Imagine you’re an artist. You could spend days working on a masterpiece, pouring your energy and focus into every brushstroke. But what if a computer could whip up something equally impressive in under a minute?
It’s not science fiction; it’s the power and reality of artificial intelligence (AI).
But which is using more energy and impacting our planet more — you with your paintbrush and canvas or the computer?
Carbon Footprints: Humans vs. AI Models
When exploring the actual costs of energy consumption, the numbers tell a predictable story.
Let’s start with the human brain, a marvel of energy efficiency, operating on approximately 12 watts of power.
Contrast this with consumer-grade computers with high-performance GPUs like the NVidia 3090, which can consume around 650 watts when running compute-intensive tasks. That’s over 50 times the energy consumption of the human brain for a single computer setup.
But the stakes are even higher when we focus on the business world.
Consider an NVidia h100 in a data center, which alone demands around 700 watts of power. A supercomputer cluster, powered by an astounding 10,000 h100 GPUs, can ramp up that energy requirement to staggering heights — equivalent to half a million times the power your brain uses.
These figures don’t merely hint at the escalating energy demands of advanced computing; they shine a glaring spotlight on the sustainability problem at the heart of the AI revolution.
We’re collectively faced with a crucial question: How do we reconcile the meteoric rise of AI capabilities with the pressing need for environmental responsibility?
As we race towards an increasingly digital future, we must grapple with the actual costs of our technological appetite for energy.
The Energy Efficiency Dilemma in AI and Human Activity
As the old saying goes, you can’t improve what you don’t measure. However, comparing human carbon footprints to AI models is inherently flawed due to the complexities of defining boundaries in a rapidly evolving area that lacks transparency and accessible data.
Creating an accurate carbon footprint estimate for AI remains challenging without detailed information about hardware, energy consumption, and energy sources.
Researchers from the University of California-Irvine and MIT released a study that has sparked a spirited discussion among top AI professionals. The research challenges preconceived notions about the energy consumption of generative AI models like ChatGPT.
According to the findings, the carbon dioxide equivalents (CO2e) emitted by ChatGPT when generating a page of text are 130 to 1500 times lower than those emitted by a human engaged in the same activity.
The study concludes that AI technology holds promise for accomplishing a variety of tasks while generating substantially lower carbon emissions than human activities.
Why AI and Human Abilities Aren’t Easily Comparable
The conversation about AI’s energy and resource costs versus humans is far from black and white; it’s a nuanced landscape with varying benchmarks. For example, AI systems have long outpaced humans in performance and energy efficiency in some domains like chess.
Yet, this doesn’t provide a comprehensive view of AI’s capabilities across the vast spectrum of human activities.
Imagine a task where an AI model has been trained for a year and performed it billions of times for different people, while a human might have been training for the same task for 30 years but only performed it for dozens. AI might seem more efficient in this specific context, but it’s crucial to remember that humans perform a multitude of tasks that AI can’t handle — yet.
Comparing energy consumption in such cases becomes difficult not just due to these technical disparities but also because of ethical considerations. Humans have lives beyond their “tasks,” and these lives consume resources that are typically not accounted for in a direct comparison.
Sustainability in the AI Age: More Questions Than Answers
The comparison between AI and human resource consumption is fraught with technical and ethical complexities. Currently, AI is optimized for specialized tasks rather than the broad range of activities that constitute human life. So, while AI may appear more efficient in specific benchmarks, it’s not necessarily a straightforward or fair comparison.
On the ethical front, humans have a right to exist and consume resources for survival and well-being, something that’s not typically ‘switched off’ in the way we might power down a machine. Without delving into ethically murky waters, we can’t adequately measure a human’s lifetime resource consumption against an AI model.
As pointed out by AI researcher and climate lead at HuggingFace, Sasha Luccioni:
“You can’t compare the carbon emissions of people and objects. Humans are more than just the work that they do.”
Man, this preprint is really the gift that keeps on giving.
In case people missed my previous PSA : you can't compare the carbon emissions of people and objects. Humans are more than just the work that they do.
(Also, that paper makes a lot of false assumptions in general) https://t.co/bZA414J9YI
— Sasha Luccioni, PhD 💻🌎🦋✨🤗 (@SashaMTL) September 19, 2023
The missing links in data availability severely limit our ability to comprehensively analyze environmental effects. It will take a transparent, science-based approach to overcome these complexities.
The question that confronts us is much bigger than who or what is more energy efficient. We must avoid picking sides and engage with the complexities underpinning our choices.
If AI is to be our future, let it be a future we choose consciously, knowing full well the costs — not just in watts or carbon emissions, but in the very texture of our human experience and the shared responsibility we hold for the planet we call home.