Even before the pandemic hit, the adoption of artificial intelligence (AI) was gaining momentum with automated driving as the showcase application of its capabilities and disruptive power. With market tailwinds such as the increased availability of VC funding (especially in the US), the sector seems ripe to transform from being an experimental technology to becoming mainstream. Canada is targeting the AI sector by providing government funding under its superclusters program. Scale AI has received $230M from the Canadian government and $50M from Quebec. Canadian universities have some of the best researchers in specific areas like deep learning and Canada is among the top 10 countries by published journal articles in the field of AI. Five Canadian startups also feature in the list of top 100 startups in AI (a list dominated by the US-based companies). There has been discussion on how to regulate AI in Canada and about responsible use of AI in government. With both technology and market forces changing so fast, it is often hard to imagine how the future of AI will look once everything is settled.
Confidence in AI and use of data as fuel
AI as a technology has been in use in business and consumer applications for years without many of us noticing. There is no universally agreed upon dividing line between what qualifies as AI versus intelligent automation. Software algorithms and models have been guiding us in things like which route to take at what time to make deliveries, who should pay what interest rate, or who gets stopped at the airport for additional screening. Until now many of these recommendations were being sold and used as “decision support systems”. Even in chatbots, one of the most used AI technologies, there would be a “human agent” who would use the software to find out the answers to customer queries. But with enough data and cheap computing power, it has become possible to remove humans from the interaction altogether. This confidence in AI and how it evolves over time will determine the future of AI. This confidence will vary across applications as well.
It is important to remember that the precision and accuracy of any AI predictions are highly dependent on data used in training. The fiasco of a world-leading software company’s AI tagging a photo of a Black couple as gorillas is a good reminder of the importance of this crucial dependency on data. Therefore, the companies and countries that have access to usable data or who provide and sell products and services to refine raw data into model-usable data will have a competitive advantage. Data is the new oil.
Canada doesn’t have the advantage of a large population (like India or China) that can help it gain access to or control over the data that could be used for training the AI models for future applications. Therefore, it would be better if Canadian companies focus on developing unique, IP-protected tools that could be used by everyone around the world, especially in the areas where there is not yet much data.
One potentially rewarding area could be climate action. Climate AI is an emerging field with a few Canadian companies offering software to plan, track and positively improve any business’ environmental impact. Most of the current AI work seems to be focussed on manufacturing, construction, smart cities, disaster planning, and mitigation efforts, since that is what seems to have the most value-for-money. But with more funding available and climate events becoming more severe and frequent, we are likely to see AI being used to find the best way forward in other areas as well.
This could include:
- AI that enables companies to manage and improve their Environmental, Social and Governance (ESG) targets, and deliver on their Corporate Social Responsibility (CSR) commitments
- AI that helps governments implement their policy plans efficiently, track indicators, and monitor progress at different levels
- AI that guides civic society (individual users) to contribute not only through responsible consumption, but also through their work, investments, votes or any other actions they could take
Intent, externality and human-assisted AI
Like any technology before it, AI is a tool that serves a purpose and any AI strategy should be based on its final usage. But this is where things get a bit tough. What is best for individual businesses might not be best for the aggregate or the entire economy. For example, the purpose of using AI could be:
- To replicate what employees do
- To enhance what employees do
- To do things employees cannot do
- To develop a competitive advantage
Depending on the industry structure and the technology, one of these intentions may be the main one or only one. But if, for example, every industry started developing labour-replacing AI, it could become a vicious cycle: one company’s firing of employees would reduce another company’s product demand, leading the second company to think about investing in labour-replacing AI.
As with any other externality, the government would likely consider whether taxing the AI could be an alternative to minimize this disruption. A few industry leaders, like Bill Gates, have already brought this suggestion forward. But Canada should consider the impact on this nascent industry carefully before rushing to such a measure. The carbon tax was necessary to reduce the overall demand of climate-change-causing fossil fuels, but there is nothing inherently negative about the building and using of AI applications. Unlike other externalities, the relationship between AI and labour is still evolving, and we (industry and the Canadian government) can still control AI to ensure that it is a productivity-enhancing technology.
“Canada can take the lead and set up a framework and infrastructure for AI validation and approval.”
