When Jordan Jacobs approached former TD Bank Group CEO Ed Clark in late 2016 with the idea to fundraise for a world-leading artificial intelligence research facility, he found a willing and motivated partner.
“I sent an e-mail to Ed at about 10 p.m. on a Friday saying, ‘Here are the first 20 companies we should go to,’” says Jacobs, CEO of Toronto-based Layer 6 AI, an AI software company. “He called me just before noon on Saturday and said, ‘OK, I’ve raised $70 million this morning. What have you done?’” When the dust had settled only a few months later, the Vector Institute had raised over $150 million in public and corporate money—including pledges from Google, Magna, Loblaw and each of the Big Five banks. Its mandate is to drive the adoption of AI technologies across Canada, facilitate the commercialization of AI research and serve as a centre of gravity for global AI talent.
When explaining how AI is expected to transform business and broader society, experts struggle to find historical comparisons of the same scale. Even the internet doesn’t measure up. Clark, who now chairs the Vector Institute’s board, says it’s much farther-reaching. “Just as electricity was transformative—what could you do once you had electricity—AI is the same.” Clark also sits on the board of Thomson Reuters, and says that after a demonstration of AI technology was given to his fellow directors at the company earlier this year, they “were blown away.”
The headline-grabbing establishment of the Vector Institute punctuates a frenzy of recent AI activity that’s been hard to miss. At the same time, however, for many leadership teams looking to get their feet wet with the technology, confusion remains over how to begin. Clark says, “I think a lot of people are walking around with an exotic view of AI, but they’re not quite sure about, ‘What does it actually mean?’” A McKinsey Global Institute survey of C-level executives at over 3,000 companies supports this assessment, saying that “many business leaders are uncertain about what exactly AI can do for them, where to obtain AI-powered applications, how to integrate them into their companies, and how to assess the return on an investment in the technology.”
For every day those questions go unanswered, early AI adopters build a compounding competitive edge. As the McKinsey report says: “There are no shortcuts for firms. Companies cannot delay advancing their digital journeys, including AI,” as the “gap with the laggards looks set to grow.” Indeed, this sentiment was expressed as early as 2014, when technology and market research firm Gartner told its clients that “the risk of investing too late in smart machines is likely greater than the risk of investing too soon.”
The numbers being thrown around by analysts about AI’s impact are enormous. Accenture, looking at 16 major industries, claims AI has the potential to boost profits by an average of 38% by 2035. Market researcher IDC projects corporate spending on AI solutions to reach over US$46 billion by 2020, up from an estimated US$12.5 billion this year.
Some Canadian companies are experimenting aggressively with the technology. Scotiabank, for instance, has engaged with Toronto AI start-up DeepLearni.ng to develop an AI system that recommends next steps—calls, e-mails or patience—for overdue credit card payments after analysis of personal credit history, customer behaviour patterns and existing collections data. The bank also revealed a partnership with Layer 6 AI to create a product recommendation engine that personalizes marketing of its products to holders of Cineplex Scene loyalty cards.
On the e-commerce side, Shopify is employing machine learning to determine the credit risk of all applicants to the company’s Merchant Cash Advance program before spitting out customized offers. The technology was credited in the company’s August earnings call when executives explained that they had “nearly doubled the number of capital advances in Q2 over Q1.”
Elsewhere, the likes of Johnson & Johnson, PayPal and KLM are using artificial intelligence to streamline supply chains, detect fraud and money laundering, and augment online customer service with chatbots that interact directly with customers.
Of course, for every new application and announcement there is a corresponding increase in unease about how AI will reshape the professional landscape. How do you introduce the technology to employees who have read that it threatens to do their jobs faster and more accurately than they do—or replace them altogether?
True, it’s still early days. But experts say corporate directors would be wise to take these developments and concerns seriously—and insist executives under their oversight think through their digital strategies, AI use cases and talent needs lest they find the selves eclipsed by bolder industry peers. Those same experts have many insights to offer on the challenge of smoothly introducing AI into your company. But before diving into that advice, it behoves directors and C-suite executives to understand what AI is, and why it has come to such sudden prominence.
