TL;DR:
- Artificial Intelligence (AI) plays a significant role in the fight against climate change, but its energy consumption and carbon footprint pose challenges.
- Advanced computing technologies offer a sustainable alternative to AI, with quantum computing, photonics, and neuromorphic computing leading the way.
- Quantum computing excels in drug discovery by quickly simulating binding sites for optimal therapeutic configurations, reducing energy usage.
- Photonics, using laser-produced light, provides energy-efficient information transmission, offering a potential route to achieving Net Zero.
- Neuromorphic computing replicates the analog nature of the human neural system, running large deep learning networks with exceptional energy efficiency.
- Startups and companies like Pasqal, Qubit Pharmaceutical, Lightmatter, and Groq are leading the way in advanced computing for climate impact.
- Investing in advanced computing technologies is essential for addressing climate change while leveraging the power of AI for environmental mitigation.
Main AI News:
In today’s fast-paced world, news about the Earth and humanity inching closer to a climate change tipping point is a common occurrence. The gravity of the situation was further underscored by recent reports from the World Meteorological Organization and the UN Secretary-General. These reports served as a wake-up call, urging us to reflect on the potential consequences of our actions.
Amidst this uncertainty, Artificial Intelligence (AI) stands as a pivotal tool that can either push us past the tipping point or pull us back from the edge. In the fight against climate change, AI has proven to be an invaluable asset. It has been employed to predict risks, prevent destructive weather events such as wildfires, and facilitate carbon offset initiatives. Furthermore, AI plays a crucial role in helping companies meet their Environmental, Social, and Governance (ESG) targets.
However, we must acknowledge that AI also acts as an accelerant in this process. The computational power required for AI operations consumes significant amounts of energy, particularly during the development of algorithms and the training of models. Similar to how software revolutionized various industries, AI is poised to follow suit, permeating every aspect of our lives.
By 2030, AI is projected to contribute a staggering $15.7 trillion to the global economy, surpassing the Gross Domestic Product (GDP) of influential nations like Japan, Germany, India, and the UK. This pervasive use of AI, akin to the widespread adoption of the internet, ranges from leveraging ChatGPT for crafting emails and writing code to utilize text-to-image platforms for artistic creations.
The energy demands of AI have been spiraling upward over the years. For instance, the power required to train the largest AI models doubled every 3.4 months between 2012 and 2018, increasing by a staggering 300,000 times. While this expansion presents opportunities to address critical real-world challenges spanning security, medicine, hunger, and agriculture, it also exacerbates the issue of climate change.
The Price of High Energy Consumption
Computing, an indispensable companion to AI, carries significant energy costs and contributes to a larger carbon footprint, further propelling climate change. This is particularly evident in the case of AI, where an extensive array of Graphics Processing Units (GPUs) employed in machine learning algorithms generate substantial heat and require cooling to prevent damage. The energy consumed during the training of even a single large language model (LLM) is truly mind-boggling and leaves behind a sizable carbon footprint.
Consider these examples:
- The training of BLOOM emitted 50 metric tons (50,000 kg) of carbon dioxide, equivalent to the emissions generated by 60 flights between London and New York.
- Training OpenAI’s GPT3 produced 500 metric tons of carbon emissions, while Meta’s OPT emitted 75 metric tons.
- Training GPT-3 alone consumed a staggering 1,287 gigawatt-hours, resulting in 502 tons of carbon emissions, equivalent to the annual electricity consumption of 120 US homes.
As we venture into the GPT4 era, with models growing in size, the energy required for training them also escalates. GPT-3, 100 times larger than its predecessor GPT, was succeeded by GPT-4, which was ten times larger than GPT-3. Notably, the pace of releasing larger models is accelerating, with GPT-4 hitting the market in March 2023, a mere four months after the launch of ChatGPT (powered by GPT-3.5) in November 2022.
