En Route to Making AI Sustainable – Together!
Table of Contents
Transparency as a Catalyst for Trustworthy and Sustainable AI
Rethinking Stakeholder Perceptions in Sustainability and Data Science
Artificial Intelligence is often hailed to be the tool that will accelerate the world’s journey towards a more sustainable future, but how it will do that remains mostly vague.
While AI holds great potential, it also has a troubling side: privacy risks, possible propagation of inequalities and discrimination mark AI’s rapid development. Regarding sustainability, its hefty environmental footprint can’t be overlooked — by 2030, data centres powering AI are expected to spew 2.5 billion tonnes of greenhouse gases, according to a report by Morgan Stanley.
Without proper oversight, AI might fast-track us to more problems, rather than solutions.
To harness AI’s potential for sustainability, we must build ecosystems that foster collaboration, learn from best practices, and innovate on how to implement this powerful technology.
DSC DACH - AI-POWERED SUSTAINABILITY & RESPONSIBLE AI
The Data Science Conference (DSC) DACH invited leiwand.ai to join and engage with some of the leading minds in AI & Data Science, connecting and inspiring each other in shaping the future of sustainable Artificial Intelligence.
As experts on fair and trustworthy AI, we took responsibility in sharing our insights on the importance of transparency in AI development and deployment. We also gained a lot of new knowledge from other participants:
From using AI systems to improve the efficiency of the energy sector, to decolonising digital archives by cultivating collaboration of affected communities and researchers, we had the pleasure to network, exchange ideas and engage with a plethora of different fields and perspectives.
The event was held in Vienna at the Marriott Hotel, bringing together data & AI professionals, where we learned about the most recent advancements in sustainable AI and emerging trends.
This report summarises some of the insights we gained from the event.
Transparency as a Catalyst for Trustworthy and Sustainable AI
“True transparency requires courage, but it leads to more informed decisions about the beneficial use of AI systems”
Our CTO Rania Wazir held a keynote on transparency as catalyst for trustworthy and sustainable AI, asking the hard-hitting questions that AI developers, deployers and society as a whole need to answer with regard to regulations such as the EU AI Act.
She raised critical questions about whether transparency involves documentation, access, or something deeper, and how it relates to other trustworthiness factors such as human agency, fairness, sustainability, privacy, and cybersecurity.
Rania delved into the meaning of transparency in AI, considering the inherent limitations these systems have, such as reproducing bias, leaking private data, malfunctions or the carbon footprint of AI data centres. Quoting the mathematician Cathy O‘Neill, she argued that AI systems need to be considered untrustworthy until proven otherwise, if we want to make them sustainable.
Additionally, she reflected whether transparency should be viewed as a fundamental right or a privilege, emphasizing that true transparency requires courage and essentially fosters more informed decisions about the ethical use of AI systems, which can be directed for the benefit of people and the planet.
Totally Rational: The biggest misconception of how stakeholders perceive our sustainability and data science projects
“If decision makers don’t understand your data project, then they don’t need more information. They need less of it”
How do we convince stakeholders of our sustainable and data science projects?
As important as our data science projects may be, getting across what sort of benefits they eventually will have for people is often tricky.
Dr. Julia Zukrigl enlightened us on how we have to communicate our goals to stakeholders: leaning on behavioural science research, instincts of flight & fight and creating “emotional visions” about how people benefit from your project when advocating for it.
As humans, we are strongly motivated to think and act on situations according to how our brains interpret them in the very first moments: Is this a threat? If not, is it of interest to me? Zukrigl said it is important to manage this perceived threat first when engaging with stakeholders.
It is therefore also paramount to accept that “emotional” side of our brains when pitching our ideas. This means that rather than solely reciting hard facts and leaning on objective argumentation, stakeholders need “emotional visions” about people who will gain something from your project. Stories about individual customers and anecdotes are therefore far better of a communication strategy than dumping data on them, Zukrigl suggests.
Sustainable AI in the Age of Digital Historical Archives
“Racism and sexism are part of the foundation of colonialism, and they will be reproduced by AI systems if we train them uncritically on archival data”
Dr. Petra Weschenfelder from the University of Vienna made us aware of how AI systems cannot be used blindly when digitizing archives.
As with current AI systems that run on data full of bias, archival material may reproduce sexism and racism in AI systems when used to train them. This is important outside the realm of archival work, as she demonstrated.
