图形学与人机交互2019ise.ncsu.edu编辑精选
自动化的讽刺（Ironies of Automation，1983）
Ironies of Automation (1983)
入门 · 约 30 分钟读完前置知识：自动化系统基础、人机交互概念、过程控制常识
导读
这篇 1983 年的经典论文讨论自动化（尤其过程控制）中的悖论：自动化越先进，人类操作员在异常处理和人工接管中的作用可能越关键；自动化常把常规任务拿走，留下监控、诊断和罕见故障处理等更难任务，并可能造成技能退化与负荷错配。论文由此质疑“全自动”能消除人的角色，强调人机交互和操作员能力设计的重要性。
HN 关注点：是因为四十年后自动驾驶、AI 取代工作和“无人化”系统仍在重复论文描述的困境：自动化没有简单消除人，而是把更难的问题留给人。
后来怎么样
它奠定了“自动化悖论”与人机监督控制的讨论框架，后续人因工程、过程控制和自动驾驶中关于人工接管、技能退化与操作员负荷的研究大量沿用其问题设定。
HN 评论精选
这几条评论里：支持 1 质疑 1 补充 1 亲历 1
支持论文核心洞见：人工接管发生在异常场景，所需技能和负荷反而更高，元层监控可能是不可能任务。
这是我最喜欢的论文之一。我认为最该被关注的一点是：当需要人工接管时，过程很可能已经出了问题，因此需要非常规操作；可以说操作员需要比平均更高而不是更低的技能，需要比平均更轻而不是更重的负荷。自动化完成工作后“剩下”的操作任务更复杂、更怪异，需要比基础任务更多的知识和技能。而且计算机能把系统带入人类不会进入的状态，所以这些技能可能超过纯手动操作所需。做驾驶辅助的人常谈这个，但系统可观测性领域关注不够。另一句：只能指望操作员在元层面监控计算机决策是否“可接受”；如果计算机做决策正是因为人类判断和直觉在此不够用，那该接受哪个决策？人类监控者被赋予了不可能的任务。这在复杂数据库模式的数据完整性上尤其明显：规模一大，变化太快也以人类无法理解的方式变化，人看到异常时，谁来判断……
mjb查看英文原文
This is one of my favorite papers. The core point that I think should get a lot more attention is this one: > When manual take-over is needed there is likely to be something wrong with the process, so that unusual actions will be needed to control it, and one can argue that the operator needs to be more rather than less skilled, and less rather than more loaded, than average. The "remaining" operational work once automation has done it's job is more complex and weirder and requires more knowledge and skill than the basic task. On top of that, computes can (and do!) get systems into states that humans wouldn't, so those skills may exceed the skills needed by manual operators. This is something that people who work on things like driver aids talk about a lot, but I don't see as much attention to in the systems observability field. > One can therefore only expect the operator to monitor the computer's decisions at some meta-level, to decide whether the computer's decisions are 'acceptable', If the computer is being used to make the decisions because human judgement and intuitive reasoning are not adequate in this context, then which of the decisions is to be accepted ? The human monitor has been given an impossible task. This is particularly visible with things like data integrity in a complex database schema. Above a trivial scale, things both change too fast for a human to monitor, and change in ways that don't make sense to humans. When a human sees an anomaly, who's to know if it's expected? Still a great, thought-provoking, read after 37 years.
亲历一线排障经验：论文仍是对工程傲慢的警钟，但现代传感、数据和数字孪生已部分解决其担忧。
作为曾负责给“全自动”“无人化”工厂线“重新开灯”排障的人，我认为这篇论文四十年后仍大体成立——至少可作为对工程傲慢的提醒。工厂故障的长尾仍极难彻底消除。不过，随着计算效率提高和无处不在的连接，文中许多论点其实已被大幅解决。被吹捧的专家操作员那种从高层观察、理解流程的能力（多年观察、听、甚至“感觉”机器行为），确实已被先进机器超越：机器每几毫秒采集完整运行状态的越来越细粒度快照——温度、振动、位置及其他器官/元件的感觉——永久保存，并在不断扩大的因果范围内关联解释。最好的操作员不只是响应问题，还会预测、预防或规划。海量数据、先进物理仿真和制造设备的“数字孪生”可以预先测试几乎无限场景：不仅能模拟把扳手扔进机器，还能模拟扳手以各种角度进入的后果。
wigiv查看英文原文
Speaking as someone who has been responsible for "turning the lights back on" to fix problems with "fully-automated", "lights-out" factory lines, much of this paper still rings true forty years on - if nothing else as a check against our engineering hubris. It remains tremendously difficult to quash entirely the long tail of things that can go wrong in a factory. That said, many contentions raised here really have been resolved substantially with increased computing efficiency and ubiquitous connectivity. The touted expert human operator's ability to see and understand processes from a high-level, informed by years of observing (and hearing, and "feeling") machine behavior has truly been eclipsed by an advanced machine's capacity to collect increasingly granular snapshots of its complete operating state - the temperatures, vibrations, positions, and other sensations of its various organs and elements - every few milliseconds, hold on to that data indefinitely, and correlate and interpret that data in ever-expanding radii of causation. The best human operators (of any technology) not only respond to problems, they anticipate and prevent or plan around them. Massive data, advanced physics-based simulations, and "digital twinning" capabilities of manufacturing equipment afford pre-emptive testing of virtually infinite scenarios. Not only can you simulate throwing a wrench in the works - you can simulate the effect of the wrench entering the works at every possible angle! It's not infallible, and will for a long time still require a human-in-the-loop at some level, but as the author rightly put it themselves near the end of the paper: "It would be rash to claim it as an irony that the aim of aiding human limited capacity has pushed computing to the limit of its capacity, as technology has a way of catching up with such remarks."
