A visitor touches a humanoid robot hand on display at an AI exhibition booth during the The World Artificial Intelligence Conference & High-Level Meeting on Global AI Governance, in Shanghai China, on July 4.Andy Wong/The Associated Press
Neil Desai is an executive and board member in the technology sector and previously served in senior roles with the government of Canada.
The magnitude of Canada’s economic challenges warrants bold solutions. Earlier this year the Bank of Canada sounded the alarm on our meagre labour productivity and low levels of business investment. Increasingly, the reflexive answer has been artificial intelligence (AI). There likely isn’t a board, management or cabinet meeting where the acronym isn’t dropped as a catch-all solution to a company’s profitability pursuits, the government’s plan to address its own fiscal woes or Canada’s multidecade productivity decline.
But generative AI’s current business model is inherently problematic. It could enable short-term productivity gains for organizations but at a potentially great long-term cost which is more than what Canadian executives, board members and cabinet ministers are accounting for.
This is understandable given that the most common generative AI solutions used personally, such as OpenAI’s ChatGPT, are presented as free to individuals. They are, in fact, costly for the proponents, for whom it’s a worthwhile trade-off. The AI companies are making companies gradually comfortable with adopting the new technology and training large models that generative AI is dependent on.
The “hyperscalers,” Microsoft, AWS and Google, which own a large share of the AI processing power, have to make large investments to keep up with global demand. The investment required to train large language models and process generative AI solutions, includes specialized chips, cloud servers and secure facilities as well as the substantial energy costs to power them. These long-term investments are leaps of faith that if you build them, they will come.
Microsoft, for example, reported $14-billion in capital investments, up almost 80 per cent year over year, in its last quarterly filing. To put that in perspective, their overall revenues only rose 17 per cent over this period. These investments were justified to shareholders given global demand for AI well exceeds supply at present.
The hyperscalers are creating comfortability, and perhaps even dependency, with generative AI solutions. They are incentivizing adoption at scale through discounted initial pricing as well as technical and commercial support in exchange for multiyear commitments.
If we consider how technology companies have traditionally operated, it’s not hard to understand their business strategy.
Imagine a hypothetical health diagnostics company that wants to use generative AI to help clinicians better diagnose patients for rare diseases. They build their application on top of Google Cloud using their generative AI solution, Gemini, because they offer the best deal: $50 per unit of data on a three-year contract.
That deal comes with a wide range of valuable technical resources and go-to-market support. As the three years pass, the application continuously improves and gains widespread adoption. But at renewal time, Google says the price is now $100 a unit.
The choices are 1) go to another vendor, however, this requires a complete re-architecting of the application and starting over on training the models or 2) continue to work with the existing vendor but lose money and try to pass the substantial cost increases to the customer.
Canada, as a result of our largest organizations’ long-term, superficial approach to innovation, is at risk of falling into such an AI dependency trap in the name of addressing our productivity decline.
Our country represents an untapped market for the hyperscalers given our productivity woes and low AI adoption levels. Statistics Canada reported that almost three in four Canadian businesses have not considered using generative AI.
The government of Canada is doing its part to accelerate adoption and contain costs by increasing the supply of AI processing power. They recently announced $2.4-billion for AI-related initiatives with the majority going to processing infrastructure. The market intervention further incentivizes companies to use AI in the short term, adding more uncertainty to the sustainability of their business models.
More concerning for Canadians and our largest organizations is that generative AI seems to be a general solution seeking out specific problems given its current hype. There is a myriad of innovative solutions that have been overlooked by our largest organizations ranging from the cloud to automation data analytics and even process improvements that require little to no technology. Even in the AI spectrum, machine learning has the potential to substantially improve productivity at a fraction of the cost of generative AI.
This is not to say generative AI doesn’t have a place in Canada. There are certainly transformative applications in the public and private sectors that should be explored. But leaders shouldn’t be predetermining the tool in the tool kit to use without a full understanding of the acute problem being solved and all the options to address them alongside their cost and risk profile.