Back in 2010, as classmates in our AI class (CS 486) at the University of Waterloo, we started to think about the way humans interacted with machines. Graphical User Interfaces (GUI) had been in use for 30 years and yet, they hadn’t changed much. For simple tasks they were easier to use than the command line interfaces, but for complex tasks we still resorted back to programming. We wondered why was this the case? Why couldn’t we just interact with computers the same way we interacted with each other everyday? We had to go to first principles and came to the realization that in order to achieve this level of natural interaction, we had to first develop algorithms that understand the way human beings communicate. Therefore, we had to have a very deep understanding about the fundamentals of human language; our memory and reasoning capabilities; as well the decision making process in our brain.
A couple of years later, we started to develop technology that could solve some of the basic problems of language understanding. At the time, the language understanding community (both academia and industry) was very intrigued by the early success of statistical machine learning algorithms in Personal Assistant systems like Siri. Users could make voice commands and do simple tasks like finding the weather, making a restaurant reservation or even playing some music from the phone. Besides the fact that these systems were extremely unscalable (built by engineers in a domain-by-domain fashion), brittle (keyword style queries worked) and gave users a very poor experience, these systems had a more fundamental flaw - they lacked the intelligence that humans have. In fact, this fallacy didn’t just hold for Personal Assistants, this was true for every machine out there. Machines just don’t think, reason or learn from their mistakes like we humans do. Machines neither have any common sense reasoning, nor they do have short-term, long-term or working memory like us.
In early 2014, we observed that great leaps had been achieved in the fields of computer vision and speech recognition through the application of Deep Learning algorithms. We were excited - if deep learning techniques could enable machines to see and hear like humans, then why not communicate like humans? As we all know, understanding human language is extremely complex and is ultimately the holy grail in the field of Artificial Intelligence. We finally saw a great opportunity to apply Deep Learning and Reinforcement Learning techniques to solve fundamental problems in language understanding, with the vision of creating a truly literate machine - one that could actually read, comprehend, synthesize, infer and make logical decisions like humans. This meant we had to heavily invest in research, therefore we started our Research lab in Montréal in late 2015 (in addition to our awesome engineering team in Waterloo). Our research lab, located at the epicentre of Deep Learning, is focused on advancing the state-of-the-art in deep learning for human language understanding. We have built a team of top Deep Learning Research Scientists and Engineers from around the world and built partnerships with leading academics in the field. We are extremely proud of the breakthroughs we have accomplished over the course of the year.
So where are we in our quest for achieving ‘Machine Literacy’? Well, we are just getting started and are excited about the long road ahead.
CTO and Co-Founder
CEO and Co-Founder
Maluuba is backed by an accomplished team of leading investors and some of the most respected researchers in the fields of artificial intelligence and deep learning.