and-lick interface of computers hasn’t changed in over 30 years. We wanted to go to the next level, like the interface we all saw in movies where you could just talk to your computers. And really, language was the biggest barrier.”
“Machines don’t understand language, which is why we have to program everything. We have to turn the human language into the machine language. We need to bring the machine into our world.”
One of their advisors since 2015 is widely viewed as the godfather of machine learning in Dr. Yoshua Bengio, a professor-researcher at the Universite de Montreal. When Maluuba first opened its Montreal office, it was in partnership with Bengio’s Montreal Institute for Learning Algorithms (MILA).
In fact, it was Montreal’s leading community of AI researchers that first attracted Suleman and Pasupalak to move from Kitchener-Waterloo to Montreal.
Just a year after Maluuba moved here, Google opened a research office focused on AI, and invested millions in academic efforts like MILA. Then Microsoft invested $7 million into local AI research efforts just days after it acquired Maluuba.
“We 100 per cent moved here because of Montreal’s growing AI scene,” said Mehrotra.
Suleman added that at the time they were preparing to switch their research from machine learning to deep learning “to solve some of the bigger problems we needed to solve.”
They asked themselves where they could set up the ideal research lab.
“Toronto’s scene was kind of shrinking since [Dr. Geoffrey] Hinton had just moved to Google and several others were going on sabbatical. But MILA was growing,” said Suleman.
Today, MILA has over 150 researchers, substantial government funding and money from the tech giants.
“We believed Montreal was going to be the next center for AI and we really wanted to be part of that. Toronto would have been a way easier choice, but with the amount of untapped talent that was in Montreal, we had to,” said Suleman.
Just 24 months later, the guys sit in their ninth floor office overlooking Peel street and Montreal’s downtown, a newly minted Microsoft research company, busy solving problems.
It must feel good.