Date: 13th Oct – SundayTime: 11:00-12:00Location: Ballroom A
Do we trust AI? I don’t, personally.
What is “state of the art” in AI security? Yesterday it was a PoC, not a product, today becoming a We will fix it later, tomorrow it will be a if it works, don’t touch it. And tomorrow is too late.
But what we can do for Trustworthy AI? There are just no simple answers.
You can’t install antivirus or calculate hashes to control integrity of annotated dataset. Traditional firewalls and IDS are almost useless in ML cloud internal SDN Infiniband network. Event C-level Compliance such as PCI DSS and GDPR doesn’t work for massive country-level AI deployments. What about vulnerability management for TensorFlow ML model? How it will impact ROC and AUC?..
To make it better we should rethink Cyber Resilience for AI process, systems and applications to make sure that they continuously deliver the intended outcome despite adverse cyber events. Make sure that security is genuinely integrated into innovation that AI brings into our lives. To trust AI and earn his trust, perhaps?