Mysterious. Powerful. Game-changing. Artificial intelligence (AI) attracts many vague adjectives, however enticing. And despite its rapid rise in public consciousness, it’s perceived by many as inaccessible and inscrutable — as something that’s “out there.”
AI isn’t out there. It lives at MaRS.
Dozens of MaRS-supported companies — from startups to global giants — are all about deep learning, contrastive divergence, Skip-Gram, softmax and hyperplane. These tools and concepts are central to developing and advancing AI technologies. But the average person doesn’t need to know what they mean to gain a better understanding of AI.
Enter Thinkable: AI and Human Engagement, a recent MaRS event offering talks, hands-on demos, startling insights and engaging one-on-one encounters. Nora Young, host and co-producer of CBC Radio One’s Spark, conversed with AI scientists and thinkers in front of a crowd of students, technologists, marketers, investors, designers and journalists in Autodesk‘s street front space.
The AI technologies presented at Thinkable were tied to very relatable human issues and concerns, from how we judge each other’s opinions to how we assess our mental states as we age.
TwentyBN’s Roland Memisevic, the company’s chief scientist, explained that while AI is getting more “intelligent,” it’s crucial that the technology learn basic perceptual concepts — things that a one-year-old child knows intuitively, like whether a person is opening or closing a door. TwentyBN created what it calls “the world’s largest movie studio” (thousands of volunteers around the world, online, filming things) to teach AI the stuff it needs to better understand people.
The company breaks down human physical behaviour into hundreds of gestures, labelled and catalogued through thousands of video examples, to develop AI technology that will soon be able to “watch” live video feeds or “read” the gesture of a nurse, for example, so it can deliver a required medical instrument without the need to touch a non-sterile button.
The delete button is used way too often by social media professionals tasked with the stressful job of wading through deeply offensive comments. Isar Nejadgholi, head of machine learning research at IMRSV Data Labs, demonstrated how the company’s technology may soon be able to hand that unpleasant task over to machines.
IMRSV’s AI-powered Toxic Language Identifier tags text that might be insulting, threatening, obscene or hateful. AI can’t yet determine sarcasm, irony or other subtleties of human exchange — though, sometimes we can’t either — but the company’s tech augments efforts to make online forums safer and more civil for public debate.
WinterLight Labs’ Liam Kaufman told us something surprising about the way we talk — even when we’re not trying to be controversial.
The company’s AI technology quantifies speech and language patterns to help detect and monitor cognitive and mental diseases. At Thinkable, he contrasted two interviews featuring Gene Wilder (diagnosed with Alzheimer’s several years before his death in 2016) that were filmed decades apart. Even though the audience couldn’t notice a change in the actor’s mental acuity, the AI program was able to identify in the newer video an increase in his use of pronouns and a drop in the number of clauses, both signifiers of mental decline.
Increasingly, AI is getting better at prediction. And one of Canada’s most established innovators revealed how the past can be mined for glimpses of what might be over the horizon.
Dr. Foteini Agrafioti, head of Borealis AI (RBC’s research institute), showcased Apollo, a tool that does what no single financial analyst can: read all the news and know “everything” about a given area of sector specialty. Working with massive news datasets, the program applies deep learning and statistical inference to predict the probability of future events pertinent to a company’s place in the business ecosystem. Whoa.
Closing out the evening, Hilmar Koch, Autodesk’s guru in the future of storytelling, asked the audience an important question: “Are you becoming species-fluid?” In a wide-ranging talk covering philosophy, fuzzy photos of cats and a dinosaur made out of data points, Koch deftly explored the challenge of how we might come to understand the species known as AI. He believes humans will have to reconsider how they perceive, learn and share knowledge. A journey from data to information to knowledge and finally to wisdom and reasoning is one that makes us human, but also makes us realize the depth and breadth of the tasks ahead with AI.
In summary, it was an evening of provocative discussion. Still, while AI is starting to do astonishing things, there remained a recurring theme: most of the thinking is still up to us.
Learn more about the groundbreaking technologies being developed at MaRS in the latest issue of our magazine.