Signal president Meredith Whittaker has issued a stern warning about the increasing tendency to anthropomorphise AI chatbots, stressing that they are neither sentient beings nor conscious interlocutors. In a recent interview, Whittaker argued that the 'agentic AI' vision, championed by companies like Microsoft with products such as Copilot, is essentially building a new layer of surveillance infrastructure. While Whittaker admits to using AI for basic tasks like document formatting, she draws a firm line at more profound interactions, emphasizing the importance of her own thought process and writing.
Her most pointed criticism was aimed at Microsoft AI CEO Mustafa Suleyman's suggestion that Copilot could handle tasks like Christmas shopping by accessing group chats. Whittaker highlighted the extensive access such an AI would require – credit cards, browsing history, messaging apps like Signal, personal contacts, home addresses, and calendars. She framed this level of access, particularly within the context of a secure messaging app like Signal, as a "backdoor." This is a significant concern coming from the head of Signal, a platform renowned for its end-to-end encryption, which is also a foundational technology for WhatsApp.
Whittaker's core argument is that agentic AI systems, which necessitate pervasive access to a user's digital life to function, are fundamentally incompatible with end-to-end encryption. If an AI can read messages before they are encrypted or after they are decrypted on a device, the encryption's privacy benefits are nullified. This aligns with her previous warnings about agentic AI posing a "perilous" threat to secure applications and her accusation that operating system vendors are undermining Signal's privacy guarantees. She also pointed to "prompt injection" as a likely initial exploit vector for attackers targeting encrypted messaging platforms. The broader trend of companies like Microsoft, Google, and Apple building operating systems around AI agents creates vast databases of personal information, making them prime targets. Furthermore, Whittaker cautioned against treating these systems as confidants, as their empathy is mimicked through pattern matching, not genuine underst anding, and their data handling practices are often opaque, with commercial incentives to exploit user data.
Fuente Original: https://thenextweb.com/news/signal-whittaker-ai-chatbots-not-friends-copilot-backdoor-privacy
Artículos relacionados de LaRebelión:
- AMA Warns AI NOT for Medical Diagnosis Decisions
- Linus Torvalds AI Coding Hobby Project Not Hype
- AI Chatbots Psychosis Doctors Sound Alarm
- LLMs Gone Wild Can One Long Sentence Really Make AI Chatbots Misbehave
- Illinois Prohibe la Terapia con IA Esta Regulando el Futuro de los Chatbots de Salud Mental
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario