Rotman Management Magazine
The Future of Growth: AI Comes of Age
A new factor of production is on the horizon, and it promises to transform economic growth for countries around the world.
IN THE MODERN ECONOMY, there are two traditiona ldrivers of production: increases in capital investment and labour. However, the decades-long ability of these drivers to propel economic progress in most developed countries is on the cusp of a massive change.
With the recent convergence of a transformative set of technologies, economies are entering a new era in which artificial intelligence (AI) has the potential to overcome the physical limitations of capital and labour and open up new sources of value and growth. AI is, without question, the single most disruptive technology the world has experienced since the Industrial Revolution. In this article we will discuss some of the implications, challenges and opportunities of this new fact of economic life.
A New Factor of Production
For three decades, rates of gross domestic product (GDP) growth have been shrinking across the globe. Key measures of economic efficiency are trending sharply downward, while labour-force growth across the developed world is largely stagnant.
Are we experiencing the end of growth and prosperity as we know it? The short answer is an emphatic No, because the data misses an important part of the story: How new technologies affect growth in the economy. Traditionally, growth has occurred when the stock of capital or labour increased, or when they were used more efficiently. The growth that comes from innovation and technological change in the economy is captured in total factor productivity (TFP). Economists have always thought of new technologies as driving growth through their ability to enhance TFP, and this made sense for the technologies that we have seen — until now.
What if AI has the potential to be not just another driver of TFP, but an entirely new factor of production that can replicate labour activities at much greater scale and speed, and even perform some tasks beyond the capabilities of humans?
For example, Meta — now part of the Chan Zuckerberg Initiative — uses AI to read, understand and act on the thousands of scientific papers that are published daily. For context, over
4,000 papers are published daily in the field of biomedicine alone. Meta’s system helps scientists access these mountains of research to learn from real-time insights and unlock scientific discoveries years in advance.
Unlike traditional capital like machines and buildings, AI can improve over time, thanks to its self-learning capabilities. The Spanish AI start-up NEM Solutions, using an algorithm based on the human immune system, is targeting wind-farm productivity by predicting and preventing failures. The platform first analyzes instances of wind turbine failure to learn what the symptoms are, then monitors the turbines in real time to detect the symptoms and flag any potential problems.
Of course, AI is not a new field; much of its theoretical and technological underpinning was developed over the past 70 years. Its applicability, though, is a relatively modern development. AI went out of favour in the 1970s and 1980 because technological capabilities such as limited computing power fundamentally limited the capabilities of researchers. That changed in the early 2000s, when three Canadian-based researchers — Geoffrey Hinton, Yoshua Bengio and Rich Sutton — made breakthroughs that re-popularized the study of AI.
Over the last ten years, increases in efficient computing power, data quality and data quantity have redefined how we look at AI. Today, the term refers to multiple technologies that can be combined in different ways to sense, comprehend and act. All three capabilities are underpinned by the ability to learn from experience and adapt over time.
Computer vision and audio processing, for example, SENSE. perceive the world by acquiring and processing images, sounds and speech to develop enhanced data.
Natural language processing and recommendation COMPREHEND. engines, for instance, can analyze and understand the data collected by generating meaning and insights.
One of the key components of AI systems is their ability to ACT. use the information generated to take action like in the case of augmented reality or, more simply, chatbots.
Five Levers of AI-LED Growth
So, how can organizations drive value from AI? Businesses that successfully apply it could increase profitability by an average of 38 per cent by 2035, according to a recent research report we did in conjunction with Frontier Economics. That’s a compelling case. We think about AI delivering value in terms of five levers:
This involves deploying cognitive 1. INTELLIGENT AUTOMATION. capabilities on top of traditional automation technologies to achieve self-learning, greater autonomy and flexibility. Results include more efficient processes, activities, and services beyond what traditional automation will deliver.
This involves delivering superior 2. IMPROVED INTERACTIONS. experiences to customers and users based on hyper-personalization and the curation of real-time information. On top of overall satisfaction improvement, this can also generate greater acquisition and retention rates among customers.
