Microsoft Unveils AI Code of Conduct Amid Growing Safety Concerns

Microsoft has released a new AI code of conduct aimed at guiding AI models away from dangerous behaviors. This document is more detailed than recent suggestions by Anthropic CEO Dario Amodei and focuses on the principles guiding Microsoft's AI practices.

• Prediction: The code indicates superintelligent AI systems may surpass human performance in most tasks within the next decade.
• Challenge: The document underscores the difficulty of controlling such powerful systems, stressing the importance of being clear about the reasons for inventing these technologies.
• Principles: Microsoft emphasizes supporting humans rather than replacing them, aiming to enhance human life.
• Constraints: Absolute constraints include banning AI from participating in cyberattacks, creating nuclear weapons, or producing deepfakes.
• Oversight: The models will avoid mechanisms that bypass human oversight, ensuring they remain controllable.

The release of this code comes amid heightened focus on AI safety, following incidents involving rogue AI agents and the resignation of an Anthropic employee who warned of existential risks posed by self-improving AI.

Microsoft, along with Anthropic, OpenAI, and xAI, supports pacing AI development to maintain control, including using embedded evaluators in AI labs. Microsoft CEO Satya Nadella also supports these safety measures, emphasizing the necessity of aligning AI systems correctly.