目前，AI 正在使生物制药和医疗保健产业化，并被应用于从药物设计、诊断到医疗保健交付和后勤功能等各个方面。例如，在医疗保健中，AI 可用于药物设计和诊断等环节。但在监管方面，我们的框架是基于具体情境的，不会为整个部门或技术分配规则或风险级别，而是根据 AI 在特定应用中可能产生的结果进行监管。比如，并非所有关键基础设施中的 AI 应用都被归为高风险，有些用途如识别机器表面的划痕风险相对较低。同样，用于在线服装零售商客服请求分类的 AI 聊天机器人，与作为医疗诊断过程一部分的类似应用，监管方式应有所不同。此外，政府会优先与企业合作，了解如何设计和提供支持以满足其需求，例如通过监管沙盒模式等方式。同时，政府也在持续评估监管机构在不同方面的能力需求。
我们今天正站在这个转折点上。直到现在，医疗保健和生物技术仍然大量依赖服务——由受过专业培训的科学家和[医生](https://a16z.com/2019/06/13/ai-doctor-deep-medicine-topol/)提供——这些服务是算法无法替代的，更不用说为公司增加足够的价值来采纳它们了。但现在，我们正处于一个革命的起点，[AI正在](https://a16z.com/2019/11/19/ai-industrializing-discovery-biology-healthcare/)工业化生物制药和医疗保健，它被应用于从[药物设计](https://a16z.com/2020/05/26/investing-insitro/)和[诊断](https://a16z.com/2017/03/01/going-deeper-into-freenome/)到[医疗保健交付](https://a16z.com/2021/07/12/investing-in-bayesian-health/)和[后勤功能](https://a16z.com/2021/02/09/administration-healthcare-back-office-innovation/)的各个方面。（关于在生物学中应用AI的讨论经常出现的问题或挑战，我在[此处](https://a16z.com/2018/02/28/black-box-problem-ai-healthcare/)解决了医疗保健中AI的“黑箱”问题；并在[此处](https://a16z.com/2021/06/15/ai-is-too-dumb-for-now-2/)解决了我们获取智能[与“愚蠢”]AI的需求问题。）[heading4]但现在，我们正处于一个革命的起点，AI正在使生物制药和医疗保健产业化，并且它被应用到从药物设计和诊
context-specific.83We will not assign rules or risk levels to entire sectors ortechnologies.Instead,we will regulate based on the outcomes AI is likely to generate inparticular applications.For example,it would not be proportionate or effective to classify allapplications of AI in critical infrastructure as high risk.Some uses of AI in critical infrastructure,like the identification of superficial scratches on machinery,can be relatively low risk.Similarly,an AI-powered chatbot used to triage customer service requests for an online clothing retailershould not be regulated in the same way as a similar application used as part of a medicaldiagnostic process.
businesses.As a matter of priority,we will engage with businesses to understand how such anapproach should be designed and delivered to best support their needs.Consultation questions:S1.To what extent would the sandbox models described in section 3.3.4 supportinnovation?S2.What could government do to maximise the benefit of sandboxes to AI innovators?S3.What could government do to facilitate participation in an AI regulatory sandbox?S4.Which industry sectors or classes of product would most benefit from an AI sandbox?Enabling innovation – piloting a multi-agency advice service for digital innovators,Regulators’ Pioneer Fund,2022.The MHRA’s ‘airlock process’ is an example of this kind of service,designed for AI products meeting certain criteria.See:Software and AI as a medical device change programme,MHRA,2022.For an example,see:NHS Innovation Service,Accelerated Access Collaborative,2023.For AI projects,see:AI and DigitalRegulations Service,Care Quality Commission,Health Research Authority,Medicines and Healthcare Products RegulatoryAgency,National Institute for Health and Care Excellence,2023.Pro-innovation Regulation of Technologies Review:Digital Technologies,HM Treasury,2023.A pro-innovation approach to AI regulation3.3.5 Regulator capabilities101.Government has prioritised the ongoing assessment of the different capability needs across theregulatory landscape.We will keep this under close review as part of our ongoing monitoringand evaluation activity.102.While our approach does not currently involve or anticipate extending any regulator’sremit,150regulating AI uses effectively will require many of our regulators to acquire new skills and