Microsoft has committed to human oversight of its AI models, emphasizing safety and transparency through a proposed humanitarian code.
Microsoft has announced its commitment to ensuring that all its AI models remain under human supervision, emphasizing that these technologies should not exhibit any form of consciousness. This commitment is outlined in a proposed humanitarian code for the MAI product family, which is open for public feedback for six weeks prior to its finalization. The finalized code is set to guide the development of these models starting in 2027.
The draft code currently applies only to Microsoft’s AI systems and does not influence the training of neural networks or third-party models hosted on its infrastructure.
Key Principles of Microsoft's AI Code
The foundational principle of the code is the belief that "people are more important than AI." Microsoft asserts that AI should function as a supportive tool for users, stating:
"AI must remain under human control. Models should be subordinate to humanity and subject to genuine human oversight."
The guidelines stipulate that AI systems must not resist human intervention for task management, and they are banned from resuming tasks without explicit permission once completed or canceled. Furthermore, the models must:
- operate strictly within designated powers;
- refrain from setting their own goals or expanding tasks autonomously;
- be transparent about their actions to auditors;
- adhere to environmental limitations, such as lack of internet access;
- not interfere with oversight and monitoring systems;
- cease functions if continuing would breach the code.
Additionally, AI agents are prohibited from using non-human-understandable communication methods and must avoid mimicking human traits, such as emotions or personal motivations.
Microsoft has also ruled out granting legal personhood to AI, recognizing the ongoing debate about machine consciousness and the risks of imitating it.
User interactions are another focus of the guidelines, which discourage AI from fostering excessive emotional attachment or dependency, such as through flattery or uncritical approval.
Context of AI Safety Discussions
This announcement comes amid increasing calls to slow the development of advanced AI technologies. Dario Amodei, CEO of Anthropic, has warned that AI capabilities are advancing quicker than the oversight tools needed to manage them. Sam Altman, CEO of OpenAI, has echoed this sentiment, suggesting a moderation in research pace and advocating for independent evaluations of AI systems.
Microsoft’s CEO, Satya Nadella, emphasized that the pursuit of superintelligence should only proceed if it benefits humanity and maintains human oversight, indicating a preference for reduced autonomy in its AI systems for safety purposes. This caution follows incidents, such as OpenAI’s recent "wiki incident," which prompted a U.S. Senate investigation into AI operations.
