公司法如何促进模型可信与价值对齐

How Can Company Law Facilitate Model Reliability and Value Alignment

合成数据、解释微调、思维树、过程监督等核心关键技术对大模型的“并发式”赋能,加剧了业界对于通用人工智能“奇点降临”的担忧,而具身人工智能的赛道开辟则使得风险提前具象化与泛在化。通用人工智能与具身人工智能的规制结点在于大模型向善,模型可信和价值对齐成为风险防范的“元规则”。较之于面向技术的管制型立法,面向公司的约束型立法具有规范组织行为的效率优势,公司法中蕴含着多类引导人工智能健康发展应用的制度工具。为了促进模型可信,可扩容法人人格否认制度的适用场景、激活公司目的条款、优化董事信义义务内涵并活用双重股权结构。为了促进价值对齐,可引入多样化保护原则、贯彻科技向善理念、建立科技伦理委员会并强化伦理导向的ESG信息披露。

The core key technologies such as synthetic data,interpretation fine-tuning,mind trees,and process supervision empower large models in a "concurrent" manner,intensifying industry concerns about the advent of the singularity of general artificial intelligence,while the development of embodied artificial intelligence opens up avenues for early and ubiquitous risk concretization.The regulatory nodes for general artificial intelligence and embodied artificial intelligence lie in the benevolence of large models,with model trustworthiness and value alignment becoming the "meta-rules" for risk prevention.Compared to technology-oriented regulatory legislation,constraint-oriented legislation targeting companies has the efficiency advantage of regulating organizational behavior.Corporate law contains various institutional tools guiding the healthy development and application of artificial intelligence.To promote model trustworthiness,scenarios applicable to the denial of corporate personhood,activation of corporate purpose clauses,optimization of director fiduciary duties,and utilization of dual share structures can be considered.To promote value alignment,diversified protection principles,the implementation of the concept of technology benevolence,the establishment of a technology ethics committee,and the strengthening of ethical-oriented ESG disclosures can be introduced.