Summary by Sidonie Wittman
At the AI Climate Conversation on November 21st, Assistant Professors Ifeoma Adaji (Computer Science) and Niyi Asiyanbi (Geography) from UBC Okanagan and Assistant Professor Mohammad Shahrad (Electrical and Computer Engineering), Associate Professor Ivan Beschastnikh (Computer Science) from UBC Vancouver came together to look at how AI is being used in current climate adaptation efforts, and how to quantify the climate impacts of AI usage.
They shared both the potential benefits from the use of AI (e.g. for climate adaptation measures such as wildfire management and preparedness) as well as the real climate impacts of its use (e.g. the high carbon and water costs that can be difficult to avoid).
Drs. Shahrad and Beschastnikh started the conversation with an overview on AI’s rapid increase in use in recent years, and explained that as AI continues to improve, usage will continue to grow. This increase has large carbon emissions implications, as data centres are predicted to use almost 1000 TWh per year in total by 2030 in a projection by the International Energy Agency.
Part of this growth is the use of AI in software development, with Dr. Beschastnikh saying AI usage in the sector is “essentially pervasive at this point”. Actions people take day to day, such as ordering an Uber, have a carbon footprint from the AI embedded in their formulation. The increased energy use associated with this pervasiveness is externalized, hiding it from consideration.
The Solutions Scholars Project End‐to‐End GHG Emissions Tracking of Online Services, led by Drs. Shahrad and Beschastnikh, works to understand the emissions related to cloud request processing.
For example, when a website user clicks a button on a hotel booking site, the site issues a ‘request’ to the cloud in order to construct the next page specific to that user. The focus of the Solutions Scholar project is to understand the emissions that result from constructing that next page in the cloud based on the user’s specific request. Through this work, they hope to create a system that can track the footprint of semantically meaningful user actions, such as clicking on a button to share a document in OneDrive.
In looking for solutions against the increased environmental impact of AI, Dr. Shahrad cautioned against the idea that simple increased energy efficiency would lower energy consumption. Using the concept of Jevons’ Paradox, he explained that because the market for AI tools is so competitive currently, increased efficiency would likely just allow for greater usage, cancelling out potential for energy benefits. He explained this saying “I think in the foreseeable future, any gains in efficiency of our accelerators, data centres, would just mean more and more demand. Overall, there is going to be growth in the energy consumption”.
Drs. Adaji and Asiyanbi presented potential uses of AI in wildfire preparedness, looking at their Solutions Scholars project Enhancing Wildfire Preparedness Through a Multi‐Platform Digital Tool.
Dr. Asiyanbi explained that research on machine learning usage in wildfire tracking and management goes back to the 1990s, and recent expansion of AI has greatly increased speed and accuracy in wildfire detection, mapping, and forecasting. Because of this expansion, he said that “traditional fire lookout towers are being replaced by this massive AI powered network of cameras that's tracking fire in real time and are able to communicate to coordination centres where decisions are taken”. Building on recent improvements in AI, their Solution Scholar project seeks to develop a multi-platform application with AI-powered functionalities, including a chatbot that can answer questions about wildfire safety.
While AI usage is improving wildfire management capabilities, the speakers indicated key drawbacks or deficiencies in AI usage in wildfire management. Firstly, Dr. Asiyanbi spoke on the moral hazards of techno expectations, where an overconfidence in these new tools leads to complacency in other realms of fire prevention. Relatedly, he warned that these tools could contribute to a bias towards fire suppression, as opposed to preparedness or working with natural wildfires instead of against them. He cautioned that we cannot rely on a sense of security from these AI tools, and encouraged continual work looking at social factors of wildfire preparedness.
Also, Dr. Adaji described how AI can “hallucinate”, or provide factually incorrect information. This would be potentially very harmful in the WISEC app, due to the high stakes of wildfire preparedness. To combat this, the team is using RAG, or retrieval augmented generation. This AI would prepare answers based on a log of factual information, and use previous knowledge to deliver this factual information to a query. This is opposed to other AI which only uses memory, and not a foundation of factual information.
Overall, the speakers spoke to many benefits from using AI, and ways to mitigate the drawbacks in AI usage. However, they also cautioned against reliance on AI, with Dr. Shahrad saying “sometimes not using AI actually means being cheaper, or being faster. So it is important that we understand how to use it like any other tool that we use around us. It’s… something to be using judiciously”.
To learn more about these projects, see our Solutions Scholars page. Thank you to the speakers for their expert insight, and thank you to all participants who came and asked questions.
Speakers:
Dr. Ifeoma Adaji (she/her)
Professor, Department of Computer Science, Mathematics, Physics, and Statistics (Irving K. Barber Faculty of Science), UBC.
Ifeoma Adaji is an Assistant Professor of Computer Science at UBC Okanagan, affiliated with the Master of Data Science (MDS) program and the Department of Computer Science, Mathematics, Physics and Statistics. Her academic appointment focuses on data science, human-computer interaction, and ethical computing. She earned a PhD in Computer Science from the University of Saskatchewan, Canada, and a Master’s degree from the University of Aberdeen, UK. Her research explores the design and development of persuasive technologies and behavior-change systems, including serious games and mobile/web applications that influence user behavior for societal benefit. She also studies user behavior modelling, social computing, and the ethics, trust, fairness, accountability, and transparency of artificial intelligence (FATE).
Dr. Adeniyi P. Asiyanbi (he/him)
Professor, Department of Community, Culture and Global Studies (Irving K. Barber Faculty of Arts and Social Sciences), UBC.
Adeniyi P. Asiyanbi is an Assistant Professor in Geography within the Department of Community, Culture and Global Studies at UBC Okanagan. He teaches undergraduate and graduate courses in human geography, climate change and society, development geography, and political ecology. He holds a PhD in Geography and an MSc in Environment and Development, both from King’s College London, UK, and a BSc in Forestry (First Class) from the University of Ibadan, Nigeria. His research applies critical geography and political ecology to investigate the intersection of climate change, forests, governance, development and social justice. Asiyanbi critically examines how forests are positioned as both “climate solutions” (e.g., REDD+, carbon forestry) and “climate-related risk” (e.g., wildfire governance), exploring the power dynamics, rationalities and impacts of environmental policies and interventions across contexts in Nigeria, Ghana and Canada.
Dr. Mohammad Shahrad (he/him)
Professor, Department of Electrical and Computer Engineering (Faculty of Applied Science), UBC.
Mohammad Shahrad is an Assistant Professor in the Electrical and Computer Engineering Department at UBC Vancouver. At the intersection of software systems and computer architecture, his research is focused on building resource-efficient cloud computing systems. Mohammad’s recent focus has been on enabling and efficiently managing emerging cloud services, such as serverless computing. He has collaborated with different industry teams in this space. Mohammad’s research has been deployed in production, won the USENIX Community Award, and been featured in the CACM Research Highlights.
Dr. Ivan Beschastnikh (he/him)
Professor, Department of Computer Science (Faculty of Science), UBC.
Ivan Beschastnikh is an Associate Professor in the Department of Computer Science at UBC Vancouver. Beschastnikh’s research spans distributed systems, software engineering, formal methods, program analysis, security and privacy, and machine learning for systems. His group focuses on building real software systems and empirical tools that improve developers’ capacity to build, debug, and understand complex software. He is involved with research clusters including UBC’s Systopia Lab, Software Practices Lab (SPL), the Data Science Institute, and the Blockchain@UBC initiative.