AI 与机器学习medium.com编辑精选
人工智能：革命尚未发生（2018）
Artificial Intelligence – The Revolution Hasn’t Happened Yet (2018)
2018 — 2022，它一次次回到首页
黑点是上首页（≥ 50 分）的那次投稿，空心点是没冲上去的投稿，点越大分数越高。
导读
这是 Michael I. Jordan 2018 年发表的文章，讨论为什么当前所谓 AI 革命其实尚未真正到来：人们常把机器学习、深度学习与“人工智能”混为一谈，而实际进展更多是智能增强和智能基础设施。文章主张把 AI 当作工程学科来建设，像土木工程一样关注可靠性、数据溯源、模型生命周期和治理，而非只追求更大数据与更强算力。
为什么经久不衰：它经久不衰，是因为在每一轮 AI 热潮里都提醒我们：演示能力不等于可靠系统，真正的进步取决于工程化、数据来源与全生命周期验证。
HN 评论精选
点出数据溯源与智能增强/智能基础设施，比空泛的 AI 叙事更贴近实际价值。
文章提到的问题不只是数据分析本身，还有数据库研究者所说的“provenance”（数据溯源）：数据从何而来、得出了哪些推断、这些推断与当前情境有多相关。训练有素的人也许能逐案判断，但设计行星尺度的医疗系统时，不能依赖这种细致的人工监督。我也喜欢“智能增强”（IA）这个说法；我工作过的几家公司爱在营销里滥用 AI，而实际更有价值的是通过“智能基础设施”（II）实现的 IA，II 往往只是给原本只开放给一两个用户的数据库加一层网页视图。
klenwell查看英文原文
The problem had to do not just with data analysis per se, but with what database researchers call “provenance” — broadly, where did data arise, what inferences were drawn from the data, and how relevant are those inferences to the present situation? While a trained human might be able to work all of this out on a case-by-case basis, the issue was that of designing a planetary-scale medical system that could do this without the need for such detailed human oversight. I'm not a data scientist and I've never encountered that term "provenance" before but I've encountered the problem he talks about in the wild here and there and have searched for a good way to describe it. His ultrasound example is a great, chilling, example of it. I also like the term "Intelligence Augmentation" (IA). I've worked for a couple companies who liberally sprinkled the term AI in their marketing content. I always rolled my eyes when I came across it or it came up in say a job interview. What we were really doing, more practically and valuably, was this: IA through II (Intelligent Infrastructure), where the Intelligent Infrastructure was little more than a web view on a database that was previously obscured or somewhat arbitrarily constrained to one or two users.
机器学习工程化必须持续验证分布漂移，不能假设部署环境长期不变。
主流机器学习对分布外数据很脆弱，这是最基本的错误来源之一；很少有领域能长期依赖静态环境。若要把机器学习当成工程学科，就需要建立贯穿模型生命周期的验证实践。Automunge 这类开源数据管道库可自动评估特征集的数据属性漂移（披露：我是 Automunge 创始人）。
nicholast查看英文原文
The brittleness of mainstream ML to out of distribution data is one of the most fundamental channels for error. There are very few domains where a static environment can be depended on over the long term. If machine learning is to be approached as an engineering discipline there will need to be practices established for validating models throughout their life cycle. One potential resource that can support this type of systematic evaluation is the Automunge open source library for assembling data pipelines, which has automatic support for evaluating data property drift in feature sets serving as basis for a model. (disclosure I am founder of Automunge)
深度学习不等于自适应智能，工程范式虽合理却可能与大数据商业逻辑冲突。
“自适应智能”或许可描述为：接受少量指令、收集信息并采取行动完成任务的能力，也就是下属做的事。但深度学习几乎相反：它从巨大的现有数据流出发，外推数据中蕴含的判断，能复制人类分类和决策，却不涉及自适应智能；它“泛化”了什么，数据科学家也说不清。文章提出用“工程”范式替代当前的“智能”范式，这更合理，但我怀疑能否被接受：神经网络 AI 像是无限数据收集意识形态的补充，限制 AI 该做什么就会限制大数据的好处。
joe_the_user查看英文原文
How would one put it? "Adaptive Intelligence" might be described as the ability to be given a few instructions, gather some information and take actions that accomplish the instructions. It's what "underlings", "minions" do. But if we look at deep learning, it's almost the opposite of this. Deep learning begins with an existing stream of data, a huge stream, large enough that the system can just extrapolate what's in the data, include data leads to what judgements. And that works for categorization and decision making the duplicates what decisions humans make or even duplicates what works, what wins in a complex interaction process. But all that doesn't involve any amount of adaptive intelligence. It "generalizes" something but our data scientists have no idea exactly what. The article proposes an "engineering" paradigm as an alternative to the present "intelligence" paradigm. That seems more sensible, yes. But I'm doubtful this could accepted. Neural network AI seems like a supplement to the ideology of unlimited data collection. If you put a limit on what "AI" should do, you'll put a limit on the benefits of "big data".
