Thursday, 14 September 2023

The Thirsty ChatGPT! Cultivate Responsible Prompt Engineering! Be Accountable for your Prompts and behind Water footprint!

Dear Friends and Students,

During a recent visit to a friend's home, I witnessed an interesting scenario. A 10-year-old boy was playfully engaging with ChatGPT, and his parents were thrilled to see him embrace this AI tool. While technology adoption is a positive development, I believe it's important to educate the entire family about a crucial aspect of these interactions.

In technical terms, our interactions with ChatGPT through the textbox are referred to as "prompts." In this instance, the boy asked a simple question, "How are you?" and received a response from ChatGPT: "I'm just a computer program, so I don't have feelings, but I'm here to help you with any questions or tasks you have. How can I assist you today?" This exchange was fun and entertaining, but it's essential to understand that for every 20-50 such conversations or prompts, ChatGPT consumes approximately half a litre of water.

What may surprise you is that ChatGPT's water consumption is related to the cooling of the data center machines that power the AI. In fact, Microsoft's global water usage increased by 34 percent from 2021 to 2022, reaching nearly 1.7 billion gallons (source: www.businesstoday.in).

It's not just ChatGPT; every AI-powered data center has a significant water footprint. Consider the numbers: Google Bard has about 1,00,000 daily active users, while ChatGPT boasts over 10 crore active users. Microsoft Bing Chat has an estimated 50,000 daily active users, Meta Llama 2 has around 25,000, Claude has roughly 10,000, and GitHub CoPilot has about 5,000 daily active users.

Starting with the estimated daily active users of ChatGPT and Bard, if each user generates an average of 10 prompts per day, the total number of prompts from these two engines alone would reach 130 million per day. Adding in prompts from other competitors like Microsoft Bing Chat, Meta Llama 2, Claude, and GitHub CoPilot, the daily total exceeds 135 million. To put this into perspective, 130 million prompts require a staggering 65,00,000 liters of water. Assuming an average person consumes 3 liters of water per day, this amount could sustain approximately 21,66,667 people daily.

I've been asked whether AI models like ChatGPT are sustainable. The answer is Yes/No, depending on the context. AI can be sustainable when used for vital tasks. For instance, when AI analyses medical images such as X-rays, CT scans, and MRIs to help doctors detect diseases like cancer, fractures, and neurological disorders, it greatly enhances healthcare accuracy and efficiency. However, we might consider delaying AI-powered meme generators, AI toys, and entertainment robots to reduce the water footprint. We can also postpone AI chatbots for casual conversations and AI virtual assistants for entertainment and convenience.

In conclusion, AI developers and users should be transparent about their products' water footprint. Users must be aware of the environmental impact of their choices. To address the water footprint of AI, developers, users, policymakers, and stakeholders must collaborate and take responsible actions.

Warm regards,

Ravi Saripalle

No comments:

Post a Comment