Maggie Appleton 写设计、人类学和智能体界面时，习惯先画清楚隐喻，再谈工具。把「Empty Pointers and Constellations of AI」整理成可阅读的中文笔记：问题在哪、界面默认会把人带去哪、落地时该改什么。原站导航、评论回响和广告已去掉。

问题怎么被看见

Most people who work in the field of “Artificial Intelligence” rarely use the term. Referring to something as “AI” is a flag you probably don’t know much about the industry.

People in the field instead talk about specific technologies and techniques: machine learning, deep learning, neural networks, natural language processing (NLP), transformers, embeddings, reinforcement learning, large language models (LLM), robotics, latent space diffusion, and generative adversarial networks (GAN).

默认界面在强化什么

AI becomes a catchall term for this wide array of disciplines that are only united by a hand-wavey philosophy about getting machines to act smart. Prominent researchers Kate Crawford and Meredith Whittaker poetically call it a “constellation of technologies.” The AI Now Report , Crawford and Whittaker (2016) 10ya A bit like how “cancer” is the catch-all phrase for a group of diseases – each with its own set of characteristics, prognoses, and treatment opti

Artificial intelligence traditionally means any computational system that mimics human-like intelligence. At the moment we take that to mean understanding language, recognizing patterns, perceiving the world, reasoning, and decision-making. But we end up persistently moving the goalposts on “intelligence.”

可以怎么改

All computing is to some degree artificially intelligent. At one point a TI-83 calculator was artificial intelligence. When Google search first came out it was artificial intelligence. Frankly, a modern refrigerator counts as artificial intelligence: it’s programmed to achieve an objective by taking action and uses environmental feedback to adjust its behaviour. These things are now so banal and pervasive we are unimpressed by their “intelligence.”

AI becomes a stand-in term for whatever technologies and techniques are new, shiny, and just beyond the grasp of our understanding. We use it to gesture at a future we cannot fully comprehend or currently realise. As soon as we do, it will no longer be “AI.”

落到产品里的动作

Many people have pointed to this lack of specific meaning behind the moniker. Here’s an excerpt from a (paraphrased) conversation between Stewart Russell and Sean Carroll on AI as a shifting continuum: Russell wrote one of the original textbooks on AI and has too many accolades in the field to list off. Also wrote Human Compatible – an accessible non-fiction book on AI alignment

Is there some way of definitively saying when a certain computer program is AI versus just a regular computer program? I mean in some sense isn't a pocket calculator optimized to add numbers together correctly?

落地时建议先做的 5 件事

1. 先写清这个工具在强化思考还是在替人思考。
2. 默认交互不要只会谄媚：该追问、该给反例、该要求证据。
3. 智能体界面要露出推理步骤，而不是只给一个光滑答案。
4. 知识发布用可生长的笔记，而不是一次性营销长文。
5. 改产品前先画隐喻：用户以为自己在做什么，系统实际在做什么。

和智能体产品怎么接

龙虾PRO做 OpenClaw 落地时，最该从这类笔记里拿走的是「别把聊天框当唯一界面」。数字员工要能追问、能验证、能把过程摊开，而不是只负责说好话。

本文侧重全链路风控方法论。落地时请用自身业务单据做回放验证，不要把示例阈值直接当生产策略。 相关：风控体检 · 方案资源