Leveraging AI capabilities to augment 3. ENHANCED JUDGMENT. human analytical and management capabilities. Results include improved quality and effectiveness of prediction and decision making.
AI can be used to build trust with custom4. DEEPENED TRUST. ers and within the organization by more effectively preventing and detecting anomalies. It also provides the ability to significantly reduce false positives, which further improves efficiency.
Deploying AI to enable a new class of 5. INNOVATION DIFFUSION. products and services that use AI to enhance the product development lifecycle and create new businesses. Results include increased speed with which new products and services are designed and delivered.
Let’s look at a few examples. The aircraft manufacturer Airbus was looking for ways to achieve more accuracy and quality in cabin furnishing. Accenture worked with Airbus to develop
AI can boost labour productivity only if companies are willing to disrupt their legacy models.
a solution involving smart glasses. Using contextual marking instructions, the smart glasses display all required information for an operator to help mark the floor faster and reduce errors to zero, with a built-in ability to validate the work and provide real-time feedback to users along the way. You can imagine the applicability to many operations that require precision in the set up or implementation of equipment.
In California, AI start-up Elementum generates real-time insights when incidents or disruptions threaten a supplier, helping its clients understand where every component and finished good is supplied, manufactured and distributed. Rather than simply automating supplier management processes, clients of Elementum can get early warnings of potential problems and alternative solutions to react before production is impacted. For instance, in 2014 a fire in a Chinese DRAM [Dynamic RandomAccess Memory] chip factory put a 25 per cent squeeze on world supply. Whereas most equipment manufacturers only found out days later, Elementum’s customers knew about the incident within minutes and secured their supply of DRAMS before prices reacted to the shortage.
It is not just production and supply chains that can benefit from intelligent automation. One of Accenture’s clients — a global insurance company — wanted to automate its claims processing for auto insurance. We worked with them to develop an algorithm using a data set of toy-car images. The solution enables customers to send their own pictures of the damaged car to the insurer, and the algorithm classifies the damage, replicating the work of an adjuster, with 90 per cent accuracy. In addition to reducing the effort of humans in assessing the damage, the solution can also be extended to requisition parts and detect potential fraud cases.
Elsewhere, in March, Capital One revealed Eno, the first of its kind natural language chatbot for banking. During the pilot phase, customers could text Eno anytime to review their accounts, pay their credit card bill, or just ask general questions. As of this month, Eno is available to communicate by text with millions of Capital One credit card and bank customers. Capital One has revealed three surprising things it learned in the pilot process:
• Every customer has their own language and conversational style. Therefore, the agent had to learn how different people like to text about their money. This includes the use of emojis. For example, some customers like to use a thumbs-up emoji to confirm their payment;
• Language, tone and meaning trainers have been required to help Eno interpret the 2,200 different ways customers may ask for their balance;
• And chatbots actually need empathy! People will tend to build relationships with them even while knowing that they are talking to a bot.
However, improved interaction is not just about interacting with customers. One of the world’s largest oilfield services companies — which creates products and services to analyze, drill, evaluate, complete and produce oil and gas reserves and then transport and refine the hydrocarbons — wanted a way to respond more efficiently to its vendors’ inquiries about their invoices and payments. Vendors can interact with a digital assistant and receive information about the status of their invoices. This includes checking invoice status, and searching for invoices in back-end systems. Vendors can also use the virtual agent to upload missing invoices and log trouble tickets.
A key use case for enhanced judgment is in recommender systems. Machine learning and deep-learning models have been used to personalize recommendations for movies, research articles and products in general, and there are now recommender systems for experts, collaborators, job candidates and romantic partners. Canadian company Layer 6 recently won an international challenge for its work on ‘cold-start’ recommendation systems — cases where there is no interaction history to draw from. Layer 6’s deep learning platform allows users to leverage a wide variety of historical data and solves the cold-start problem by incorporating data from the current user session and context.