For example, since AI becomes better at predicting things as more people use it, companies will be willing to pay for what is called “supervised-learning.” The government can make it mandatory as part of product licensing to determine whether the AI is proven to work without human supervision. Canada can take the lead and set up a framework and infrastructure for AI validation and approval. Following the framework used for the pharma industry, we could have a regulator or an industry consortium that would work together with AI software companies in identifying:
- The set of actions that an AI algorithm is taking that could impact humans
- The data that was used to train the AI
- The representativeness of the data and the prediction results
- The applicability of the AI to broader users and for other usages
AI inputs vs AI outcomes
Despite the buzz around AI and the amount of capital being invested in the sector, much of it is based on expectations. Once the novelty of the idea wears off and when the true cost of implementation is taken into consideration, not all AI technologies will be beneficial. For example, the cost-benefit analysis of chatbots versus human agents is tilted because the time cost is borne by the consumer. But if a $100/hr analyst must spend hours to figure out how to ask the right question from an AI-based document search engine to identify the relevant ESG compliance checklist, it will be easier to pick up a phone and ask a human expert. Therefore, the focus will eventually shift to return on investment.
Currently there is no way to judge how valuable an AI technology is. In fact, most of the AI industry seems to be navel-gazing. The companies are building platforms that can integrate with other platforms to feed in the data needed to train the models (which is another platform), in turn powering the applications (yet another set of platforms). With the buzzwords like data lake, data stack, MLOps, AI-cloud and more becoming commonplace, it is hard to separate substance from hype. A lot of these companies are being assessed based on their input and not really based on the output. Even though many VC-funded startups, including those in Canada, have slides on Total Addressable Market (TAM), most of these numbers are estimates rather than based on historical data.
Valuing any new product or service is hard to do, especially if there is no direct substitute available. What price point should the companies use? Tesla offers its AI through its auto-pilot feature for free and so does Google through its Google Assistance feature. But whether Tesla can charge a premium compared to other car makers because of this feature would depend on
- How novel the AI is
- How much it would cost for others to replicate it
- What human intervention it is replacing and how much that effort-saving is worth
If Google started charging its users for its voice search service, many would stop using it. Google gets more value from improving their AI models through the data a user generates and therefore would rather give it for free than risk losing the users and having to spend money to improve the models.
“Industrial AI applications in healthcare, manufacturing, smart home, food and agriculture, transportation, supply chain logistics, energy and mining are a few sectors where Canada can take a lead in the future of AI.”
Canadian companies who are willing to operate in the AI-outcome space are more likely to benefit at the other end of the technology hype-cycle. Industrial AI applications in healthcare, manufacturing, smart home, food and agriculture, transportation, supply chain logistics, energy and mining are a few sectors where Canada can take a lead in the future of AI. Instead of focusing solely on historical data, companies who can show the benefits of using the AI in real-world applications or companies that improve their prediction results using evidence will be able to become global leaders. If the Canadian government starts using an AI to do comparative analysis of the outcomes across cities or different departments, it could guide policymakers around the world on which of the several climate change actions are likely to have the most impact.
Using AI to enhance competitive advantages
Canada has one of the highest tertiary education numbers in the world. Our universities produce graduates in a wide variety of fields. A lot of CEOs and even policymakers have been arguing that we need to shift the focus and either change the program composition at the universities or launch retraining programs to make these graduates employable in the technology sector. This is akin to playing catch-up. In the technology space, many countries suffer from US-envy and want to emulate the success of Silicon Valley. This goes against the notion of comparative advantage and gains from trade. It is fine to realign one’s competitive advantage if it is done to move higher up the value chain, but not to emulate others. Since AI applications will be powered by the underlying data, the consumer software leaders like Facebook, Google and Apple or e-commerce giants like Amazon will always have the lead in finding insights into people’s activities. Canada’s Shopify is growing at an amazing rate and hopefully it should be able to craft a good AI strategy driven by the huge amount of data of its merchants and their customers. But Canadian governments and Canadian companies should try to leverage their biggest competitive advantage, which is highly-educated Canadians.
At present, AI is very good at performing routine tasks: looking for information based on keywords, identifying pictures, and following conditional commands . But if it achieves its potential, the future of AI will look very different. Rather than suggesting a patient visit a doctor since a mole could be cancerous, it will suggest that it is time to take a photo to check for any moles. This “actionable wisdom” will come from AI models that will be developed based on the codified knowledge of all industries and all walks of life. Given Canada’s mature industrialization, and the breadth and diversity of its population, it could lead the world in the next evolution of AI applications.
In the future, AI will become a “complementary input” to almost everything. Much like driving a car or using a mobile phone. Everyone will learn it and use it without needing to know the technical details. Canadian companies in different sectors could be generating these “recipes for life” or “blueprints to do anything” developed using human-assisted AI.