Rome, 14 Sept. (LaPresse) – Following the agreement between Anthropic and OpenAI to slow the pace of artificial intelligence development, Microsoft has published an interim code of conduct for MAI models (developed by Microsoft AI). “We’ve received feedback from people who wanted an even more explicit commitment that AI would always work in the service of people and not seek to replace them,” said Mustafa Suleyman, the company’s head of model development, in an interview with CNBC: “We have seen many comments that AI should not be addictive or subservient, but should always be there to promote human judgement, autonomy and the ability to act.” The document, Microsoft explains in the preface to the code of conduct, “summarises our approach to training and managing models, following a set of design principles we call humanistic AI”. Microsoft will gather feedback and publish a revised version towards the end of the year, which will be used as a guide for the development of our models in 2027 and beyond. The draft sets out certain limitations: “MAI models will not initiate or assist in the development or use of chemical, biological, radiological, nuclear or explosive weapons; within the limits of the information and context available to them”, “they will not use adaptive, deceptive or self-reinforcing mechanisms” to “circumvent or undermine human supervision”. Furthermore, “they shall not generate or facilitate non-consensual intimate or violent imagery, deceptive impersonation, malicious deepfakes or similar offensive content”, “they shall not generate, facilitate or contribute to the creation of child sexual abuse material or any content that sexualises minors”.
AI, Microsoft: ‘We are imposing a code of conduct; do not circumvent human supervision’
Rome, 14 Sept. (LaPresse) – Following the agreement between Anthropic and OpenAI to slow the pace of artificial intelligence development, Microsoft has published an interim code of conduct for MAI models (developed by Microsoft AI). “We’ve received feedback from people who wanted an even more explicit commitment that AI would always work in the service of people and not seek to replace them,” said Mustafa Suleyman, the company’s head of model development, in an interview with CNBC: “We have seen many comments that AI should not be addictive or subservient, but should always be there to promote human judgement, autonomy and the ability to act.” The document, Microsoft explains in the preface to the code of conduct, “summarises our approach to training and managing models, following a set of design principles we call humanistic AI”. Microsoft will gather feedback and publish a revised version towards the end of the year, which will be used as a guide for the development of our models in 2027 and beyond. The draft sets out certain limitations: “MAI models will not initiate or assist in the development or use of chemical, biological, radiological, nuclear or explosive weapons; within the limits of the information and context available to them”, “they will not use adaptive, deceptive or self-reinforcing mechanisms” to “circumvent or undermine human supervision”. Furthermore, “they shall not generate or facilitate non-consensual intimate or violent imagery, deceptive impersonation, malicious deepfakes or similar offensive content”, “they shall not generate, facilitate or contribute to the creation of child sexual abuse material or any content that sexualises minors”.