Microsoft unveils code of conduct for AI models
Microsoft has published a code of conduct for artificial intelligence models intended to steer them away from dangerous behavior. The document defines the values and restrictions that guide the company in training models, TechCrunch reports.
Absolute restrictions for models
Under Microsoft’s system, each model will have a general code of conduct that will take priority over the wishes of individual users and specific tasks. Absolute restrictions include a ban on using models for cyberattacks, activities related to nuclear weapons, and the creation of deepfakes.
The document also contains provisions designed to prevent an overall loss of human control over AI. Models must not use adaptive, deceptive, self-reinforcing mechanisms, collusion, or other methods to evade or overcome human oversight. Authorized people or systems must retain the ability to reliably direct, modify, and shut down models.
More current news is available on the UA.News Telegram channel Telegram.
Principles of AI development
The code stipulates that models should support people rather than replace them, and should promote human well-being. The document predicts that over the next decade, superintelligent AI systems will surpass humans in performing most tasks. Containing, controlling, and aligning such systems is described in the code as one of humanity’s greatest challenges.
Microsoft, together with Anthropic, OpenAI, and xAI, generally supports an approach of cautious development of advanced AI systems. Microsoft CEO Satya Nadella also expressed support for research focused on AI alignment and the idea of embedded evaluators in AI laboratories.