质疑反驳：用少数高技能人换大量低技能人并提升产能是净胜利，现代 AI 会进一步改变局面。
是的，不幸的是这篇论文很容易反驳。我只需读摘要和第一段。第一段最后一句：“控制系统越先进，人类操作员的贡献可能越关键。”这完全正确。与在血汗工厂当齿轮的人相比，配合自动化系统工作的人类操作员需要技能更高、受教育更多、更聪明。但这仍不能支持“这扩大了与人类操作员相关的问题”这一论点，原因很简单：你用 5-10 个（或更少）高技能工程师和有经验的技术人员取代 1000 个低技能血汗工人，同时把生产系统吞吐量提高一个数量级以上。这是彻底的胜利。我甚至还没提现代 AI（基于 ML/DL/DRL/CV 等）的潜力和作用，它会把自动化推向下一层次。推荐看 CGPGrey 'Humans Need Not Apply'，看 15 秒处农业劳动力图表就明白我的意思。
fizixer查看英文原文
Yeah, unfortunately it's fairly easy to rebut this paper. I only had to read the abstract and the first paragraph. The last line of the first paragraph: > ... the more advanced a control system is, so the more crucial may be the contribution of the human operator. This is absolutely true. The human operator working with an automation system needs to be way more skilled, more educated, and smarter compared to a human working as a cog in a sweatshop. But this still doesn't support the thesis that it expands the problems associated with human operator, for the very simple reason: You are replacing 1000 low-skilled sweatshop workers with 5-10 (or less) highly-skilled engineers and experienced technicians. And at the same time increasing the throughput of your production system by an order of magnitude or more. It's a complete win. And I didn't even need to invoke the role and/or potential of modern AI (i.e., one based on ML/DL/DRL/CV etc, etc), which is going to take the automation to the next level. Highly recommended watch: CGPGrey 'Humans Need Not Apply' https://www.youtube.com/watch?v=7Pq-S557XQU (in the video just look at the human labor chart regarding agriculture, at 15 second mark, and you'll understand what I'm saying).
补充提供后续回顾文献，说明论文至今仍有现实意义。
一些回顾这篇论文及其持续相关性的较新论文：2017 年的 Ironies of Automation: Still Unresolved After All Years（原文如此？）https://doi.org/10.1109/THMS.2017.2732506（没找到开放获取链接 :/）；2012 年的 The ironies of automation... still going strong at 30? - http://johnrooksby.org/papers/ECCE2012_baxter_ironies.pdf
detaro查看英文原文
Some newer papers looking back at this and it's continuing relevance: 2017: Ironies of Automation: Still Unresolved After All These Years https://doi.org/10.1109/THMS.2017.2732506 (couldn't find an openly accessible link :/) 2012: The ironies of automation... still going strong at 30? - http://johnrooksby.org/papers/ECCE2012_baxter_ironies.pdf
相关资源
从这篇论文所有 HN 讨论帖的全部评论里整理出来的链接。
历次讨论
- 2019Hacker News 讨论帖81 分 · 27 评
- 2020Hacker News 讨论帖66 分 · 11 评
同领域还值得读
像素不是小方块 (1995)
A Pixel Is Not A Little Square (1995)
这篇 1995 年的经典备忘录讨论数字图像中像素应如何理解：像素不是有面积的小方块，而是对连续图像的采样点，重建连续图像需要合适的重建滤波器。
图形学与人机交互2014 · 372 分 · 241 评进阶 · 约 20 分钟编辑精选GDPR 同意框中的暗黑模式
Dark patterns in GDPR consent boxes
研究 GDPR 同意框中的暗黑模式，即网站如何通过界面设计引导用户同意 Cookie 与追踪。
图形学与人机交互2020 · 328 分 · 122 评 · 引用 6进阶 · 约 30 分钟用稀疏条带在 CPU 上实现高性能 2D 图形渲染（硕士论文）
High-performance 2D graphics rendering on the CPU using sparse strips
这篇硕士论文研究如何在 CPU 上实现高性能的 2D 图形渲染，核心思路是用“稀疏条带（sparse strips）”结构来表示和光栅化路径，以降低内存占用并提升渲染速度。
图形学与人机交互2025 · 281 分 · 35 评专业 · 约 90 分钟GDPR 后的暗模式：同意弹窗及其影响
Dark patterns after the GDPR: consent pop-ups and their influence
研究 GDPR 生效后网站 consent pop-ups 中的 dark patterns，以及这些弹窗如何影响用户的同意选择。
图形学与人机交互2021 · 251 分 · 228 评进阶 · 约 60 分钟光线追踪一个微型程序化星球
Ray Tracing a Tiny Procedural Planet
研究如何对一个微小的程序化星球进行光线追踪渲染，属于计算机图形学中程序化生成与渲染的示例。
图形学与人机交互2019 · 247 分 · 9 评进阶 · 约 60 分钟