Presenters
Source
AI for Good? Let’s Talk Applied AI for Sustainability 🌍💡
The buzz around Artificial Intelligence (AI) is undeniable. We hear grand promises of AI solving climate change, boosting biodiversity, and finding new vaccines. But beneath the hype, especially from big tech companies, lies a more nuanced reality. The massive adoption of AI carries significant environmental impacts, from water usage to carbon emissions and energy consumption.
On the Green IO show, Gaël sat down with Jeremy Tamanini, a seasoned sustainability consultant and host of the Applied AI for Sustainability meetups, to cut through the noise and explore real use cases of AI for good. Jeremy is also a contributor to the E Standard, working on a framework for reporting AI’s environmental indicators.
The Carbon Accounting Conundrum: Saving Time, Not Just the Planet ⏱️💰
One of the most concrete “no-brainer” use cases for AI, according to Jeremy, lies in carbon accounting. Sustainability professionals globally spend an average of 50% of their time on compliance. This involves a laborious process of collecting, synthesizing, and cleaning data from across an organization, often for different reporting standards.
AI tools are emerging to dramatically reduce the time spent on tasks like:
- Communicating with suppliers to gather Scope 3 emissions data.
- Attributing emission factors to various data points.
- Formulating and synthesizing data into different reporting structures.
Platforms like Greenly and Persefoni are already embedding these AI tools, offering significant time savings for sustainability professionals. While the direct environmental outcome might not be immediately obvious, this saved time can free up these professionals to focus on more strategic sustainability initiatives.
The Challenge: Jeremy points out a lack of transparency around the environmental footprint of these AI-powered tools themselves. The cost-benefit analysis of AI integration, considering both its footprint and its “handprint” (positive impact), is still largely unmeasured.
Beyond the Hype: Defining “AI” and “For Good” 🤔✨
Jeremy emphasizes the need to move beyond broad terms like “AI for good.”
- “AI” is Not Monolithic: The term AI encompasses a vast range of technologies. The environmental impact and potential benefits vary dramatically.
- “For Good” is Nuanced: What seems like a positive application can have unintended negative consequences. For instance, time savings in carbon accounting could lead to headcount reductions, potentially diminishing the influence of sustainability teams.
Jeremy’s work focuses on applied AI for sustainability. This means exploring targeted AI applications that can genuinely contribute to sustainability goals. He highlights that many organizations are still in the early stages of understanding and piloting these applications, despite the pressure from big tech to adopt AI rapidly. The conversation is not about whether AI is “good” or “bad,” but about understanding its nuanced impacts and exploring its potential through evidence-based exploration.
Tangible Wins: Where Applied AI is Making a Difference 🚀🌱
Beyond carbon accounting, Jeremy shares several compelling use cases:
1. Energy Efficiency ⚡🏢
AI has been aiding energy efficiency in enterprises for a long time.
- Detecting Equipment Failure: Machine learning can identify equipment failures that lead to energy inefficiency (Scope 1 emissions).
- Optimizing Building Energy Use: AI can optimize heating and cooling in commercial buildings. Shockingly, around 30% of energy for heating and cooling in US commercial buildings is spent on empty rooms. AI offers a powerful way to tackle this waste.
2. Scope 3 Emissions Tracking 🛰️🏭
Gathering and measuring Scope 3 emissions from the upstream supply chain is notoriously difficult. AI, often combined with satellite imagery and computer vision, is making inroads:
- Analyzing Smoke Stacks: AI can learn to identify emission patterns from smoke stacks and attach estimated numbers related to energy use and GHG emissions by comparing them to similar sites.
- Bridging Data Gaps: When direct data from suppliers is unavailable or unreliable, especially for global supply chains, satellite views combined with AI can provide valuable estimates, filling crucial blanks in carbon accounting.
3. Nature Monitoring and Conservation 🌳🌊
The marriage of satellite observations with machine learning AI offers significant advantages:
- Ecosystem Visibility: AI helps us see and track forests, oceans, biodiversity, and land use with greater clarity.
- Optimizing Ranger Patrols: Tools are being developed to optimize the routes of rangers in protected ecosystems, increasing their probability of intersecting with illegal poaching activities.
4. Carbon Credits and Land Use 💰🌲
Satellite-based views of land areas and forests provide greater granularity for carbon credits, addressing scrutiny around their quality and claims. This offers a more transparent and detailed picture for the marketplace.
The Data Dilemma: Measuring Impact and the Pilot Failure Rate 📊📉
A significant challenge Jeremy highlights is the lack of data when it comes to the positive impacts of AI in sustainability. While the negative impacts of AI (like data center footprints) are becoming more visible, quantifying the benefits of applied AI is difficult because it’s rarely tracked as a separate line item.
Furthermore, a widely cited MIT study suggested that up to 95% of AI pilots fail. Jeremy stresses that this is a normal part of innovation and entrepreneurship. However, for these tools to scale and make an impact, organizations need senior management support to allow for experimentation and the possibility of failure.
Navigating the AI Landscape: LLMs vs. Targeted ML 🤖🆚🧠
Jeremy draws a stark contrast between Large Language Models (LLMs) and smaller, more targeted machine learning (ML) applications:
- LLMs: He sees huge associated environmental impacts and limited case studies of good environmental outcomes.
- Targeted ML Applications: These have much smaller environmental impacts in terms of resource requirements but potentially much larger positive impacts in areas like energy efficiency, nature monitoring, and compliance time savings.
The common misconception is that “AI” solely refers to LLMs. This oversight leads to a lack of discernment. Jeremy advocates for a more mindful approach, using LLMs only when truly necessary and exploring the vast potential of other AI types.
The Future of Applied AI for Sustainability: Discernment and Frameworks 🧭🛠️
The conversation concludes with a look towards the future:
- Discernment is Key: We need to evolve towards using AI, especially LLMs, more judiciously. Simply using an LLM as a replacement for a web search has significant environmental implications.
- Need for Frameworks: Jeremy agrees with the idea of developing checklists or frameworks to help product and sustainability managers decide when and if AI is the right solution for a given problem, weighing the benefits against the costs and potential impacts.
- Custom LLMs: While broad LLMs have concerns, smaller, custom-trained LLMs (e.g., trained on specific regulations) show promise.
- The “Token Extravagance”: The current era of excessive token consumption for LLMs, as exemplified by the Nvidia CEO’s comments, is seen as a period of “crazy extravagance.” This echoes the early days of the internet boom, where infrastructure was built before widespread, applied applications emerged.
The journey of applied AI for sustainability is still in its early days. It requires continuous learning, transparency, and a willingness to experiment. By understanding the nuances, focusing on targeted applications, and demanding better data, we can steer AI towards truly beneficial outcomes for our planet.
Connect with Jeremy Tamanini: You can connect with Jeremy on LinkedIn. He also recommends following Green IO and its speakers for ongoing insights into this evolving field.
Learn More:
- Data Center Watch (NGO)
- Greenly
- Persefoni
- GHG Protocol