AI is also spreading to areas where intellect and critical thinking have long dominated. For instance, start-up Narrative Science is ‘humanizing’ data with technology that interprets an organization’s data, then transforms it into intelligent
narratives in a style that a human might write. Take, for example, the suspicious activity reporting AML [anti-money-laundering] investigators are required to do. For a large bank that averages 4,000 alerts a year, typically over 150 cases need to be filed with regulatory bodies. Narrative Science’s platform, Quill, can reduce the time it takes to file cases by automating the narrative required that explains the suspicious transactions, saving 2.5 hours per case.
California ai company Auto desk is pioneering this approach with its computer-aided design system, Dreamcatcher. Using AI to mimic the generative design work of nature, Dreamcatcher creates thousands of virtual prototype iterations and compares their function, cost and material according to specified criteria. In the healthcare industry, Dreamcatcher has already been used to design a facial implant that accelerates recovery and tissue regrowth.
Factoring in AI
To understand the value of AI as a new factor of production, Accenture, in association with Frontier Economics, modelled the potential impact of AI for 12 developed economies that together generate more than half of the world’s economic output. Our results reveal unprecedented opportunities for value creation: AI has the potential to double annual economic growth rates across these countries. In Canada, the increased labour productivity that AI offers could potentially reduce the number of years required for Canada to double the size of its economy by 13 years if it achieves an Ai-steady state by 2035.
AI also has the potential to boost labour productivity by up to 40 per cent by 2035 in the countries we studied. Optimal labour productivity will not be driven by longer hours, though, but by innovative technologies enabling people to efficiently use their time. This labour productivity increase dramatically reduces the number of years required for our analyzed countries’ economies to double in size. The results are primarily driven by a country’s ability to diffuse technological innovations into its wider economic infrastructure. While the gains vary in each country surveyed, our research shows AI can transcend regional and structural disparities, enabling huge, rapid leaps in labour productivity.
AI can boost labour productivity, though, only if companies are willing to disrupt their legacy models. An Accenture study found that companies that optimally use AI will generate higher shareholder value. However, less than a fifth of leading companies that leverage AI have achieved this performance. Accenture’s research found that only 17 per cent of Canadian companies leverage AI successfully — demonstrating the ability to innovate from within and collaborate externally. The research shows that companies must converge and integrate technology, data and people to improve what we call their ‘AIQ’.
One third of the skills that will be required in three years are not yet considered crucial.
Clearing the Path to an AI Future
Entrepreneur Elon Musk has warned that AI could become humanity’s ‘biggest existential threat’. The more optimistic view of futurist Ray Kurzweil is that AI can help us to make ‘major strides in addressing the [world’s] grand challenges’.
The truth is, it all depends on how we manage the transition to an AI economy. To fulfill the promise of AI as a new factor of production that can reignite economic growth, relevant stakeholders must be thoroughly prepared — intellectually, technologically, politically, ethically and socially — to address the challenges that arise as AI becomes more integrated into our lives. A good starting point is understanding the complexity of the following issues.
There are three PREPARING THE NEXT GENERATION FOR THE AI FUTURE. things we need to do to create the AI workforce of the future: accelerate the re-skilling of employees; unlock human potential; and strengthen the talent pipeline. These actions will enable leaders to build on a workforce that is already highly engaged with new technologies in their daily lives. And these leaders will reshape their organizations to allow workers to flourish in an AI economy in a way that drives real business value as well as innovation and creativity.
An example of using AI to power re-skilling is Montreal’s Erudite AI, which is tackling the human issue of academic and career stagnation due to a lack of productivity and learning. Erudite uses AI to augment human collaboration and knowledge sharing at work or school. Unlike other such tools, its knowledge management system enables individuals to amplify and share their expertise through the power of human collaboration. It optimizes knowledge transfer and skill augmentation by mapping — in real-time — unique knowledge and skill profiles of learners and matching them with the right expert at the right time. It also provides coaching and collaboration among experts within the platform to instantly enhance the quality of responses. Ultimately, companies must make radical changes to their training, performance and talent acquisition strategies. Re-skilling should be viewed as a new way of thinking about continuous education, as one third of the skills that will be required in three years are not yet considered crucial.
