The recent explosion of data center demand, including, and especially, in the Midwest, has been primarily driven by the growth of Artificial Intelligence (AI). These data centers have served as a distinct reminder that technology and the internet is a physical place, an embodied resource. They are raising concerns over power supply and water use – large language models, like the ones used to power generative AI, can take 10x more energy than traditional search engines, and just 15 queries with ChatGPT could evaporate and consume a glass-worth of water. The rapid growth of data centers has led utilities in the RE-AMP Network footprint to extend the life of coal plants and propose new gas plants, getting the network farther from our North Star goal. 

The AI boom is being driven by trillion-dollar tech companies that, in many cases, have integrated AI into their products even without clear market signals, and with only dubious benefits to the product – take, for instance, Google’s “AI Overview,” a feature rolled out in May 2024 to appear automatically before search results and was repeatedly shown to provide misinformation at the time of its rollout (some of which was subsequently fixed, although inaccuracies persist and now there are also ads). In the face of a wave of new AI offerings and the consolidation of power by technology billionaires with control over major internet platforms who stand to profit from the growth of AI, the trajectory of AI adoption can feel inevitable. But AI is still at a critical juncture, where institutions across the country are now deciding how and how much to use AI. We recognize that individual or even collective institutional action on the consumption side is not nearly sufficient to address issues of energy and material resource use. But as resources are still consumed on the individual and institutional level, we also recognize the impact that our movement can have collectively.

Our decision-making around AI use also cannot be limited to the material impacts of our own queries, as each query feeds back into the system, helping support the larger, growing ecosystem. While some AI offerings have features to switch on privacy mode or opt out of being used for training, not all ask for consent, and practices are not always transparent.

As we consider how to approach the impact of this technology on our work and on our world, we have landed on 7 key considerations to consider before adopting AI into our workflows:

• Security: AI may store sensitive data, which could be at risk of being breached or hacked. All AI tools must stay contained to their intended scope, and access to sensitive data – including names, addresses, and passwords – must be avoided. AI software can not only process information intentionally entered, but can gain access to a wider variety of personal information. This lack of privacy is, perhaps, already built into our social awareness – ”my TV is spying on me” is now, practically, a given. AI accelerates those concerns, and also provides a tool to easily process and monitor the information, leading to a higher level of surveillance.
• Sustainability: As detailed above and in extensive reporting, these technologies rely on considerable material, water, and energy consumption. AI has the unprecedented ability to create digital waste – slop of no (or negative) value that not only takes resources to produce, but may also be stored indefinitely in data centers’ rapidly expanding footprint.
• Confidentiality: Information input into any AI may enter the public domain. This can release non-public information, breach regulatory requirements, and compromise information and strategy critical to the movement. The companies creating these technologies have shown themselves to be willing to use user data to train their models, thus exposing that data in the output of the AI.
• Accuracy: AI is a predictive tool, relying on complex algorithms to generate content from strings of most-likely words. This means it may provide inaccurate, unreliable, or misleading information (also known as “hallucinations”). AI models may have difficulty providing real references or assessing the quality of sources. The ease of using AI may encourage people to accept outputs that are “good enough” – and because they require little to no effort, these outputs can proliferate exponentially. The resulting harm is therefore not only the production of clear misinformation which a human could hypothetically fact-check, but also lowered standards, and generally muddied waters. As more and more data is synthesized and output is made using AI, the algorithms will be trained on a larger portion of AI’s own output, heightening those concerns. Even in the best AI summaries that only replace busywork, its use demands that we contend with the value of human labor, creativity, knowledge, and thought.
• Bias: All AI should be assumed to carry bias. Due to the nature of training data, AI may produce biased, discriminatory, or offensive content. AI can synthesize vast amounts of data, and can be better than humans at pattern recognition. But when trained on existing data, it can also be even better than humans at reproducing existing disparities, without the same accountability (such as in housing or policing). When augmenting human decision-making with AI, AI’s fundamental lack of accountability can be particularly concerning, and may amplify pre-existing biases in its dataset, and in society at large.
• Equity: In addition to equity concerns in sustainability, bias, and democracy, much of AI training data has been sourced from individuals without their consent, and may legally be considered copyright infringement. AI outputs compete with and direct traffic away from human labor. As a result, generative AI has been especially noteworthy for taking work opportunities and deflating wages for creatives. In the optimistic view of AI, the tools can increase equity by making a wider variety of labor accessible and lead to improvements for lower-skilled workers. However, AI’s widespread adoption could also contribute to widening the wealth gap and accelerating existing disparities.
• Democracy: The popularization of AI has amplified threats to our democracy, including through misinformation, the creation of fake “bot” accounts, and voter suppression. The increased adoption of AI into government and everyday decision-making, and outsourcing of governance to poorly understood algorithms would erode democratic processes and our opportunities for democratic engagement.

Building on these guiding considerations, we are implementing our own AI policy for RE-AMP staff and at RE-AMP meetings. We are sharing the policy with the network both as a way to set an expectation for our meetings going forward, and in the hopes that it may provide a useful example for your own work.