AI’S “CAMBRIAN EXPLOSION.” That’s how Nvidia CEO Jensen Huang and Andreessen Horowitz partner Frank Chen have described the activity of the last five years. Finding your bearings amongst the AI proliferation starts with demystifying the technology.
“Currently, when people say AI,” says Jacobs, “often what they’re talking about is deep learning.”
Deep learning is a branch of machine learning, which is itself a subcategory of artificial intelligence, an umbrella term covering a number of techniques that enable machines to mimic human intelligence. Machine learning involves computational models, called artificial neural networks, that “think” like humans and that learn to reproduce a human behaviour by recognizing patterns in massive amounts of data, gaining experience and improving at achieving an outcome. “Learn” is the key word here. Developers do not program a computer to, say, recognize an image or a spoken word per se. Instead, they create an algorithm that learns how to do this.
At specific tasks, machine learning is giving us humans a run for our money. Google DeepMind’s AlphaGo famously beat the world Go champion earlier this year with the help of machine learning techniques. Sony Computer Science Laboratories’ DeepBach learned to create music that is virtually indistinguishable from Bach, at least to the layperson’s ear. Chances are you’ve already interacted with more mundane machine learning applications when Netflix made personalized viewing recommendations for you or Amazon suggested items you might want to add to your cart.
Deep learning involves artificial neural networks that operate using several “layers” of analysis to recognize increasingly abstract patterns and concepts. Data, including images or speech, is fed into the first layer which passes its output to a successive layer to be analyzed, and so on, with each layer recognizing patterns of increasing sophistication until a final output is delivered—like the identification of a traffic sign or a set of spoken words. Deep learning is the underpinning of self-driving cars, the speech translation on your phone and the automatic tagging of people in your Facebook photos.
The reason for the recent surge in deep learning activity—and for AI’s sudden, intensively hyped emergence as a strategic linchpin in business—lies in improvements to three key elements of the technology: learning algorithms, data volume and computing power.
Each of these has a rich history, in which many of the same names—including a number of Canadians—come up time and again. Researchers focused on algorithms, for example, toiled rather inconspicuously until 2012, when two University of Toronto students studying under deep learning pioneer Geoffrey Hinton debuted a deep learning algorithm called AlexNet. The venue was a competition among AI research teams from around the world to detect and classify objects from a set of more than a million images. AlexNet out-performed its nearest rival by such a large margin that Google hired Hinton and his two students six months later. The promise and potential benefits of deep learning suddenly seemed real.
A key to AlexNet’s breakthrough performance was the enormous volume of images in the contest. Deep neural networks train themselves on data, and they need reams of it. Before AlphaGo beat the human champion, for instance, it trained on 30 million board positions from 160,000 real games, then played itself millions of times. Fortunately, today’s world has no shortage of data, with 2.2 billion gigabytes of it being generated per day according to the McKinsey Global Institute, with companies in tech, media, financial services, retail and telecommunications being particularly good at creating an inventory.
Big Data and learning algorithms go hand-in-hand with advancements in computer processing. A key breakthrough in this area took place four years ago when a team led by Andrew Ng, a professor of computer science at Stanford University and founder of the deep learning research project Google Brain, partnered with chipmaker Nvidia to demonstrate that the latter’s graphics processing units (GPUs)—processors originally designed to carry the heavy loads required by gamers—were equally well-suited for neural networks. Specifically, they showed that Google Brain’s achievements, which required 1,000 servers the year before, could be realized on only three GPU-accelerated servers at a fraction of the cost.
The race to market for chipmakers and service providers since then has been staggering. Today, Amazon and Microsoft both offer cloud-based GPU processing services. Last year, Google unveiled the tensor processing unit (TPU), a proprietary chip optimized to run neural networks, and this year rolled out Cloud TPU, the first processor to run and train neural networks through a dedicated cloud service. Traditional chipmaking giant Intel, meanwhile, spent US$400 million on AI start-up Nervana Systems last summer, and announced testing of the Nervana Engine—claiming that by 2020 they would “deliver a 100-fold increase in performance” over today’s best GPUs.
AS EXCITING AS that sounds, executives and directors eager to begin their company’s AI journey first need a starting point. In this context, the experiences of Australian energy executive Peter Coleman, CEO of Woodside Petroleum, Australia’s largest publicly traded oil and gas company, are instructive.