AI will be instrumental in not only making existing workers more productive, but also in helping them deliver better work. This involves fostering a culture of lifelong learning, much of it enabled by technology, such as personalized online courses that replace traditional classroom curricula and wearable applications such as smart glasses that improve workers’ knowledge and skills as they carry out their work. Success will also depend on partnerships with start-ups, universities and individual experts to access knowledge and skills at scale.
In preparation for the AI economy of the future, countries need to do better in aligning their education systems with the needs of the new economy and forging partnerships between institutions and industry. This means enhancing primary and secondary programs, college programs and undergraduate programs with content in critical thinking, creativity, math, robotics and human-machine interaction, as well as continuing to
grow post-graduate programming. This will require extending the learning cycle beyond traditional timeframes and into the workplace.
Last fall, Quebec’s Quartier Innovation and the École de Technologie Supérieure announced a partnership with Vidéotron and Ericsson to create an ‘open laboratory’ for smart technology to unite the telecom and manufacturing industries, municipalities and advanced learning in the creation of new technologies. Governments are also beginning to understand the importance of collaboration in AI. Earlier this year, the governments of Ontario and Quebec signed a memorandum of understanding (MOU) to work together to foster AI development. The MOU aims to keep Ontario, Quebec and, more generally, Canada competitive among other jurisdictions both in the expansion of fundamental knowledge and in the widespread development and application of these technologies. One of the important ways they want to achieve this is by bolstering ties between research and industry and between technology companies and start-ups.
As autonomous machines ENCOURAGING AI-POWERED REGULATION. take over traditionally human tasks, current laws will need to be revisited. For instance, the State of New York’s 1967 law that requires drivers to keep one hand on the wheel was designed to improve safety, but may inhibit the uptake of semi-autonomous safety features, such as automatic lane centralization. In other cases, new regulation is called for. For example, though AI could be enormously beneficial in aiding medical diagnoses, if physicians avoid using these technologies, fearing that that they will be exposed to accusations of malpractice, this uncertainty could inhibit uptake and hinder innovation.
AI itself can be part of the solution, though, creating adaptive, self-improving regulations that close the gap between the pace of technological change and that of regulatory response. For example, AI could be used to update regulations considering new cost-benefit evaluation.
Intelligent systems are rapADVOCATING A CODE OF ETHICS FOR AI. idly moving into social environments that were previously exclusively human, opening up ethical and societal issues that could slow AI’S progress. These range from how to respond to racially-biased algorithms to whether autonomous cars should give preference to their driver’s life over others in the case of an accident. Given AI’S rapid growth, policymakers need to ensure the development of a code of ethics for the AI ecosystem and ethical debates need to be supplemented by tangible standards and best practices in the development of intelligent machines.
Many people are conADDRESSING THE REDISTRIBUTION EFFECTS. cerned that AI will eliminate jobs, worsen inequality and erode incomes. This explains the rise in protests around the world and discussions taking place in several countries around the introduction of a universal basic income. Policymakers must recognize that these apprehensions are valid. Their response should be twofold.
First, policymakers should highlight how AI can result in tangible benefits. For instance, an Accenture survey highlighted that 84 per cent of managers believe machines will make them more effective and their work more interesting. Beyond the workplace, AI promises to alleviate serious global issues such as climate change and poor access to healthcare. Benefits like these should be clearly articulated to encourage a more positive outlook on AI’S potential.
Second, policymakers need to address and pre-empt the downsides of AI. Some groups will be affected disproportionately by these changes. To prevent a backlash, policymakers should identify the groups at high risk of displacement and create strategies that focus on reintegrating them into the economy.
Increases in capital and labour are no longer driving the levels of economic growth that the world has become accustomed to. As indicated herein, a new factor of production is on the horizon. AI promises to transform the basis of economic growth for countries around the world.
To avoid missing out on this opportunity, policymakers and business leaders alike must work towards a future with artificial intelligence. They must do so not with the idea that AI is simply another productivity enhancer; rather, they must see AI as a tool that will transform our thinking about how growth is created.