多数 AI 能力只在演示中成立，真实场景的可靠性和适用范围被严重高估。
今天“AI 革命”的主要问题是：实践者声称它能做的 95% 的事情，在真实世界所需的条件范围内都远不够准确；超过一半甚至在手工挑选样例的演示之外根本不工作，而似乎没人在意。这把适用范围缩到能容忍高错误率的极小缝隙，即便在那里也很难做，AI 沦为“锦上添花”而非融入整体。少数领域性能足够实用，其中一个（人脸识别与追踪）正在中国引发“革命”，但我不确定那是我们想要的革命。
throwaway84742查看英文原文
The main problem with “AI revolution” today is that 95% of things that practitioners tell you it can do are just nowhere close to sufficient accuracy under the range of conditions they’d need to work in in the real world. Easily more than a half of them don’t even work at all outside a demo with hand picked examples. And nobody seems to care. This narrows down the scope of applicability to a tiny sliver where you can tolerate high error rates, and even there it’s a hard slog to get anything done, and AI is relegated to the role of the “cherry on top”, rather than baked into the pie. There are a few areas where performance is good enough to be practically useful. One of those (facial recognition and tracking) is currently causing a “revolution” in China. I’m just not sure it’s the kind of revolution we want.
历次讨论
同类还值得读
如何写一个拼写纠正器
How to Write a Spelling Corrector
Peter Norvig 用一个几十行的 Python 程序演示如何做拼写纠正：先用编辑距离（删除、插入、替换、换位）生成所有可能的候选词，再用从大量文本中统计出的词频概率，选出…
AI 与机器学习上首页 10 次 · 2007–2022norvig.com编辑精选面向黑客的概率编程与贝叶斯方法
Probabilistic Programming and Bayesian Methods for Hackers (2013)
《黑客的贝叶斯方法与概率编程》是一本开源的贝叶斯统计入门书，以 Python/PyMC 代码和交互式 Jupyter Notebook 为主线，强调“先实践、后数学”的学习路径，…
AI 与机器学习上首页 5 次 · 2013–2018camdavidsonpilon.github.io编辑精选神经网络黑客指南
Hacker's Guide to Neural Networks
Andrej Karpathy 写的神经网络入门教程，面向程序员，从最简单的表达式求导讲起，逐步推导出反向传播和梯度下降，最后引出神经网络。
AI 与机器学习上首页 4 次 · 2014–2019karpathy.github.io编辑精选神经网络能逼近任意函数的可视化证明
A visual proof that neural nets can approximate any function
这一章来自 Michael Nielsen 的在线书《Neural Networks and Deep Learning》，用 sigmoid 单元叠加成局部隆起/平滑阶跃的图示…
AI 与机器学习上首页 4 次 · 2014–2022neuralnetworksanddeeplearning.com编辑精选论乔姆斯基与统计学习的两种文化
On Chomsky and the Two Cultures of Statistical Learning
Norvig 回应 Chomsky 对统计语言模型批评的文章，围绕“统计学习能否解释语言”展开，比较生成与解释、少参数模型与 Shannon 式高参数统计模型。
AI 与机器学习上首页 4 次 · 2011–2025norvig.com编辑精选