AI has burst onto the scene, infiltrating society as both a powerful tool and an irresistible toy. Among these AI marvels, ChatGPT has stolen the limelight, captivating the masses. Mind-boggling reports reveal that only in January, ChatGPT was summoned an astonishing 590 million times (Carr 21.3.2023).

But let’s not be blindsided by the allure of AI. Its resurgence brings forth a host of challenges. Behind every captivating image, thought-provoking opinion, and insightful essay lies a voracious appetite for electricity, fueling the rise in CO2 emissions from AI technology.

Bloomberg’s revelations are even more alarming, as they expose AI’s insatiable energy demands, surpassing any other form of computing. Training a single model consumes more energy than a hundred American households consume in a year (Saul & Bass 9.3.2023).

And believe it or not, Bloomberg’s estimate might be conservative, merely scratching the surface of AI’s energy gluttony.

It is easy to forget your own carbon footprint

Amidst this digital frenzy, it’s all too easy to overlook our own carbon footprint and turn to ChatGPT for answers. Yet, in truth, we remain oblivious to the energy consumption entailed by each task it undertakes. We don’t really understand yet how much it consumes.

A member of the Danish Data Science Community’s Board, Kasper Ludvigsen, takes us even deeper into the rabbit hole, speculating that ChatGPT’s energy consumption in a single month outstripped that of 175,000 individuals, based on the average energy consumption of Danish citizens (Ludvigsen 1.3.2023).

Some estimates even go further, painting a graver picture. Ludvigsen himself has also revised his figures, hinting at a consumption level potentially soaring into the millions of kilowatt-hours per month (Ludvigsen 6.3.2023).

As I delved into this issue, it forced me to confront my own behavior in the digital realm and the sustainability of my actions. I stand as a testament to the fact that even those who consider themselves environmentally conscious and well-informed can easily overlook their own carbon footprint and its dire consequences when it comes to technology.

It’s important to remember that for traditional information searches, we can still rely on regular search engines or open online encyclopedias like Wikipedia. Furthermore, we should question the ethics of engaging with electricity-hungry entities like ChatGPT. Perhaps machine learning models are better suited for purposes other than finding cake recipes.

AI also needs transparency

Despite its rapid expansion, the realm of AI struggles to keep pace with transparency. Currently, accurately measuring the energy consumption of machine learning models remains an enigma. The emissions stemming from AI development companies fluctuate, influenced by the energy sources they rely upon.

Data centers fueled by coal or natural gas spew significantly more pollution into our already burdened atmosphere compared to their solar or wind-powered counterparts. Numerous researchers express grave concerns about the opaque nature of AI and the impending threat of uncontrollable energy consumption in the near future (Saul & Bass, 9.3.2023).

Take, for instance, the case of Google, where AI gobbles up a whopping 15% of the company’s total electricity consumption (Patterson 15.2.2022). Although this percentage seems modest in the grand scheme of tech giants, it underscores the pressing need for enhanced transparency.

While Alphabet, Google’s parent company, sets an example by disclosing its energy consumption, few companies follow suit. Wired magazine resonates with this hidden reality of generative AI competition. As computational power and demand surge, so too does the imposing carbon footprint (Stokel-Walker 10.2.2023).

Unless we rein in the rampant energy consumption of AI, legislation may eventually need to step in and play its part. It remains vital to remember that governments and countless corporations have pledged their commitment to achieving carbon neutrality in the years to come.

In this equation, energy-hungry entities like ChatGPT simply do not fit the bill.

References

Carr, D. F. 21.3.2023. ChatGPT Tops 25 Million Daily Visits. Similarweb Blog.

Ludvigsen, K. G. A. 1.3.2023. ChatGPT’s Electricity Consumption. Medium.

Ludvigsen, K. G. A. 6.3.2023. ChatGPT’s electricity consumption, pt. II. Medium.

Pattersson, D. 15.2.2022. Good News About the Carbon Footprint of Machine Learning Training. Google Research Blog.

Saul, J. & Bass, D. 9.3.2023. Artificial Intelligence Is Booming—So Is Its Carbon Footprint. Bloomberg.

Stokel-Walker, C. 10.2.2023. The Generative AI Race Has a Dirty Secret. Wired.