The sustainability challenges raised by AI

While a primitive concept of artificial intelligence (AI) can be traced back to the Bell Laboratories in the 1950s, it was not until this year that many of us experienced its potential for the first time, with the launch of OpenAI’s ChatGPT and other large language models. In simple terms, the simulation of human intelligence by machines requires computer programs that are trained on large amounts of data to infer or problem solve with minimal human intervention. As we embark upon the next wave of technological innovation, we reflect upon the balance of ESG risks and opportunities on the horizon.

Climate Change

Artificial intelligence may be the next frontier for fighting climate change. A recent study from the World Economic Forum[1] concludes that digital technologies, such as AI, can reduce greenhouse gas emissions by up to 20% by 2050 in the three highest-emitting sectors: energy, mobility and materials. In brief, AI can be used to better track and report greenhouse gas emissions, improve circularity and reduce emissions. A tangible example of this would be Johnson Controls International (JCI), which can deploy AI across its heating, ventilation, and air conditioning (HVAC) control systems to deliver significant improvement in building energy efficiency. The company’s suite of OpenBlue digital solutions allows customers to reduce operating costs and carbon emissions and improve indoor air quality. Using AI, HVAC systems can respond to data from internal sensors monitoring temperature and humidity and combine that with additional data, such as weather forecasts, occupancy levels, energy source and energy cost to optimize equipment efficiency.

Energy efficiency is especially relevant when considering that data centers running AI training and inference models are expected to consume an increasing amount of energy, not just for processing data but also for cooling equipment. Indeed, a peer-reviewed analysis[2] published last month lays out a range of scenarios around the growing energy footprint of artificial intelligence. In a base-case scenario, AI servers could use between 85 to 134 terawatt hours of energy annually by 2027, which represents around 0.5% of the world’s current electricity use and approximates what Argentina, the Netherlands and Sweden each use in a year.

 Jobs and Workforce

While initially disruptive, industrial or technology revolutions have historically translated into an overall growth in employment opportunities. As evidence, a recent study by MIT economist David Autor indicates that 60% of today’s workers are employed in occupations that did not exist in 1940 – implying that 85% of employment growth over the last 80 years can be explained by technology-led creation of new jobs. Viewed differently, AI could be the new demographic that supplies the labor shortage caused by slower population growth and aging populations around the world. Unlike prior technological innovations, which have historically disrupted blue-collar jobs, AI is likely to impact white-collar jobs – several studies have identified knowledge roles in the administrative, computer, mathematical, business, design and media domains as most likely to be impacted. As noted earlier, history would suggest that AI is unlikely to result in a reduction of aggregate employment, but rather lead to the creation of new roles. As such, what we are focused on is the corporate response to these workforce changes with respect to upskilling resources for displaced workers.


The responsible phasing out of human judgment in favor of AI models carries significant risks associated with biased inputs, inaccurate results (model hallucinations), accessibility, data security and privacy and cybersecurity. Since AI models are trained on human-generated data, we run the risk of perpetuating biases that could have meaningful ramifications for a range of end-use applications, such as financial (i.e. biasing lending practices) or legal (i.e. biasing legal outcomes). Furthermore, the costs associated with training AI models today are prohibitively expensive, which concentrates access to this powerful resource into the hands of large and resourceful developers that may broaden access to these tools in limited ways. Looking ahead, we await clarity on the regulatory framework in the United States, where several government agencies are working on finding a balance between encouraging innovation and mitigating the potential harms of the use of AI without guardrails, including political polarization, privacy violations and social inequity.


[2] Joule, The Growing Energy Footprint of Artificial Intelligence, Alex de Vries, October 10, 2023