Commentary: How steep are AI's social and environmental costs?

Generative artificial intelligence, or GenAI, keeps making headlines. Every day, we hear more success stories about how GenAI is transforming sectors such as health care, infrastructure, business and commerce, education and research. We can applaud this technological progress, but is there anything we should be concerned about?

Much has been said about how GenAI systems can threaten the integrity of research, de-skill the workforce, redefine our jobs and abuse our personal data— but these are potential consequences. Regardless of where and how GenAI is employed or what the consequences might be, the development and deployment of these models come with enormous social and environmental impacts — a topic that does not receive enough attention.

Here are three reasons why concern is called for.

First, GenAI requires hardware known as graphics processing units, or GPUs. Nvidia, a multinational corporation based in California, is the leading producer of GPUs. As an example of its impact, Nvidia is building a new production facility in Taichung, Taiwan, a city comparable to Chicago in population, that, once completed, will consume nearly 25% of the city’s electricity and 6% of its water. These GPUs will have to be housed in data centers. In 2022, data centers consumed 1% to 1.3% of global electricity demand. This consumption is projected to grow by 160% by 2030, leading to an estimated 2.5 billion metric tons of carbon dioxide emissions.

In a recent development, Microsoft and G42, United Arab Emirates’ top AI firm, agreed to jointly invest $1 billion in building a data center in Kenya’s Olkaria region — a well-known geothermal hot spot that can provide affordable energy for the data center. Nevertheless, locals remain concerned about the pollution caused by geothermal energy production and its negative impacts on air, water and soil quality. The data center itself is expected to have negative impacts on the physical and psychological health of nearby communities, based on reports from places such as Chandler, Arizona.

Second, GenAI needs to be trained by data and humans who should label the data and supervise the process. Thousands of workers who train commercial GenAI models are based in low- and middle-income countries, where they are underpaid and work in conditions often described as modern-day slavery.

Furthermore, significant numbers of people are required for maintenance, prompt engineering and validation of GenAI outputs. These efforts require subject matter experts who define and validate the quality of GenAI outputs. Ongoing maintenance, such as debugging and updates, is also a continuous process that adds to the human hours needed to keep these systems operating. Such extensive human involvement highlights the often-overlooked costs and complexities associated with deploying and sustaining GenAI technologies.

Third, GenAI development and deployment need huge sums of money. According to estimates, for a midsize enterprise to initially deploy a GenAI system, $80,000 to $190,000 of investment is needed for hardware, development and data preparation costs. This is excluding the $5,000 to $15,000 for annual maintenance and ongoing costs. And yet, this would still be cheaper than using commercial models such as OpenAI for business purposes. Some experts even believe that the costs of many current and future GenAI developments may exceed the value these systems add to workflows.

The environmental impacts of developing and employing GenAI are immediate and irreversible, but the human and financial burdens need more time to manifest themselves. In the short term, financial and human resources will have to be diverted from other critical projects, which will shift priorities in companies, industries and the public sector. This redirection could stall or slow down initiatives aimed at corporate social responsibility and workers’ welfare in the private sector or education and health care in the public sector — areas that may be more immediately beneficial to society at large.

The long-term implications could go either way, though. Should investments in GenAI pay off, societies could save money through greater equity and efficiency and increased rates of automation and innovation. These savings could potentially be invested in socially responsible or environmentally friendly initiatives. However, if they do not pay off, we will have collectively made a small group of high-tech developers and multinationals richer and more powerful, effectively shooting ourselves in the foot. In that scenario, the wider social benefits would remain unrealized, leading us to question the true value of GenAI advancements and ask ourselves: Were these innovations truly beneficial for all? As some experts have warned, “new technologies have always worsened disparities,” and GenAI might be no exception.

Even if we give GenAI the benefit of the doubt, navigating this period remains extremely challenging. It’s a time when the focus and attention given to GenAI may overshadow other pressing social needs. For this reason, we need to pay more attention to social and environmental impacts and push those who are considering GenAI development and deployment to also include these impacts in their assessments.

So where do we go from here? It is essential to approach GenAI with caution and awareness. We should work with organizations that have a clear strategy toward using GenAI and incorporate ethical practices found in recommendations such as the UNESCO Global AI Ethics and Governance Observatory or the National Institute of Standards and Technology’s AI Risk Management Framework.

And finally, let’s remain critical about technology and educate ourselves by following the works of such notable voices in this field as Nick Bostrom, Mark Coeckelbergh, Timnit Gebru and Shoshana Zuboff.

____

Mohammad Hosseini, Ph.D., is an assistant professor in the Department of Preventive Medicine at Northwestern University’s Feinberg School of Medicine. Kristi Holmes, Ph.D., is a professor of preventive medicine and the director of Galter Health Sciences Library at Northwestern’s Feinberg School of Medicine.

___