Artificial Intelligence (AI) has the potential to significantly benefit or harm humanity. At present, a few for-profit companies largely control the development and use of this technology, and therefore determine its outcomes. In an effort to diversify and democratize work on AI, various groups are building open AI systems, investigating their risks, and discussing their ethics. In this paper, we demonstrate how blockchain technology can facilitate and formalize these efforts. Concretely, we analyze multiple use-cases for blockchain in AI research and development, including decentralized governance, the creation of immutable audit trails, and access to more diverse and representative datasets. We argue that decentralizing AI can help mitigate AI risks and ethical concerns, while also introducing new issues that should be considered in future work.
翻译:人工智能(AI)有可能大大造福或伤害人类。目前,少数营利性公司在很大程度上控制了这一技术的开发和使用,因此决定了它的结果。为了实现AI工作的多样化和民主化,各团体正在建立开放的AI系统,调查其风险,并讨论其道德问题。在本文件中,我们展示了闭锁技术如何促进和正式确定这些工作。具体地说,我们分析了AI研发中障碍环节的多重使用案例,包括分散管理、建立不可改变的审计线索以及获取更多样化和更具代表性的数据集。我们说,分散AI可以帮助减轻AI的风险和伦理问题,同时也提出了在未来工作中应考虑的新问题。