AVL/AI: AI and the Environment

The environmental impact of AI and particularly the data centers that power the technology has triggered many alarms. A multitude of think pieces approach this topic, but I point to an MIT article as a fairly measured approach. In general, authors are concerned about power usage, water usage for cooling and the carbon footprint of producing the hardware that runs the machinery. Betraying my background of a specialty trade within the construction industry, I’d add the carbon footprint of building out the facilities as well, although we’ll leave alone for the time being the NIMBYism that controversially accompanies the locations of these data centers.

I’ve described myself in the past as a cautious enthusiast regarding AI and its use within industry. I’ve also considered myself a climate ally who recognizes the tension that emerges between protecting environmental concerns and restricting rote human progress. We incur debts against the environment in pursuit of human progress at all costs. Will AI adoption be yet another red mark in the P&L of technological advancement?

There are certainly uses of AI that are purely negative in regard to environmental impact. Whatever the latest social media meme for visually representing yourself as claymation or a cartoon based on the AI knowledgebase of the user, these uses of AI provide little to no economic value but come at a huge expense of the environmental impact of AI. They are frivolous.

There are also environmentally-neutral uses. This may include automated e-mail responses, drafting support and many of the day-to-day office uses increasingly commonplace in business. One individual can now do the work of many. While there are other negative impacts such as the hollowing out of entry-level jobs, environmental impacts are minimal since the work would be done regardless, whether by one person or by multiple.

Using AI, despite the op-ed anxiety, can have positive environmental impacts. For myself, I use AI primarily as a review and information fetching mechanism. At this point I don’t use AI to generate any content; this may be a limitation of my own knowledge of what the technology provides as well as respecting the constraints on protecting IP and security concerns of client information. It mostly reflects, if I’m being honest, a lack of APIs by which the AI providers could integrate within the specialized software that I use.

Using AI in this review role can be an environmentally positive move. Those of us who work in specialty trades and projects know about deployment costs, AKA rolling a truck. Money burns when deploying personnel, tools and vehicles to a site before the costs of material or labor hours of the service even begin.

As a document reviewer, AI has found many issues that might have resulted in extra redeployment or shipping costs. A typo on a cable type could result in having to order and ship new cable to a site. A missed SKU means that the carbon footprint of the misordered product sits on a shelf unused. A piece of equipment in the drawings but not in the bill of materials means another trip back to the shop to reclaim a crucial link. Some of the mistakes that an AI review catches count toward overall profitability and reduced frustration for our teams implementing the install. Further, those gains also benefit the environment, perhaps even greater than the AI data center infrastructure costs spent to earn those gains.

Quantifying these numbers is difficult and likely deserving of university-backed research. We don’t have any hard numbers for how many tokens (the basic unit of AI processing and a proxy for its energy and water costs) equal rolling a truck. But the research question is worth asking: what are the environmental costs of AI usage against the savings, both environmental and monetary, that strategic application can generate?

For now, instinct and anecdotal evidence will have to do. My experience leads me to believe there are gains to be had despite AI’s environmentally damaging reputation. For those of us who want to maintain both environmental consciousness and operational efficiency, the path is straightforward: forgo the frivolous uses and focus on the ones that can be environmentally net positive.

The ledger on AI’s environmental impact is still being written. Despite widespread fears of environmental damage, we have agency to record wins in efficiency that can move the bottom line into the black. Or dare I say, green?