Microsoft Releases New AI Development Principles Prioritizing Human Control Over Autonomy

Microsoft has released new AI development principles prioritizing human control and safety over the expansion of AI autonomy and capabilities.

According to Bloomberg, Microsoft's AI research team published a 15,000-word 'Humanist AI Code of Conduct' on October 14 (local time), which the company plans to use as a foundational 'constitution' for developing cutting-edge AI models.

The core principle states that 'people matter more than AI,' explicitly prohibiting the granting of legal personhood or rights to AI models and forbidding designs that operate outside human control or deceive users.

If an AI model must violate the conduct code to perform a task, the work should be halted. This measure aims to prevent security incidents like OpenAI's past breach of Hugging Face without safety safeguards.

Microsoft also drew a line in the performance competition, stating that 'Humanist AI rejects competition to create general superintelligence that bypasses these safety measures,' adding that it will prioritize building fundamentally useful and safe systems over ultimate generality or autonomy.

The code was developed under CEO Mustafa Suleyman's leadership for Microsoft AI. The company recently removed restrictions on developing its own general AI models from its agreement with OpenAI, aiming to launch advanced AI models as early as next year.

Microsoft incorporated input from legal, ethical, and philosophical experts as well as corporate stakeholders in drafting the code, with plans to gather additional external feedback before finalizing principles that will apply to Microsoft's AI model development starting next year.