在本次访谈中，知名人工智慧学者蒂姆尼特·盖布鲁（Timnit Gebru）指出，当前科技界热议的「人工智慧生存风险」与「机器之神」论调本质上是一种行销手段与监管俘虏策略。科技巨头及其投资人一方面夸大技术的潜在威胁与乌托邦愿景以吸引资本和忽视现行法律，另一方面则借此转移大众对现实危害的注意力，掩盖了包括环境破坏、资料窃取以及资料标注劳工剥削等迫在眉睫的真实问题。

面对外界批评其早期提出的「随机鹦鹉」理论已无法适用于现代具备「推理」或「递回智能」的模型，盖布鲁反驳称大型语言模型的基本原理从未改变，依然只是基于机率输出最可能的文本序列。她强调「思维」与「推理」仅是业界的渴望性命名，诸多科学实证研究已证实，只要基准测试稍作微调，所谓的模型推理能力便会瓦解，过度信任自动化系统只会导致误诊等严重现实错误。

针对人工智慧治理与未来发展，盖布鲁主张政府应落实基本且具体的监管措施，包括打击虚假行销、强制要求资料来源透明与完整文档化，以及禁止劳工剥削。她认为真正带来希望的是那些在基层积极抵抗科技巨头叙事、拒绝被动接受现状的人群，她倡导发挥人类能动性与集体力量，致力于创造不破坏环境且造福社区的替代性科技未来。

In this interview, prominent AI researcher Timnit Gebru argues that the prevailing existential risk and machine-god narratives propagated by tech leaders serve as calculated marketing tactics and regulatory capture strategies. Major tech figures and investors hype either dystopian extinction threats or utopian promises to attract capital and bypass existing legal frameworks, which effectively diverts public and regulatory scrutiny away from urgent real-world harms such as environmental destruction, data theft, and the exploitation of data annotators.

Responding to criticisms that her seminal stochastic parrots concept is outdated in an era of purported AI reasoning and recursive intelligence, Gebru insists that the core architecture of large language models remains identical: statistical systems outputting probable text sequences without genuine understanding. She dismisses claims of machine reasoning as aspirational marketing, noting that empirical evaluations reveal model performance crumbles under minor benchmark tweaks, while excessive automation bias fosters dangerous real-world errors such as medical misdiagnoses.

Regarding AI governance and future outlook, Gebru advocates for straightforward, enforceable regulations centered on deceptive marketing prohibitions, full training data documentation and transparency, and labor protections against worker exploitation. Her optimism lies not in elite industry promises, but in human agency and grassroots collective resistance, where independent communities actively build sustainable, community-oriented technological alternatives that reject environmental degradation and corporate extraction.