Can Generative AI & ChatGPT Help Protect Our Planet?
Have you noticed the increasing impact of artificial intelligence (AI) applications in our society? The November 2022 launch of ChatGPT, an AI chatbot developed by the Elon Musk- and Microsoft-backed entity OpenAI, skyrocketed this and similar tools into public consciousness.
Many people and companies are now finding intersections between generative AI and sustainability. Could AI tools, including ChatGPT, be applied to the efforts to protect our planet?
What Are Generative AI and ChatGPT?
Generative AI is a type of machine learning technology that builds on the predictive text capabilities you are familiar with in word processing and email applications. Generative AIs use large language models — collections of written, audio, images, and other content — to generate content, including text, code, music, images, and video based on prompts entered by a user.
ChatGPT is trained on a vast dataset of human-created text, which allows it to generate natural-sounding text. Users can prompt ChatGPT to chat, interpret images, and write jokes, emails, and social media captions, among other things. Its impressive ability to write code based on descriptions of a desired piece of software will help speed software development — and may put some software engineers out of a job. Other generative AI tools are specialized for use in creating presentations or generating short videos.
While these tools have many benefits, including the ability to automate tasks and save time, they also have ethical drawbacks. For example, they can plagiarize, be incorrect, and amplify the biases of the creators of the data used to train them. As of March 2023, ChatGPT cannot provide reliable information about the world after 2021, when its dataset was collected.
When it comes to generative AI, the possibilities are evolving in real time.
How Is Generative AI Changing Computing?
While the science of AI has been around since the 1960s, the launch of ChatGPT was a watershed moment that accelerated awareness and uptake of this and similar tools across industries from marketing to education to law. ChatGPT is now integrated into tools like the Bing AI chatbot and Office365, where it will give suggestions and offer automations. At the same time, Google is introducing Bard, an Internet-connected chatbot that uses Google’s own generative AI model, called LaMDA.
AI, which is proficient at identifying patterns in large amounts of data, has long been used to analyze the climate, predict weather, and other types of analysis. Many startup companies are using AI in their climate technology.
Generative AI specifically is newer to the game, but it already has a few climate applications. NotCo, a food-tech startup, is using a patented AI tool named Giuseppe to generate plant-based recipes for products of animal origin, like milk, meat, and even mayonnaise, in partnership with Kraft-Heinz. To help people understand climate impacts, researchers from Montreal Institute for Learning Algorithms and ConscientAI Labs used generative adversarial networks, a type of generative AI, to generate imagery that showed how extreme weather patterns can affect people’s homes.
ChatGPT has already been used to write a draft piece of legislation to regulate itself and other generative AI. Tools trained on specialized topics could potentially help write or refine environmentally focused legislation in the future. A 2023 study in Environmental Science & Technology hypothesized various uses for the generative AI tool. They include helping environmental researchers create digestible summaries of research papers, explain basic research concepts, and bridge a gap in programming skills.
What’s the Carbon Footprint of ChatGPT?
It’s difficult to address the use of generative AI in climate solutions without addressing the technology’s heavy computing demands, and thus the high carbon footprint, required to train and run the many existing AI models. The industry is expected to grow 44% by 2025, after AI computer resource consumption increased by 300,000 times since 2012. The good news is that AI technology has become more efficient, requiring 44 times less computing power over the past 10 years.
OpenAI has not disclosed ChatGPT’s energy usage. When asked about its footprint, ChatGPT replied, “As an AI language model developed by OpenAI, I do not have a physical body or personal actions that generate carbon emissions. My carbon footprint would be limited to the energy consumption of the computers and servers used to run me and process requests. The exact carbon footprint would depend on factors such as the energy sources used to power the computers and the efficiency of the data centers.”
