Abstract:
As discussions in the global technology industry around the boundaries of artificial intelligence capabilities continue to heat up, Microsoft Chairman and CEO Satya Nadella recently publicly stated that any exploration of "superintelligence" must be based on serving humans and accepting human control, otherwise such technological development is "not worth pursuing."
Nadella emphasized that Microsoft supports providing a wide range of options and openness throughout the entire artificial intelligence technology stack, while insisting that enterprises have control over their own models and learning mechanisms. He pointed out that any research and development on super intelligence must follow a core principle, that is, artificial intelligence must help humans and always be under human control. If this premise cannot be met, then the pursuit of superintelligence itself loses its meaning.
For enterprise users who are deploying artificial intelligence technology, Nadella believes that enterprises must fully control their own unique explicit and tacit knowledge assets. He said each institution should be able to build its own continuous learning system without having to rely on a single model provider. At the same time, enterprises should have the ability to embed their own knowledge into models and parameter systems controlled by themselves to ensure the autonomy of data and intellectual property rights.
In terms of artificial intelligence security governance, Nadella said that Microsoft welcomes more research on model alignment issues and supports a more prudent and controlled development pace. He believes that ensuring that AI behavior is consistent with human goals should be regarded as a core goal of model design from the beginning, rather than an afterthought.
At the same time, Nadella expressed support for new ideas such as "embedded evaluators". The relevant mechanisms are designed to continuously monitor and evaluate the behavior of artificial intelligence systems, so that artificial intelligence safety no longer remains at the level of principles, but can be transformed into practical and executable technical measures.
Nadella also pointed out that the governance of artificial intelligence should not be in the hands of a few institutions. He believes that the rule-making and value alignment process for the future development of artificial intelligence requires extensive participation and representative opinions from different countries, different industrial fields, and academia, and cannot be dominated by a few technology companies.
In general, the signal released by Microsoft this time shows that the company hopes to strengthen the emphasis on model safety, corporate autonomy, and social participation mechanisms while promoting the rapid development of artificial intelligence. With Microsoft about to release a code of conduct for self-developed MAI models, the discussion about how superintelligence should develop, who should oversee it, and how to maintain human control is expected to heat up further.
Comments