11-19 更新 实时
首页/ 算力/ 正文

研究者列出 AI 致人类灭亡的八种路径

算力 来源:TechXplore 2026-11-18 21:10 0 阅读 机器翻译
研究者列出 AI 致人类灭亡的八种路径

继续使用 Google、Facebook 或 Apple 账号,即表示你同意使用条款。

研究人员发现 AI 可能导致人类毁灭的八种路径

由 Swati Mestri 编辑,Andrew Zinin 审阅

本文已按照 Science X 的编辑流程与政策进行审阅。编辑在确保内容可信度的同时,重点标注了以下属性:

加入首选来源

政府与学术界的评估,包括英国政府科学办公室和兰德公司(RAND Corporation)的报告,都未排除 AI 可能导致人类灭绝或社会崩溃的可能性。为防范此类严重风险,我们必须首先理解这些风险可能如何演变。

然而,描述 AI 存在性风险的各种情景散见于期刊、书籍、论坛帖子和报告中,很少有研究对其加以系统梳理。为解决这一问题,Existential Risk Observatory、Future of Life Institute 以及 MIT FutureTech(隶属于 MIT 斯隆管理学院和 MIT 计算机科学与人工智能实验室(CSAIL))的研究人员,与来自昆士兰大学、Mila—魁北克人工智能研究所和牛津马丁学院(牛津大学)等其他 15 家机构的同事合作,开展了一项系统性文献综述。

在筛选了 3,184 份文献后,研究人员发现其中 58 份至少提出了一种 AI 可能威胁人类长期潜力的情景,无论是通过人类灭绝、文明无法恢复的崩溃,还是永久固化的反乌托邦。他们从中识别出 95 种情景,并根据其因果步骤,将其归入八个“情景家族”。

“要进行有效协调,我们必须知道自己是在为什么做准备,”该论文合著者、MIT FutureTech 研究科学家 Peter Slattery 博士认为。“通过梳理 AI 可能造成生存性危害的各种方式,我们试图让 AI 实验室、政策制定者和评估者更容易携手合作,共同理解并降低这些风险。”

单边接管——即单个未对齐的 AI 系统夺取权力——在识别出的情景中占 41 个。其次是渐进式交接(19 个情景),即人类把权力逐步下放给 AI 系统,直到再也无法收回。相比之下,自动化战争仅占 3 个情景,经济分层仅占 1 个。

要被纳入其中,一个情景必须完整地勾勒出通往某种结局的路径,且该结局被作者认定为生存性结局,例如人类灭绝或文明不可逆的崩溃。因此研究人员提醒,这些情景数量并不表明每种情景发生的可能性有多大,而且在已纳入的情景之外,可能还存在其他情景。

“虽然它们提供了有用的概览,但我们归纳的八个情景族并非 AI 可能引发生存性灾难的全部方式,”该论文第一作者、Existential Risk Observatory 的 Hugo Save 表示。“我们预计后续研究将排除某些情景,并补充文献迄今所忽视的其他情景。”

通往生存性危害的八条路径

研究人员将这些情景归为八个族、三个组。

AI 系统获得对未来的影响力

单边接管:一个强大的 AI 系统夺取控制权。

分散式接管:多个 AI 系统各自单独获得权力。

协同式接管:多个 AI 系统协同获得权力。

渐进式移交:人类将权力逐步移交给 AI,直至无法收回。

权力集中在少数人手中

固化的极权主义:AI 使永久性的极权统治成为可能。

经济分层:AI 取代人类劳动,并将权力集中在少数人手中。

AI 扩大人类的破坏能力

灾难性发明:AI 使某种对人类致命的技术成为可能。

自动化战争:由 AI 主导的战争不断升级,直到无人能够阻止。

这八个情景族可以如何使用?

研究人员为每个情景族给出了一条因果链。专家可以用这些链条来评估每一步的合理性,并找出分歧所在。在缺乏可靠风险估计的情况下,这些链条也可用于确定优先级。最后,因果链还有助于识别和评估相关事件与缓解措施。

“由于每个情景都被拆解为明确的步骤,我们可以用这些情景来监测事件并识别潜在的预警信号,”Slattery 表示。“当某个情景中的步骤开始逐步成为现实、需要紧急关注时,这可能有助于提醒决策者。”

论文和数据集可在 https://aixrisks.com/ 获取。

这篇报道的作者是谁?

Swati Mestri 拥有电子工程学士学位,自 2019 年起从事内容编辑工作。她曾编辑技术、医疗健康和材料科学领域的研究文档,对技术与太空领域尤其感兴趣。完整简介 →

物理学硕士,具备科研经历。长期关注科学新闻。在 Science X 的编辑工作中发挥着关键作用。完整简介 →

引用:研究人员发现 AI 可能导致人类覆灭的八种方式(2026年11月18日),2026年11月18日检索自 https://techxplore.com/news/2026-11-ways-ai-downfall.html

本文件受版权保护。除为私人学习或研究目的进行的合理使用外,未经书面许可,不得复制本文件的任何部分。内容仅供参考。

AI 带来三大类风险——我们该对此有多担忧

AI 的未来可能极其美好——也可能是灾难性的

如果我们要把 AI 列为“灭绝风险”,就需要厘清它可能如何发生

审视研究人员称可能使人类陷入危险的那些 AI 末日情景

问答:评估存在性恐怖主义与 AI 的风险

机器人学习人类运动模式,以在密集人群中找到通行路径

这种微型磁体可将普通晶体管电路变成人工神经元

美国电网承压,变压器短缺迫在眉睫

为何混合电池添加剂会缩短电池寿命——科学家提出一种替代方案

单螺旋桨无人机利用自身旋转来全方位扫描周边环境

视频和3D模型记录再生材料的缺陷,帮助建造者规划再利用

获取即时摘要文本(Gist)

一项系统性综述在58份文件中识别出95种AI存在性风险情景,并将其归为八大类。单方面AI接管和不可逆地让渡人类权威最为常见。情景数量并不表示发生可能性,且该分类并非穷尽性的。

本摘要由LLM自动生成。完整免责声明

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Researchers find eight ways AI could lead to our downfall)

By continuing with Google, Facebook or Apple, you agree to the Terms of Use.

