AI & Productivity

We need to approach this with some nuance and some caution.

Photo by Pawel Czerwinski on Unsplash

What impact is AI going to have on productivity? It is one of the big questions facing us. If AI is a general-purpose technology, such as electricity, then there is the prospect of it unleashing huge productivity gains as previous GPTs did in the past. If that bares out, then there is a real prospect that AI will help break the globe out of the rut of declining productivity growth (something by no means limited to Canada). Given the connection between productivity and living standards, an AI productivity bonanza could be a great thing, provided welfare support and job market impacts are adequately taken into account.

However, that future is far from certain. Despite a short-lived productivity boost, the Internet and other digital technologies have not delivered the growth that many expected of them. While there is much debate about the reasons for this, one argument that has stuck with me is about the two sides of the internet. Advanced by Ezra Klein (in a podcast that is now paywalled, so I can’t find the transcript), he argued that while, on the one hand, the internet represents the total of human knowledge at your fingertips and the ability to instantly communicate with anyone around the globe, on the other, it represents a whole new world of distraction, with business models and algorithms tailored to steal your attention. While all that knowledge might be at your fingertips, it has become easier, not harder, to be misinformed thanks to the digital ecosystems that have been built and how they have been used and abused by individuals, companies, and states.

I can see many parallels with AI already. Sure, AI might be able to boost productivity massively in some cases, but it isn’t going to come without massive downsides, too. A newly published report by Peter Nicholson for the Public Policy Forum sets out a few positive use cases:

• Processing data and information of virtually any kind.

• Augmenting the productivity of software engineers by significantly increasing output volume without sacrificing quality.

• Boosting productivity in manufacturing and in goods production generally.

• Improving marketing strategies, for example preparation of promotional materials (text, image, video) or micro-targeted customer identification and inducements.

• Enabling widespread, high-quality language translation implemented in real time with voice synthesis (including via a smart phone app).

• Improved services and capabilities in the financial sector.

• Enriching and personalizing education from the early years through adult learning and job training.

• Boosting productivity across the health-care system.

• Improving the quality and efficiency of government services.

• Equipping various kinds of robots with the capability to perform flexibly in unstructured environments

• Amplifying innovation itself.

If AI lives up to its promises, then these could be transformative. However, there are plenty of other use cases that aren’t looking so great for society’s productivity or general well-being. These include:

• The rapid proliferation of AI-generated election misinformation.

• The utilization of AI by cyber criminals for sophisticated phishing/social engineering attacks and voice/video cloning scams.

• The rise of deepfake ads exploiting names, images, footage, and voices of celebrities.

• The potential for the misuse or overuse of AI in education to harm students’ critical thinking and problem-solving capabilities and the risk it diminishes human interaction in the learning process, leading to a loss of social skills and interpersonal development.

• The amplification of bias, skewing predictive healthcare algorithms, producing biased results in job applicant tracking systems, reinforcing gender bias in search engine ad algorithms, producing biased images of people in specialized professions, and reinforcing existing patterns of racial profiling in policing.

• The use of AI in war for targeting and to enable domicide.

• The risk that AI traders lead to herd-like behaviour that threatens the stability of financial systems.

• The potential for AI to undermine scientific research by producing ever larger numbers of papers of dubious quality, perpetuating a research system that prioritizes quantity of papers over quality.

In addition, AI’s extensive use of water and energy might accelerate climate change and have negative environmental impacts, far outweighing the potential for AI to help on climate.

This is all without necessarily accepting the argument that AI poses an existential risk to humanity or probing its impact on inequality, which is intimately linked with productivity. 

We shouldn’t be blinded by the prospect of AI as a panacea for our productivity problems, especially given recent research by the Dais on (pre-ChatGPT) AI adoption by Canadian firms. They find no significant evidence that total factor productivity has risen in firms that have adopted AI. As I’ve argued before, I don’t think it is a given that Canadian companies will deploy AI in a way that actually helps productivity.

We need to approach AI with our eyes open. Yes, it has the potential to be an incredibly powerful tool for the economy and societal good, but it also has the potential to unleash a lot of harm. Chasing productivity mirages should not be the North Star of our AI policies.