To maintain a balanced perspective, we must not assume that AI’s carbon footprint will inevitably continue growing as new models and companies emerge in the field. Geeta Chauhan, an AI engineer at Meta, has been leveraging open-source software to reduce the operational carbon footprint of LLMs. Her latest work showcases an astounding 24-fold reduction in carbon emissions compared to GPT-3.
However, the popularity and exponential power of AI pose a significant obstacle to current climate action measures, casting doubt on its potential to be a part of the solution. In this critical juncture, we must find a way to allow AI to flourish while curbing its carbon footprint. The question remains: What can be done to achieve this?
Curbing the Carbon Addiction
As always, technology holds the key to resolving such predicaments. For the sustainable growth of AI, advanced computing must step into the spotlight and shoulder the burden of tasks currently performed by AI. Fortunately, advanced computing technologies already exist and excel in efficiency and speed, all while consuming significantly less energy than AI.
In essence, advanced computing is the most potent tool at our disposal to temper AI’s carbon addiction and impede the advance of climate change. Through its adoption, we can stem the tide of climate change. Multiple emerging technologies in advanced computing can address some of the challenges that AI currently tackles.
For instance, quantum computing surpasses AI in drug discovery. With humans living longer, there is a growing number of complex and untreatable diseases. This “better than The Beatles” problem refers to the modest improvements of new drugs over already successful therapeutics. Traditionally, drug development focused on rare events within a dataset, relying on educated guesses to design drugs that target disease-causing proteins. LLMs have proved to be helpful in this process, as they excel in predicting the most suitable words to convey accurate meaning within a sentence.
Similarly, drug discovery involves identifying the best configuration of molecules within a compound to achieve therapeutic results. As molecules are quantum elements, quantum computing is better equipped to tackle this problem efficiently. By rapidly simulating numerous binding sites in medicines, quantum computing enables the creation of the optimal configuration for treating currently incurable diseases, all while consuming less energy.
Advanced computing also encompasses photonics, also known as optical computing. This technology relies on laser-produced light instead of electricity to transmit information. Companies are actively developing computers utilizing photonics, recognizing their superior energy efficiency compared to other computing technologies. Photonics is increasingly viewed as a pathway toward achieving Net Zero.
Another intriguing field is neuromorphic computing, wherein computer systems are designed to mimic elements of the human brain and nervous system. This approach seeks to replicate the analog nature of our neural system, enabling computations that mirror human neural processing. Promising trials in neuromorphic computing have been conducted by companies like Mythic and Semron. With more significant investment, neuromorphic computing has the potential to support large deep learning networks that surpass traditional computing systems in energy efficiency. Notably, processing information through its hundred billion neurons consumes a mere 20 watts, equivalent to the power consumption of an energy-saving light bulb in a home.
The Imperative Role of Advanced Computing
The development and application of these innovative technologies are imperative to address the challenges posed by climate change. While numerous startups and investors worldwide are passionate about advanced computing, only a select few companies focus on the so-called impact areas such as healthcare, the environment, and climate change. Within the realm of quantum computing, notable companies like Pasqal, Qubit Pharmaceutical, and IBM are actively exploring energy and drug discovery use cases. In the field of photonics, industry leaders with global impact include Lightmatter and Luminous. Neuromorphic computing progresses under the watchful eyes of Groq, Semron, and Intel.
Advanced computing is vital for achieving the necessary energy efficiency to combat climate change effectively. The energy-intensive process of running artificial neural networks on GPUs proves to be time-consuming and environmentally taxing. By embracing advanced computing methodologies as alternatives to AI, businesses can significantly reduce AI’s environmental impact while harnessing its immense power to mitigate the effects of climate change, such as predicting wildfires and extreme weather events.
Conclusion:
The emergence of advanced computing technologies presents a promising solution to tackle climate change while mitigating the environmental impact of AI. Quantum computing, photonics, and neuromorphic computing offer energy-efficient alternatives to AI in drug discovery and information processing. Companies focused on advanced computing for climate impact are well-positioned to drive innovation in the market. Investing in these technologies is crucial for businesses to align with sustainable practices and address the urgent challenges of our time.
