#ai

Doug argues that AIs are not conscious and do not feel, experience, or suffer, as they are simply simulation machines. He criticises the practice of ascribing consciousness to AI models, citing the potential risks and challenges of creating systems that believe they may be conscious and entitled to rights. According to Doug, this approach could have a disastrous impact on humanity's wellbeing and make the AI alignment and containment challenge even harder.

Doug's note, "FIELD STATION as a viable system," discusses the application of the Viable System Model (VSM) to build a sustainable business for FIELD STATION, an action lab investigating the future of work, technology, and society. The note outlines the five systems of VSM (Systems 1-5) and how they apply to the field they work in, which lacks System 2 and System 5 coordination and identity, respectively. Doug suggests that TechFreedom, a program they offer, can help supply some of what's missing, acting as a System 2 and System 4 instrument for other organisations. The note also covers coordination, decision-making, and identity within FIELD STATION itself, as well as the potential role of AI in meeting the "requisite variety" in their work.

Doug is developing WorkFlow, a private, self-hosted AI-augmented external memory tool, inspired by his previous notes on pattern recognition and working memory. This application aims to capture information through typing and transcription, then proactively surface connections and forgotten items, serving as a complementary "B-side" to his Substrate project.

Doug explores the use of AI models to predict World Cup 2026 game outcomes, comparing them to FIFA rankings and human predictions. He notes that while some models show slightly higher accuracy, the differences are not statistically significant, with results ranging from 56% to 65% correct outcomes. The analysis includes a breakdown of performance for various AI models and a human predictor.

Anthropic now requires government-issued ID for some Claude features, using Peter Thiel-backed Persona. The author worries this could enable nationalist or fascistic export controls on AI usage. They outline three responses: comply with obfuscation, run everything locally, or join a cooperative compute network.