Using the ML CO2 Impact calculator, data scientist Kasper Groes Albin Ludvigsen estimated ChatGPT’s footprint at 23.04 kilograms (50.8 pounds) of carbon-equivalent emissions (CO2e) daily, and 8.4 tons of CO2e annually. He added an update in March 2023, saying his estimate assumed that ChatGPT uses 16 graphics processing units (GPUs). Due to the chatbot’s popularity, it’s now been estimated at more than 29,000 GPUs — over 1,800 times more than Ludvigsen’s original estimate.
While data centers are currently estimated to use 3% to 5% of global electricity, a paper from the University of Massachusetts Amherst has stated that a single training run of a single AI model “can emit as much carbon as five cars in their lifetimes.” The problem is that all successful AI models require intensive, repeated training. Researchers have estimated that GPT-3, an earlier language model used to power ChatGPT, consumed 936 MWh during training alone, enough electricity to power 1,123 homes for a month. The question society will have to answer is whether the training of an AI produces greater efficiency in other parts of the economy that justifies its carbon footprint. The jury is out on this question.
What Does ChatGPT Say About the Environment?
We asked ChatGPT a few questions about terms like “sustainability” and “ESG,” as well as how to reduce carbon footprint, divest from fossil fuels, and make investment decisions. Because ChatGPT synthesizes the human-produced text it was trained on, there are sometimes hiccups relating to “hallucinated” (incorrect) and/or plagiarized statements. However, the chatbot’s tonal responses to these prompts were positive and encouraging, often conversational, and largely accurate.
ChatGPT on Sustainability
For example, its response to, “Can you summarize what sustainability is?” relies on the 1987 UN definition and reflects our use in this Earth911 article. Its high-level summary contained no explicit plagiarism.
The biggest fault we found in our interaction with ChatGPT was its vagueness and repetition, which improved with subsequent prompts, because it learns from the prompts as it works. For example, we began by asking what individuals can do to reduce our carbon footprint. The AI returned reasonable bullet-pointed ideas, like “reduce energy consumption,” “use sustainable transportation,” and “conserve water.” When we rephrased the question, it repeated some points while adding more detail, like “Use energy-efficient appliances: Replace old appliances with Energy Star certified models that use less energy and reduce greenhouse gas emissions.”
While ChatGPT can pull ideas from the sites whose data it was trained on, it cannot necessarily reason about new sustainability ideas, unless it gains access to recent research and other information from which it might mine a new sustainability practice. This may become possible as the technology develops.
ChatGPT on Fossil Fuels
When we asked, “How can we stop supporting fossil fuels?” ChatGPT gave a number of suggestions — from reducing personal energy consumption to supporting politicians, policies, and initiatives that support clean and renewable energy. It ended with the helpful reminder, “Remember, individual actions alone may not be enough to stop supporting fossil fuels. We need systemic change, and it requires collective action and political will to create a sustainable future for all.”
ChatGPT on ESG Investing
ChatGPT can also only give individual recommendations to a limited extent. For example, when asked to suggest an investment portfolio that supports environmental, social, and governance criteria (which it had detailed in a different prompt about ESG), it suggested investing in ESG mutual funds, green bonds, and socially responsible investing accounts. It also noted that it can’t offer investment advice, which is likely a liability safeguard put in place by OpenAI.
The Future of Generative AI for Sustainability
Some believe that the advent of AI, including generative AI, may upend the evolution of our society even more than electricity or the internet.
We recommend beginning to learn about this important, emerging tool. Today’s experience with AI can help you spot its weaknesses, such as propagating misinformation or plagiarism. Think of it as a new form of media literacy — and keep an eye on AI’s impact on the environment.
There are a few ways to follow new developments. Look into organizations working on issues like AI ethics and inclusion, such as Partnership on AI (which brings together diverse voices on AI), AI Now Institute (which conducts research and public engagement towards AI accountability), and AI4ALL (which offers educational resources for young people and teachers). You might also consider subscribing to a newsletter like TLDR AI (“AI, ML, and Data Science in 5 Min”), Unwind AI (which shares bullet-pointed trends in AI), or One Useful Thing (which translates the academic research of a Wharton School professor into AI insights).
Feature photo: Diego – stock.adobe.com