Generative AI has the power to create original content and is often discussed from the perspective of artists' and copyright holders' rights, as well as labor. However, its environmental impact has not been as widely considered.
Let's explore the intriguing insights provided by research on AI's impact on energy and the environment, as it becomes a powerful support for our creative activities.
Energy Consumption Varies by Task
In a study conducted by researchers at AI startup Hugging Face in collaboration with scientists from Carnegie Mellon University, the focus was on measuring the carbon dioxide emissions based on the tasks performed by AI.
The study revealed that simple tasks like text classification emit 0.2 to 0.5g of carbon dioxide per 1,000 queries, whereas image generation, which has gained attention recently, can emit up to 1,000g of carbon dioxide for generating 1,000 images.
In other words, simple tasks like text classification are energy-efficient with relatively low emissions, while complex tasks like image generation significantly increase energy consumption and emissions.
From the perspective of energy consumption, generating 1,000 images requires 2.907 kWh. This may not seem like much, but to fully charge the battery of a Tesla Model 3, an electric vehicle, 50 kWh is needed, which equates to only 17,200 images.
Platforms like DALL-E, Midjourney, and Adobe Firefly are generating vast amounts of images, consuming energy equivalent to charging thousands or even tens of thousands of electric vehicles daily, assuming the same energy requirements as this study.
Not Just This, Hidden Environmental Impacts
AI consumes energy when performing tasks, but it requires even more energy for model training and deployment.
While calculating the cost of model training is said to be difficult, it is known that energy consumption is exponentially higher than performing tasks.
For instance, training the GPT-3 model, which uses 175 billion parameters for the increasingly popular ChatGPT, requires 1,287 MWh.
Creativity Generated by AI and Humans
When considering AI from the perspective of environmental impact, there are indeed issues such as energy consumption and carbon dioxide emissions.
However, the benefits brought by AI are significant, such as enabling image editing and synthesis with just text instructions, without the need for specialized knowledge or advanced skills.
The future development of AI may depend on how we choose to engage with it.