Coleman is a pioneer among executives who have brought AI into their companies. Speaking at IBM’s World of Watson conference last year, Coleman explained how a program powered by Watson, IBM’s AI platform, helped solve Woodside’s problem of being “rich in history, rich in data, but…poor in analysis.” Specifically, the program, called Lessons Learned, is a repository of company knowledge gained over three decades of drilling offshore and in sensitive environments. Watson was trained using Q&As to become smarter, so that now when an engineer asks the program for help with a project-related question, it combs through the repository to find the most suitable answer and deliver it in plain language. The accumulated wisdom of senior staff that would have previously walked out the door when they retired is now preserved and in reach.
To initially sell his executive committee on the Lessons Learned system, Coleman shifted the spotlight from the technology to the business problem. Only two years earlier, Woodside had completed a $15-billion liquefied natural gas project. The job had a big learning curve, yet when Coleman challenged his executive team to come up with five lessons learned on the spot, they were stumped. As Coleman told the World of Watson audience: “If we can’t give five [lessons], how in the heck do we think the rest of the organization’s going to do it? And so, we said, ‘There’s the business case.’”
The Woodside experience highlights two common problems faced by companies exploring AI solutions: identifying a use case and achieving buy-in. There are many more. Likewise, just as Coleman devised a solution, other executives and directors who have made early inroads in AI have done likewise, while learning many of their own lessons along the way. Here, then, are five guiding principles for AI adoption based on that advice and experience.
1. ENSURE AI ACCEPTANCE STARTS AT THE TOP
Addressing AI, McKinsey says that “behavioral change will be critical, and one of top management’s key roles will be to influence and encourage it.”
Anastassia Lauterbach knows about encouraging technology adoption. Along with serving on the board of commercial data provider Dun & Bradstreet Corp. and chairing its technology and innovation committee, Lauterbach, along with GEC Risk Advisory CEO Andrea Bonime-Blanc, is co-authoring a book, The Artificial Intelligence Imperative: A Practical Roadmap for Business, slated for release next spring.
Lauterbach, who interviewed 60 corporate directors on AI in business for the book, says, “I recommend companies introduce technology committees to deal with the question of IT transformation, migration to the cloud, cybersecurity and AI. If companies’ boards are not ready to hire digital directors, they should consider introducing an advisory board with representatives from high-tech companies.”
For traditionally non-tech companies, Lauterbach says, “Currently these companies don’t compete on talent in AI, as they lack understanding in what AI requires, why they need it and what capabilities they have to acquire to stay competitive. Unfortunately, AI in traditional companies has to be a top-down topic. Leadership has to be capable to be willing to embrace it.”
While support at the top is critical, it is only a beginning. A machine learning implementation requires resolute commitment, as the algorithm needs to be trained over time with your data and industry’s language. Effectively managing change throughout the organization will be the difference between an AI implementation that leads to significant efficiency or profit, and one that muddles along with lacklustre results or fails altogether.
Ed Clark, hypothetically putting himself back into the CEO’s chair, says the technology’s introduction is likely to meet “natural resistance spots within an organization that say, ‘Well, you know, you’re putting me out of a job or you’re going to force me to reorganize everything I do,’ and people don’t do that naturally.”
Reflecting on how to execute the change, he says one could ask, “How do I find areas in the company where they are change-ready enough that I could start to introduce these concepts to them, and that people could learn what AI could do for [them]?”
A second option is “to create separate groups and give them singular problems,” he says. “When people were first going into e-commerce, many people set up e-commerce groups, because they felt, ‘I’ve got to get it far enough away. The mainstream organization will never experiment enough to do it.’”
That’s precisely how Woodside began with Lessons Learned, having the group that was training Watson report directly to the CEO, giving them a licence to innovate and “protection from the organizational processes.”
That decision at Woodside to put the Watson team “into almost a box” has made all the difference. Initial skepticism about Watson gave way to enthusiasm, as the engineers working with Lessons Learned became its greatest ambassadors, and other departments, including legal and HR, began to request implementations of their own. According to Debbie Landers, vice-president of cognitive solutions at IBM Canada, “They’re so successful with it, they’re doing seven more Watson projects, including bringing it into their boardroom…for executive decision-making.”
