AI and deepfakes on show at Davos

The World Economic Forum is held every January in the Swiss resort town of Davos which this year hosted almost 3000 delegates, including more than 60 world leaders. One of the four key themes in 2024 was Artificial Intelligence as a Driving Force for the Economy and Society.

Deepfake Arms Race

In a year when more than 60 countries, including the US and the UK and India, as well as the European Union, are set to stage national elections, the World Economic Forum has listed misinformation as the number one threat of 2024. Some political commentators have already warned the world to get ready for 2024’s Deepfake Elections.

Hosted at Davos by the WSL Institute for Snow and Avalanche Research SLF, EPFL researchers from Professor Sabine Süsstrunk’s Image and Visual Representation Laboratory (IVRL) offered WEF delegates, including Switzerland’s Federal Councillor Guy Parmelin, the opportunity to experience being deepfaked first hand, as well as learn about how EPFL is working on their detection.

“It's always hard bridging our very technical deepfake research with a general audience, so this maze has helped reduce the research to very fundamental aspects, really highlighting what the tech is capable of. I hope this experience stays with them, so they think twice about the authenticity of a very real looking dancing pope video. It’s so easy to be fooled and that’s why deepfakes are such a threat,” said Peter Grönquist a research engineer with IVRL of the EPFL School of Computer and Communication Sciences (IC).

People can protect themselves from mis-and-disinformation spread through deepfakes. “Check the source and be aware of the capabilities of the technology. It’s essential that people have the opportunity to educate themselves on what technology exists at the moment and is feasible in the future. Once one is aware of deepfake methods, it becomes much harder to be tricked,” Grönquist continued.

From Lausanne to Davos and perhaps beyond, the maze was originally conceived by IC Head of Communications, Deirdre Rochat, as a playful way to raise awareness.

“I think the laughter that occurred at each different deepfake mirror really left an impression and the maze shows the reality of what deepfakes can do. For an installation that started at an Open House day on campus to one that is now helping other events communicate the vital research being done on misinformation we’re proud of the impact that it has already had,” said Rochat.

Swiss researchjoins forces on AI at a national level

To address challenges such as those of AI generated content the AI House Davos hosted an extensive agenda on exploring how to use AI to benefit all, balance innovation with risks and how it will interface with other emerging technologies such as quantum computing and biotechnology.

One new partnership on display was the Swiss AI Initiative, a pooling together of Swiss research resources in AI on a national level, including the EPFL and ETH Zurich AI Centers, for Swiss AI sovereignty.

The Initiative will position Switzerland as a leading global hub for the development and implementation of transparent and reliable artificial intelligence. The new ALPS supercomputer, based at the Swiss National Supercomputing Centre is providing the supporting world-class infrastructure. It’s one of the world’s most powerful computers and gives Swiss scientists access to the sort of computing power only available to the world’s biggest tech companies.

The aim of the initiative is to develop and train transparent and open source new large language models (LLM) where it is clear how the models were trained, what data was used and how results are recovered. Many EPFL professors are co-leading projects using ALPS, including on Foundation Models for health, medicine and education.

In addition, researchers will explore fundamental questions in the development and use of LLMs, such as: What form will future interaction between humans and AI take? What is the appropriate ethical framework? How do we manage security and data privacy? What new approaches can be used to scale up models and make them more energy efficient?

Safer AI for the common good

Another joint initiative announced at Davos, led by the Swiss Federal Department of Foreign Affairs, EPFL and ETH Zurich is the Pilot Gen AI Redteaming Network, that aims to unite tech leaders, AI experts, and policymakers in exploring AI safety testing and risk management.

Professor Carmela Troncoso, Head of IC’s Security and Privacy Engineering Lab (SPRING) is one of the academic leads for the Swiss Call for Trust and Transparency in AI which is leading the Pilot Gen AI Redteaming Network that will see researchers work with the giants of AI, including: Microsoft, IBM, Google, Amazon, AlephAlpha and HuggingFace.

“This network is unprecedented on a global scale and we aim to make the LLMs behind all the AI systems that we use as safe as possible. Collaboration like this is very important so that academia and industry can share expertise, explore current problems and work out how to mitigate them to create safer AI for the common good. But let’s be clear, this initiative doesn’t replace regulation, it’s complimentary,” Troncoso explained.

AI for the UN Global Goals

Also launched at the World Economic Forum by the Swiss Federal Department of Foreign Affairs, EPFL and ETHZ was the International Computation and AI Network (ICAIN) to develop AI technologies that will be sustainable, accessible to all and benefit society as a whole, contributing to the UN’s Global Goals.

An initial pilot project is the collaboration between ICAIN and Data Science Africa (DSA), a pan-African organisation of data scientists across the continent, offering training programmes and supporting non-profit research projects in the area of machine learning and data sciences. The first joint project aims to use AI to make agriculture more resilient to the negative effects of climate change.