About 10 years go, long before she took the lead at the Omnia artificial intelligence practice of Deloitte, Shelby Austin was working as a partner in a law firm, reviewing documents and preparing for trial. The documents — all 30,000 of them — were in binders, and they were being divided “relevant,” “privileged” and “hot documents,” which would get a pink sticky note.
“It took weeks,” Austin recalled in her keynote at the combined Big Data Toronto and AI Toronto conferences this week. “I would stumble home for a shower and Diet Coke. When I learned there was a form of machine learning that could be applied to review and classify the documents automatically in some fashion.”
Austin would later quit her job and launch a firm that was later acquired by Deloitte. At Omnia, she now finds herself caught between excited hype among some players in the AI space and deep skepticism from other organizations. In some respects, she said, it was not unlike her initial exposure to AI.
“I didn’t realize there was an AI winter,” she said, referring to a period of reducing funding and interest in areas like AI. “I just saw soem really useful science that could solve my problem.”
At Omnia, Austin is now helping customers think about the ways machine learning and other technologies could be used to boost revenue, reduce costs or enhance the customer experience. Getting started often means converting what she called “non-prediction” problems into something where AI could bring value. She gave the example of autonomous vehicles, which existed in factories for a long time but were never considered something that could run on the road.
“We couldn’t make enough if/then statements to make it worthwhile,” she said. “But then you reframe the problem: what would a good human driver do? This same thinking can be applied to loads of problems.”
Omnia is suggesting business professionals might approach some problems where the ability to make predictions haven’t been considered by asking “what would a good employee do?” or, in the same of government “what would a good citizen do?”
At companies like e-commerce giant Shopify, it’s more a case of asking what a bad person might do. In a panel discussion at the conference, Shopify data science lead Sarah Siu said the team has created an AI-based tool to identify potentially fraudulent transactions.
“These are things we have a lot of data on, and can understand the patterns with what’s happening with these transactions,” she said. “It helps (our merchants) run their business better without having to worry.”
Siu said Shopify also sees AI as a means to scale — in other words, growing without adding a lot of additional resources. Rather than bringing on more support staff to its help desk, for instance, AI is being used to offer smarter recommendations to merchants on how solve common problems using the documentation available to them.
As more of the predictive capabilities within AI are applied to these kinds of problems, the relationship between emerging firms and large, traditional enterprises may begin to change, said Stephan Piron, whose firm DeepLearni.ng announced its rebrand as Dessa at the conference.
“Startups are supposed to come in and destroy and disrupt the big ones,” he pointed out. “But the data sets that are the fuel for AI forces companies to work better together.”
Dessa is already working with a mutual fund company, for instance, to help use AI in image recognition so that back office functions involving fax machines can be replaced, and a telco that uses AI to predict who will reach out to its call centre.
If we want to avoid another AI winter, Austin said she and her team at Omnia see a number of hurdles to overcome. This includes the fact that the AI vendor landscape is fragmented, there is shortage of talent, largely only internal uses, difficulty with integrations and challenges with user adoption and change management.
“These are the unsexy problems of doing business,” she said, “and have nothing to do with the hype.”
Big Data Toronto and AI Toronto wrapped up on Wednesday.