Is It Safe? Misinformation, Terrorism and the Quest for Safe Chatbots
Track 2 – 3:10pm (please note this is only a 30 min slot)
Bio
Privacy freak obsessed with where legislation meets technology and practical implications of innovation clashing with real life. Started in robotics, graduated in cyber, before working in IT infrastructure and security. Drawing, singing and renovating furniture for fun.
Talk Abstract
Have you ever spoken or written to the AI chat of your choice and wondered why it is so misleading? Why does it give you such nonsensical answers so confidently? Why does it seem to, colloquially speaking, suck up to you so much?
Or maybe you read in the news about yet another trove of concerning safety and security issues AI chatbots suffer from? Does the prospect of such AI being put in charge of anything more serious than a vending machine fill you with concern? Maybe you're worried about the advice AI provides. Or perhaps you are concerned about disinformation it spreads. You'd like to see AI being safer than it is now and only see it getting worse.
If so, this is the talk for you. You're not going crazy; you are not a technophobe nor paranoid. There is a reason why those chatbots "behave" the way they do.
The talk will explain how the AI works and why such safety issues are bound to crop up. You will learn the challenges AI trainers face. You will see the insider practices which make the chatbots speak the way they do and learn why AI vendors require them to be that way. You will also learn what measures and attitudes have a realistic chance of mitigating the risks of the AI being relied upon as a source of information.
In essence, you will find the answer to the question "Is it safe?”
Intended Audience
General public, though some knowledge of LLMs and AI red teaming will be beneficial to understand the context
How NSFW is this talk?
(How spicy is your talk in chilli's?)
PG13; some mention of violence and terrorism