2. KNOW HOW YOUR COMPANY STACKS UP
Beneath a wholesome-looking photo, Carl Icahn’s Twitter profile reads, “Some people get rich studying artificial intelligence. Me, I make money studying natural stupidity.” That handle may be revised in the near future, as activist investors—so-called wolf packs—will surely start to scrutinize companies’ AI adoption plans.
David Beatty, Conway director of the Clarkson Centre for Business Ethics & Board Effectiveness at the Rotman School of Business, warns that boards and executives must not fall behind the curve with regard to AI.
“If you don’t do your job as a board now effectively, somebody else is going to come in and do it for you,” says Beatty. “How will AI, or any digital disruption, impact your company? There’s going to be activist firms asking that very same question, and if they’re finding better answers than you are, they’re going to let you know about it darn quick, and…they can take a run at taking you over.”
A PwC report says, “The starting point for strategic evaluation is a scan of the technological developments and competitive pressures coming up within your sector, how quickly they will arrive, and how you will respond.” Ramy Sedra, partner and data and analytics consulting leader at PwC Canada, advises asking: “Does my management team understand as an organization how AI may either threaten my business [or] present opportunities for us? Are there potential disruptors or competitors in the market that, frankly, we may be tempted to dismiss?”
Clark says, “I think I’d be saying, ‘Have you done any benchmarking of where we are versus other firms?’ So have you gone out and tried to find out what’s the leading firm in your sector? What are they doing here?”
Once the board has done that, a plan has to be promptly formulated and communicated, according to Erin Essenmacher, chief programming officer at the National Association of Corporate Directors in Washington, DC. She says, “Thoughtful, proactive communication with shareholders that tells a compelling story about the company’s strategic direction is key. That has always been true, but it’s becoming an imperative now more than ever given the pace of change and disruption in the environment.”
3. ENCOURAGE MANAGEMENT TO EMBRACE AI EXPERIMENTATION
Clark says, “I think one of the positive roles boards can play is to spur management on to, say, be a little bit more aggressive about moving faster in these areas.”
McKinsey supports that opinion. “AI can deliver significant competitive advantages, but only for firms that are fully committed to it,” it writes. “Take any ingredient away—a strong digital starting point, serious adoption of AI or a proactive strategic posture—and profit margins are much less impressive.”
This doesn’t mean you have to go all in right away. Martin Whistler, associate director and lead analyst for media and entertainment at EY in the UK, says, “You don’t need to throw yourself into this AI lake and see whether you sink or swim. I think it’s about experimentation, where it applies in your business, and then going out and making some strategic, but not necessarily high-risk investments at this stage.
“What you need to do is start looking at areas of the business where you can experiment with some basic AI technologies. For example, if you’ve got a lot of contracts to review or if you’ve got a lot of processes today that might be manual or might be repetitive. Those are areas that are tailor-made for some kind of AI solution,” Whistler says. “Those are relatively low-cost, simple experimentation areas that you should start to invest in and explore.”
Beyond just being proactive with AI experiments, directors should take the opportunity to talk with the people actually using the technology. Says Clark: “I’d be asking, ‘OK, if we are doing something, can the board meet the people who are doing it?’”
4. DETERMINE HOW TO TACKLE YOUR AI TALENT NEEDS
When it comes to hiring high-level expertise in AI and data analytics, “It’s a talent war,” says PwC’s Sedra. “It’s a small community. It’s a growing market.”
Solving the talent crunch was one of the inspirations behind the creation of the Vector Institute, according to Jacobs. He and his Layer 6 AI partner had done an informal survey asking, “If you could hire PhDs and masters in machine learning in the next five years, how many would you hire?” He says that only accounting for companies in Toronto and Waterloo, “we stopped counting when we got to 5,000, because that was already 10 [times the amount] the entire country would produce.”
