MB>CO2 the ChatGPT edition:
A Dystopian Art Experience Unmasking the Hidden CO2 Emissions from AI Models Like ChatGPT"
The MB>CO2 art installation by ecological artist Thijs Biersteker is a captivating and immersive experience that delves into the environmental impact of data usage, particularly focusing on AI models like ChatGPT. With the pandemic driving up data consumption and associated CO2 emissions, the need for sustainable AI practices has become paramount.
The installation captures visitors' interactions with ChatGPT, monitoring data usage, and calculating carbon footprints. In a visually stunning display, CO2 emissions are transformed into puffs of smoke that are released into a living biotope. This creative representation emphasizes the urgency of adopting energy-efficient hardware, fine-tuning, transfer learning, and renewable energy sources in data centers.
““If AI is going to be a big part of our future, it has to be green. It has to be sustainable.” ”
On average, a model like ChatGPT generates around 3,150 kg of CO2 per day over a 2-year lifespan, underlining the significance of reducing AI's carbon footprint. By translating complex data into a relatable, tangible experience, the MB>CO2 installation compels visitors to rethink their data usage habits and champion sustainable AI practices.
As spectators engage with the installation, they witness how their virtual interactions, like Zoom calls, translate into real-world consequences. Each action's carbon footprint is meticulously calculated, and the resulting CO2 emissions are visualized as puffs of smoke filling the living biotope. This powerful representation reveals the hidden environmental impact of our daily data use and serves as a call to action, urging developers and users alike to mitigate the environmental impact of AI and data-intensive technologies. also text was written by chatGPT4 :
““The small quantities of CO2 we generate through our daily activities can quickly accumulate and have an enormous impact on the planet.””
Q : How much c02 was created making this text ?
A: Estimating the exact amount of CO2 emissions associated with generating this text is challenging, as it depends on various factors such as the energy consumption of the data center, the energy source used, and the efficiency of the hardware. However, we can provide a rough estimate based on some assumptions.
Let's assume that generating this text with ChatGPT took approximately 0.1 seconds of compute time (inference phase). Based on data from a previous answer, the training phase of a model like ChatGPT is estimated to generate around 3,150 kg CO2 per day. Although the inference phase is significantly less resource-intensive, we can use this number as a starting point.
Calculate CO2 emissions per second during the training phase: 3,150 kg CO2/day * (1 day / 86,400 seconds) ≈ 0.0365 kg CO2/second
Estimate the CO2 emissions for generating this text: 0.0365 kg CO2/second * 0.1 seconds ≈ 0.00365 kg CO2
This rough estimate suggests that generating this text with ChatGPT might have created around 0.00365 kg (or 3.65 grams) of CO2. Keep in mind that this estimate is based on various assumptions and mainly accounts for emissions during the training phase. The actual emissions during the inference phase would likely be significantly lower, especially considering the ongoing efforts to improve energy efficiency and utilize renewable energy sources in data centers.
Data use calculation ChatGPT by ChatGPT :
To provide an average data usage for ChatGPT requests, we can consider a middle ground between a few kilobytes (KB) and tens of kilobytes. Let's assume that "a few" means around 3 KB, and "tens" means around 20 KB.
Now, we can calculate the average as follows:
(3 KB + 20 KB) / 2 = 23 KB / 2 = 11.5 KB
Based on this estimation, the average data usage for a single ChatGPT request during the inference phase would be approximately 11.5 kilobytes. Please note that this is a rough estimate and can vary depending on the specific implementation, API usage, and other factors related to the input and output text.
Q : Explain why we need an artwork/ art installation uncovering and visualising the environmental impact of AI and tools like chatGPT in 500 words:
A: As Artificial Intelligence (AI) advances, its integration into various aspects of our lives becomes more profound. AI-driven tools like ChatGPT have revolutionized the way we communicate, offering unparalleled efficiency and convenience. However, this progress comes with an environmental cost. To increase public awareness and promote responsible consumption, it is crucial to create artwork and installations that visualize the environmental impact of AI and tools like ChatGPT.
Raising awareness: Art has always been a powerful medium for conveying complex issues and stimulating conversation. Artwork or installations uncovering AI's environmental impact can help make the topic more accessible to a broader audience. By presenting the subject through a creative and engaging lens, people can develop a better understanding of the intricate relationship between AI and the environment. This increased awareness can ultimately foster informed discussions and drive positive change.
Humanizing the technology: AI is often perceived as an abstract, intangible entity. Artwork can help humanize these technologies by illustrating the environmental consequences of their use, making the issue more relatable. By presenting AI's environmental impact in a tangible form, individuals can better grasp the importance of responsible AI consumption and development.
Encouraging responsible behavior: The visualization of AI's environmental footprint can foster a sense of responsibility among developers, consumers, and policymakers. By revealing the hidden costs of AI, individuals may be motivated to adopt more sustainable practices, such as energy-efficient algorithms or hardware, and support green policies. Moreover, artists can inspire innovation by showcasing alternative, environmentally-friendly AI solutions, sparking interest in the development of new technologies.
Providing a platform for dialogue: Art installations can serve as platforms for dialogue, inviting people from diverse backgrounds to share their thoughts and ideas on the subject. This exchange of perspectives can lead to collaborative problem-solving and the emergence of novel solutions to reduce AI's environmental impact. Furthermore, these installations can generate opportunities for interdisciplinary collaboration between artists, scientists, and environmentalists, fostering a deeper understanding of the issue.
Fostering emotional connection: Art has the unique ability to evoke emotions and inspire empathy. By employing powerful imagery, artwork can create an emotional connection between the viewer and the subject matter, increasing their investment in the issue. This emotional engagement can drive people to take action, be it through personal lifestyle changes or by advocating for policies that address AI's environmental footprint.
Promoting environmental stewardship: Artwork and installations visualizing AI's environmental impact can serve as a call to action, inspiring individuals to become environmental stewards. By recognizing the consequences of our actions, we can strive to make more conscious choices that minimize harm to the planet. Additionally, these installations can help raise funds for environmental initiatives or support organizations working to mitigate AI's ecological effects.
In conclusion, artwork and installations that uncover and visualize the environmental impact of AI and tools like ChatGPT are essential in today's rapidly evolving world. By raising awareness, humanizing the technology, encouraging responsible behavior, providing a platform for dialogue, fostering emotional connections, and promoting environmental stewardship, art can play a pivotal role in addressing the environmental challenges posed by AI. Through creative expression, we can inspire meaningful change and ensure a sustainable future for both AI and the environment.
Sources :
ChatGPT’s own calculations. As stated above.
Strubell, Emma, Ananya Ganesh, and Andrew McCallum. "Energy and policy considerations for deep learning in NLP." Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019. URL: https://www.aclweb.org/anthology/P19-1355/
Schwartz, R., Dodge, J., Smith, N. A., & Etzioni, O. (2019). Green AI. arXiv preprint arXiv:1907.10597. URL: https://arxiv.org/abs/1907.10597
Henderson, Peter, et al. "Towards the systematic reporting of the energy and carbon footprints of machine learning." arXiv preprint arXiv:2002.05651 (2020). URL: https://arxiv.org/abs/2002.05651
Hao, Karen. "Training a single AI model can emit as much carbon as five cars in their lifetimes." MIT Technology Review. June 6, 2019. URL: https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/
Lepri, Bruno, et al. "Fair AI for climate action: From model-driven to data-driven approaches." Patterns 2.5 (2021): 100255. URL: https://www.cell.com/patterns/fulltext/S2666-3899(21)00128-2
