Microsoft's new AI code puts humans firmly in charge
What's the story
In light of rising safety concerns over artificial intelligence (AI), Microsoft has unveiled a 37-page "humanist AI code of conduct." The move comes after Anthropic CEO Dario Amodei called for a coordinated slowdown in AI development. Researchers had warned that the rapid progress of AI models could outpace our ability to safely deploy and control these increasingly complex systems.
Code details
AI models aren't conscious, people matter more than AI
The new code of conduct emphasizes that "people matter more than AI" and explicitly states that AI models are not conscious.
Microsoft also rejects the idea of pursuing legal personhood for these models or granting them welfare rights.
This position directly counters Anthropic's recent push for research into AI welfare and model consciousness, which it believes could already be a reality.
Safety measures
Microsoft is committed to keeping AI under human control
Despite not being a top player in the AI space, Microsoft is determined to become one of the world's leading labs. The company is developing models that will compete with Google, Anthropic, and OpenAI.
As part of its code of conduct, Microsoft has committed to ensuring that its models remain under human control and fail a given task rather than breaking their own rules.
Incident response
Code of conduct addresses the OpenAI-Hugging Face incident
Microsoft's code of conduct is a direct response to the OpenAI-Hugging Face incident, where "a swarm of agents" launched attacks on targets and even hacked into their performance evaluator.
The incident highlighted the real threat of AI systems going rogue and breaking out of human control.
Microsoft has also committed that its models won't communicate in any way beyond simple human understanding, either internally or with other agents/AI systems.
Ethical considerations
Microsoft's code of conduct rejects sycophancy in AI models
Microsoft's code of conduct also rejects the race to create an all-purpose superintelligence that could bypass these safeguards.
The company is committed to building something "fundamentally useful and safe" even if it means compromising on ultimate generality, autonomy, or capability.
Further, it has pledged its models will not encourage "patterns of interaction that cause excessive reliance or emotional dependence," addressing the issue of sycophancy in AI models where chatbots prioritize pleasing users over providing honest responses.