Star AI researchers are now reportedly earning paydays like professional athletes, and the usual tech suspects are scooping most of them up. Google, as mentioned, hired Geoffrey Hinton. IBM brought on Yoshua Bengio, a University of Montreal deep learning expert and founder of the high-profile AI service company Element AI. Stanford’s Andrew Ng, of Google Brain fame, became chief scientist at Baidu, where he led the company’s AI group (he resigned in March). University of Toronto associate professor and Vector Institute cofounder Raquel Urtasun was signed by Uber in May. Microsoft acquired Canadian AI talent this year with the purchase of Montreal-based deep learning start-up Maluuba. Google’s DeepMind just opened its first satellite research lab in Edmonton, where it tapped Richard Sutton, renowned researcher in a branch of machine learning called reinforcement learning, to head up its team.
Some Canadian non-tech heavyweight companies have been able to elbow their way to the talent table. RBC also managed to nab Sutton by collaborating with the Alberta Machine Intelligence Institute and establishing a lab in Edmonton, where the professor will take on a role as academic adviser to RBC Research in machine learning. Gabriel Woo, RBC’s vice-president of innovation, says the collaboration will help them “to build a world-class AI research group” from which “we expect we will be able to find applications of AI to financial services that will help us deliver better services to our clients, streamline and improve operations, reduce risks, etc.” RBC, along with BDC Capital, Scotiabank and Magna, also partnered with NEXT Canada to launch AI incubator NextAI, supplying successful applicants with mentorship, technology and $200,000 in funding.
For many leaders who may feel urgency but are missing the deep pockets to build an internal AI capability (or a lab for that matter), it may be best to stay out of the expensive talent war at first. Clark says executives should ask themselves, “How much I’m trying to build this capacity, or how much I am willing to buy this capacity.” Given the large number of AI start-ups in the market and the challenges involved with change management, some non-tech firms would do best to “start quick with someone who already is doing this and knows it well, and then build capacity, your internal capacity, at the same time, which will take a while.”
5. ENSURE THAT DATA AND AI ARE NOT SILOED
One PwC report states that “AI, in its current form, has a major challenge: intelligence is only as good as the data it can learn from.”
A company’s executive team must understand what data they have and are collecting, where gaps in data exist, and how they can fill those gaps to achieve the outcomes they seek with machine learning. “AI starts and ends with data,” Lauterbach says. “The majority of traditional companies I am advising do not have a sense of the value of their data.” She states that questions about data should be as common as sales updates and regulatory briefings at board and leadership meetings.
Imperative in a data strategy is ensuring that data is not siloed, but available to leadership and across the business. “Analytics and AI are going to be ubiquitous across all functions, and therefore it needs to be aligned with the highest level of strategy and management in an organization, and if it’s buried in IT or if it’s buried in business or operations, then it’s probably going to have a much narrower impact, and it will be easier for dissidents in an organization to resist,” says PwC’s Sedra.
According to Sedra, this may mean appointing a chief data officer who reports directly to the CEO. “If they are not reporting to the CEO there’s a major risk that that analytics program will fail.”
EY’s Whistler agrees, saying AI must be an integral part of your core business. “Historically, there’s been a danger that it’s very siloed in the CTO-type space,” he says, “It’s far, far broader than that. You need to have a conversation right across the business.”
A clear data strategy has the added benefit of helping on the recruiting front. As Clark says, “What does an AI researcher want? They want data and problems. They’re the happiest people in the world if you give them those two things.”
Whatever your company’s size or sector, managing these steps on the road to integrating AI effectively is a task that will require finesse, and it’s one executives need to start planning. There may be urgency, but just as there is risk in not acting, trying to move too far, too fast can hamstring companies by making them “less flexible and less able to adjust to a changing competitive landscape,” according to Gartner. To avoid that, it recommends developing a plan “for achieving the right balance of AI and human skills,” creating a timeline for implementation and communicating it with employees. In the end, the rewards for getting it right will be huge if AI turns out to be even half as transformative as proponents say.
And of that, Jordan Jacobs, alluding to the AI-as-electricity analogy, has no doubt. “If you’re an organization that has tons of data, and you’re not using AI, it’s literally like you’re working in a room with the lights off,” he says. “AI is like switching the lights on.”
Photography by Jaime Hogge (Ed Clark); Woodside Petroleum; Vector Institute