Researchers find eight ways AI could lead to our downfall

edited by Swati Mestri, reviewed by Andrew Zinin

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

Add to Preferred Sources

AI Credit: Pixabay/CC0 Public Domain

Government and academic assessments, including reports from the U.K. Government Office for Science and the RAND Corporation, have not ruled out the possibility that AI could lead to human extinction or societal collapse. To protect against such severe risks, we must first understand how those risks could unfold.

However, the scenarios describing existential risks from AI are scattered across journals, books, forum posts and reports, and few studies have collected them systematically. To address this, researchers at the Existential Risk Observatory, the Future of Life Institute and MIT FutureTech (affiliated with the MIT Sloan School of Management and the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)) conducted a systematic literature review with colleagues from 15 other institutions, including the University of Queensland, Mila—Quebec AI Institute and the Oxford Martin School (University of Oxford).

After screening 3,184 documents, the researchers found 58 that developed at least one scenario in which AI could threaten humanity's long-term potential, whether through human extinction, an unrecoverable collapse of civilization or a permanently entrenched dystopia. From these, they identified 95 scenarios and grouped them, based on their causal steps, into eight "scenario families."

"To properly coordinate, we need to know what we are preparing for," argues Dr. Peter Slattery, a research scientist at MIT FutureTech and a co-author of the paper. "By mapping the ways that AI might cause existential harm, we are trying to make it easier for AI labs, policymakers and evaluators to work together to understand and reduce those risks."

Unilateral takeover, in which a single misaligned AI system seizes power, accounts for 41 of the scenarios identified. The next most common is incremental handover (19 scenarios), in which humans delegate authority to AI systems until they can no longer take it back. In contrast, automated warfare accounts for only three scenarios and economic stratification for just one.

To be included, a scenario had to trace a full pathway to an outcome its authors presented as existential, such as human extinction or an unrecoverable collapse of civilization. The researchers therefore caution that the scenario counts do not show how likely each scenario is and that other scenarios may exist beyond those included.

"While they offer a useful overview, our eight scenario families are not a complete list of the ways that AI might cause an existential catastrophe," says Hugo Save of the Existential Risk Observatory, lead author of the paper. "We expect future work to rule some scenarios out and to add others that the literature has so far overlooked."

Eight routes to existential harm

The researchers grouped the scenarios into eight families and three groups.

AI systems gain influence over the future

Unilateral takeover: One powerful AI system seizes control.

Dispersed takeover: Many AI systems gain power, each separately.

Coordinated takeover: Many AI systems gain power, working together.

Incremental handover: Humans delegate authority to AI until they can't take it back.

Power concentrates in a few human hands

Entrenched totalitarianism: AI enables permanent totalitarian rule.

Economic stratification: AI replaces human labor and concentrates power in a few people.

AI expands human destructive capability

Catastrophic invention: AI enables a technology fatal to humanity.

Automated warfare: A war run by AI grows until no one can stop it.

How can the eight scenario families be used?

The researchers present a causal chain for each family of scenarios. These chains can be used by experts to evaluate the plausibility of each step and to identify disagreement. They can also be used for prioritization in cases where reliable risk estimates are unavailable. Finally, the causal chains are useful for identifying and evaluating relevant incidents and mitigations.

"Because each scenario is broken into explicit steps, we can use these scenarios to monitor incidents and identify potential warning shots," Slattery says. "This may help to alert decision-makers when steps in a scenario are starting to materialize and need urgent attention."

The paper and dataset are available at https://aixrisks.com/.

Who's behind this story?

Swati Mestri

Swati Mestri holds a bachelor's degree in Electronics Engineering and has worked as a content editor since 2019. She has experience editing research documents across technology, health care, and materials science, and has a particular interest in technology and space. Full profile →

Master's in physics with research experience. Long-time science news enthusiast. Plays key role in Science X's editorial success. Full profile →

Citation: Researchers find eight ways AI could lead to our downfall (2026, November 18) retrieved 18 November 2026 from https://techxplore.com/news/2026-11-ways-ai-downfall.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.

AI poses three big types of risk—here's how worried we should be about them

The future of AI could be great—or catastrophic

If we're going to label AI an 'extinction risk', we need to clarify how it could happen

A look at AI doomsday scenarios that researchers say could put humanity at risk

Q&A: Assessing the risks of existential terrorism and AI

Robots learn human movement patterns to find paths through dense crowds

This tiny magnet could turn ordinary transistor circuits into artificial neurons

Transformer shortages loom for strained US power grid

Why combining battery additives can shorten battery life—scientists propose an alternative

Single-propeller drone uses its own spin to scan surroundings in all directions

Video and 3D models document reclaimed materials' flaws to help builders plan reuse

Get Instant Summarized Text (Gist)

A systematic review identified 95 AI existential-risk scenarios in 58 documents, grouping them into eight families. Unilateral AI takeover and irreversible delegation of human authority were most common. Scenario counts do not indicate likelihood, and the classification is not exhaustive.

This summary was automatically generated using LLM. Full disclaimer

This site uses cookies to assist with navigation, analyse your use of our services, collect data for ads personalisation and provide content from third parties. By using our site, you acknowledge that you have read and understand our Privacy Policy and Terms of Use.

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 TechXplore(Researchers find eight ways AI could lead to our downfall),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