Eclipsing Threats: AI data poisoning and planetary atmospheres

After more than 60 years of unmanned space exploration, we still have a lot to discover about the atmospheres of planets and worlds in our solar system. Whenever we try and predict from the known atmospheres what global structures we should find in the next unexplored world, we usually get it wrong, even if the basic Physics we assume is correct. What does that tell us about our level of understanding, the predictability in general of atmospheres and thereby our ability to reproduce the global 3-D structures of extrasolar planet atmospheres? And what do we need to do better?

AI is cool. Chatbots, recommender systems, and virtual assistants, to mention a few, hugely benefit from AI. However, this success can also make AI a very appealing target for attackers. For instance, attackers can manipulate training data collected from the Internet to subvert the learning process of the AI and increase its errors, in a similar way to how a malicious teacher would teach their students a wrong behaviour. Whether you are an enthusiast, sceptic, or curious, grab a pint, take a seat, and join us for an overview of how AI can be exposed to attacks. But fear not, we will also explain how we can protect AI.