Distinguished Lecture: Dawn Song Finds Challenges And Opportunities At The Intersection Of AI And Security
- Posted by Jesse Polhemus
- on Oct. 25, 2018
Click the link that follows for more Brown CS content about our Distinguished Lecture Series.
Professor Dawn Song of UC Berkeley visited Brown CS earlier this month to deliver the thirty-seventh lecture ("AI and Security: Lessons, Challenges and Future Directions") in the Distinguished Lecture Series.
Described as an innovator and "serial entrepreneur" in Professor Roberto Tamassia's introduction, Song began by setting a historical backdrop of cyberattacks that are increasing in number, financial cost, and volume of users exposed. "And as AI becomes more and more capable," she explained, "the consequences of misuse by hackers will become more severe....Security can really help AI and AI needs security." She then laid out a structured response to challenges at the intersection of the two disciplines, offering examples from her own research into secure and privacy-preserving machine learning solutions.
"One of the fundamental weaknesses of systems is humans," she said, following the observation with an overview of a project that produced an AI-enabled chatbot to monitor correspondence between the user and a third party and identify phishing attempts. She also highlighted the vulnerability of deep learning systems to adversarial examples, demonstrating how attackers could modify traffic signs to confuse self-driving cars even without having any insight into the car's programming.
But far from pessimistic, Song made an effort to depict a future in which machine learning can be less vulnerable and more privacy-preserving. One of her examples, again drawn from her research, was an intelligent and user-controlled virtual assistant that acts as an intermediary when dealing with data-mining third parties such as Google or Facebook. "This will maximize user value," she said, "not corporate value." She ended with a look at her latest venture, Oasis Labs, which is working to create a privacy-first cloud computing platform on blockchain.
We caught up with Roberto after the lecture, and he spoke highly of Song's ability to look at an issue from all sides: "Her talk eloquently brought together the vast opportunities created by AI and the major challenges in securing its deployment and use."
As the presentation drew to a close, queries from the audience came rapidly, and Song provided answers about secure hardware, the difference between AI and machine learning, and the effectiveness of chatbots against social engineers. The final question, about whether future Oasis Labs releases would work with existing social media platforms or create new ones, brought an answer that seemed to reflect one of the lecture's major themes. Faced with continuing challenges, Song said, she hopes her work will "offer meaningful alternatives" that will bring about both behavioral change and security improvements.
A recording of the lecture is available here.
For more information, click the link that follows to contact Brown CS Communication Outreach Specialist Jesse C. Polhemus.