As the Super Bowl approached and passed, it seemed that one faction of Americans was accusing Taylor Swift of practicing witchcraft on the NFL while another was slagging her for the carbon output of her private jet—reportedly about 8,300 tonnes of CO2e in 2022. And although it is fair to expect owners of private aircraft to fly responsibly, I must ask this:  What is the environmental value of not shitposting about Taylor Swift? Or for that matter, any number of topics?

The carbon cost of a single tweet is ~.026g; the cost of X (nee Twitter) is estimated at 8,200 tonnes per year; and the overall carbon cost of social media is estimated at 262 million tonnes of CO2e per year. So, if we use this social media carbon calculator, it tells us that 1 million people spending just 2 minutes a day on the 10 major social sites costs just over 8,300 tonnes of CO2e per year—roughly the same amount T Swift reportedly generated with her airplane in 2022.

I recognize that this is comparing the carbon footprint of one individual to a million individuals, but that one individual entertains millions and generates economic activity. By contrast, the social posts of a million people at any given moment are only making pollution in every sense. Clearly, it costs metric tons of carbon to produce metric tons of useless noise. And that preamble brings us to the topic of the projected increase in electricity demand for data centers to support advancements in artificial intelligence (AI). As Bloomberg reported in late January:

Electricity consumption at US data centers alone is poised to triple from 2022 levels, to as much as 390 terawatt hours by the end of the decade, according to Boston Consulting Group. That’s equal to about 7.5% of the nation’s projected electricity demand. 

In past posts about generative AI, I have opined that we do not need machines to make creative works—because we don’t—and that AI should be tasked with solving problems like curing disease or mitigating the climate crisis. On the second point, however, it seems that if an AI were asked the climate question, its only rational answer would be, “Shut me down.” If nothing else, AI could be an environmental catastrophe in the making.

“In the Kansas City area, a data center along with a factory for electric-vehicle batteries that are under construction will need so much energy the local provider put off plans to close a coal-fired power plant,” the Bloomberg article states. Because that quote cites both electric vehicles (EVs) and the data center, one must acknowledge that the environmental analysis of EVs entails a projection of carbon saved against carbon spent. But because a data center is pure carbon expenditure, that cost can only be measured against the value of the activity the center supports.

No question that data centers are infrastructure. There is no enterprise—private or public—that does not rely on networked computing, and economic activity almost always presents an environmental challenge, whether one is building a railroad or an eCommerce platform. But considering even the current energy demand, let alone the projected increase, AI pulls the issue into focus because so many of its applications are already either useless or toxic.

Useless, as stated, is the AI that generates “creative” work in lieu of the human creator, while toxic would be something like more advanced deepfakes exacerbating the disinformation crisis. Regarding the former, this flips the economic equation—i.e., carbon cost yielding lost jobs, which is arguably the opposite of economic activity. Regarding the latter, the use of AI to expand and deepen disinformation campaigns represents carbon cost in exchange for “better tools” that have already been used to weaken democracy worldwide.

In 2013, I wrote a post called Show Me the Innovation—one of many responses to the generalized argument that legal frameworks designed to protect intellectual property, privacy, information integrity, and even personal safety all stand in the way of “innovation.” The point then, as now, is that not everything produced by Big Tech is “innovative,” if we insist that word mean something. If “innovation” should improve lives and foster prosperity, isn’t it curious that social media’s carbon cost helps support anti-science agendas like climate change denial?

In a recent post about the environmental cost of data centers, Chris Castle cites Science Daily, noting that “generative AI like ChatGPT could cost 564 megawatt-hours (MWh) of electricity a day to run.” That’s more than some small countries. When coupled with the fact that data center demand is halting planned shutdowns of coal-fired plants, then it starts to look a lot like AI is helping to “innovate” the U.S. backwards, reversing the gains made over the past twenty years in carbon emissions.

Traditionally, it is possible to do a cost/benefit analysis. We burn x amount of coal to power y number of homes, or we need x amount of oil to run y amount of ground transportation. And even in the earliest days of electrification or automobiles, the benefits were self-evident. But with rapid advancements in AI, the cost is rising without clear evidence of benefit—at least not at the scale the electricity demand implies. This is because, like so many “innovations” of Big Tech, AI might be used to accomplish something extraordinary like improving medical diagnoses, but in the meantime, it will be used make what is already bad about digital life suck faster.

Photo by: dropthepress