She pointed out current data challenges, showing that AI systems trained on recursively generated data lead to their collapse. In other words, the internet is getting increasingly full of AI-generated content, which is then used to train AI, decreasing their functionality. To overcome this problem, AI developers are looking towards gaining new data from old data, in this case archives.
Needless to say, archival data is full of racist and sexist content. If we are to train AI systems on digitised archival data, we need to critically assess historical datasets, engaging with colonised communities and the implementation of ethical and legal data curation.
AI and Energy Transition: How to accelerate adoption
“AI should help us reach net-zero by 2040 or 2050”
The energy sector, much like the finance sector, is a conservative, highly regulated “slow mover” with a duty to society. It also recognizes the need to transform itself into a net-zero business.
This was the initial statement brought up in a panel discussion by Dietmar Boeckmann (Member of the Management Board BKS Bank AG), who was joined by Tarry Singh (CEO of DK AI Lab and RealAI B.V) and Mark Stefan (Senior Research Engineer at the Austrian Institute of Technology).
Moderated by Lisa Kratochwil from Verbund AG, the panellists brought their A-game in highlighting how the energy sector is evolving – albeit slowly – and how AI can be an integral part to making the sector more sustainable.
Tarry Singh argued that the energy sector needs to think about the small picture by miniaturizing projects, arguing that microgrid advancements are far more convincing and show more easily how AI can have a positive effect on the energy sector’s sustainability and efficiency. A microgrid combines renewable energy sources like solar and wind power, along with combined heat and power plants and energy storage systems, to create a localized energy network.
Even though the microgrid might be the way to go, the panellists agreed that big players are required to enhance the innovation of AI in Europe, inferring that it will take large amounts of resources to get AI systems to a point where they can become a positive and integral part of the energy sector.
Furthermore, more access to data through data sharing is needed to improve and ship new AI-based products for the energy sector, Mark Stefan assessed, reinforcing that transparency is key in improving AI products. He further stressed that AI systems shouldn’t be used just for the sake of using them: they need to be applied according to necessity and with accommodating methodologies.
Whether AI would become an important part of the energy sector remained contentious among the panellists. While some agreed that it will help the energy sector attain its net-zero goals, data also shows that spending on AI systems in the sector is below average. What all lamented is the lack of innovation in Europe, which needs a boost to stay competitive on the global stage.
Evaluation and Observability of GenAI Applications
“Faithfulness is the solution to LLM hallucination”
Igor Nikolaienko (Generative AI Architect at DHL and Deutsch Post) gave a run-down of how evaluation and observability of generative AI application is essential to optimize large language model (LLM) performance by testing different prompts and applications parameters.
To improve the evaluation of GenAI applications, he presented his vision of a new automated evaluation method that leverages synthetic data generated by LLM- generated Q&A’s, where LLMs function as “the judge” to evaluate LLM performance. He added that it is key to establish evaluation methods, define metrics and KPI’s and choose an evaluation framework when using this method.
Reducing Environmental Impact of Digital Services through Understanding Environmental Impacts of AI
“Make money, but not at the expense of everything and everyone else. What are we going to do with a supercomputer when there is no one left to enjoy it.”
Elina Stanek (sustainability lead of Women in AI Austria) sat down with Dr. Inez Harker-Schuch (Head of Strategic Initiatives Cintana Education) and Alice Schmidt (Policy Advisor and Business Consultant) to address some of the crucial questions regarding the environmental impact of AI. The two experts offered an alternative view, highlighting that technological advancement does not happen in a vacuum, innovation starts with access and education.
They promoted thoughtful and meaningful investments over hype based commercial interests that aren’t long-lived but leave tremendous damage to the planet and society. The crucial message being, that we need to understand the impact of the tools (AI) we use, to be able to reduce the environmental impact.
Smart Vision for Smart Energy: Computer Vision Applications in the Energy Sector
“Real innovation is about solving real-world problems with the right tools, not about chasing the latest buzzwords.”
Data scientist Magdalena Hutze gave us a deep dive into how her employer VERBUND, a leading energy company and one of the largest producers of electricity from hydropower in Europe, uses AI systems to improve very different kinds of workflows.
The company uses AI to streamline energy contracts, employ a robotic dog to collect data on the power plant, and use recognition systems to monitor whether fish pass through their water power plants, showing that AI systems are not just a fancy add-on, but already an essential part of the energy sector in Austria.
Fotos by DSC DACH and leiwand.ai