Co-authors Ram Shankar Siva Kumar and Dr. Hyrum Anderson discuss the promise, peril, and impact of AI. They explore data poisoning, information asymmetry in securing AI systems, and the importance of diverse perspectives in cybersecurity.
AI should be adopted responsibly and its risks understood as it becomes more mainstream.
Maintaining a healthy skepticism towards AI is crucial, as adversaries can exploit its weaknesses and trick the system into providing inaccurate or manipulated outputs.
Deep dives
The Genesis of the Book 'Not With a Bug, But With a Sticker'
Dr. Hiram Anderson and Ram Shankar Siva Kumar discuss their book 'Not With a Bug, But With a Sticker' and its focus on AI risk. They emphasize the need for adopting AI responsibly and understanding the risks as AI becomes more mainstream. The book tells stories to help readers comprehend the risks and opportunities of the AI-powered world we are entering.
The Power and Potential of AI in Solving Problems
The podcast explores the many ways AI is transforming industries and solving problems. It highlights AI-powered advances in business, science, and society, extending beyond major tech companies. AI is being used in unexpected areas, such as optimizing supply chains for McDonald's and identifying the ideal number of twists in Hershey's Twizzlers. The podcast emphasizes the economic gains and the comprehensive integration of AI into our daily lives.
The Perils of Over Trusting AI
The episode discusses the danger of over trusting AI systems. AI is designed to excel in specific tasks, but people tend to ascribe more trust to AI than it deserves. The phenomenon of automation bias arises when people rely too heavily on AI and overlook the system's limitations. Additionally, adversaries can exploit AI's weaknesses, tricking the system into providing inaccurate or manipulated outputs. These vulnerabilities highlight the importance of maintaining a healthy skepticism towards AI and understanding its limitations.
The Risk of Data Poisoning in AI Systems
The podcast explores the concept of data poisoning, a mechanism of attack on AI systems. Data poisoning occurs when adversaries insert malicious content into datasets used to train ML models. This poisoned data can mislead the model and compromise its performance and security. The book highlights the risks associated with data poisoning, citing instances where organizations had to discard ML models and rebuild from scratch due to inadvertent poisoning. The episode emphasizes the need to be aware of this threat and mitigate it proactively.
AI is perhaps the most defining technology of this generation, with the potential to solve major business, scientific, and societal issues. As this powerful technology goes mainstream – it is not without its flaws and challenges. “Not With a Bug, But With a Sticker,” co-authors Ram Shankar Siva Kumar and Dr. Hyrum Anderson join Ann on this episode of Afternoon Cyber Tea to chat about the promise, peril, and impact of AI.