从一个赛博磕头的玩笑开始，我们认真做了一次 AI 公益A cyber-kowtow joke led us to take AI for good seriously
关于一个荒诞创意的诞生，以及我们如何重新理解技术、善意与真实需求。How an absurd idea began, and how it changed the way we think about technology, good intentions, and real needs.
- CyberCare
- AI 公益
- 信息可及性
- 产品验证
- CyberCare
- AI for Good
- Information Accessibility
- Product Validation
01｜公益项目的起点，居然是磕头
最开始，这个项目其实不怎么严肃。
朋友提出了一个很有意思的想法：做一个网站，用户捐款之后，屏幕里的虚拟人就会向他磕头致谢。
一个赛博世界里的电子磕头机。
听起来有点荒诞，甚至带着某种互联网特有的抽象幽默。
我当时觉得很好玩。
但笑完之后，又忍不住多想了一会儿。
为什么一个虚拟人的夸张动作，会比一段普通的感谢文字更容易引起注意？
为什么我们愿意围观一个奇怪的互动，却可能直接划过一条真正需要帮助的公益信息？
如果一个原本无人问津的公益话题，因为一个略显荒诞的设计而被更多人看见，这算不算一种有价值的尝试？
我开始意识到，这个想法真正有趣的地方，或许根本不在于磕头。
而在于它提出了一个问题：在注意力越来越稀缺的互联网里，我们怎样让一件值得被关心的事情，先被人看见？
这成了后来 CyberCare 的起点。
02｜如果不收钱，我们还能做什么？
顺着最初的创意，我们很容易想象出一套完整的产品流程。
用户捐款，虚拟人致谢，生成分享卡片，再通过社交传播吸引更多人参与。
从产品设计的角度看，它似乎说得通。
有情绪反馈，有互动机制，也有一定的传播潜力。
但我逐渐对其中一个环节产生了疑问。
为什么一定要从捐款开始？
捐款当然是一种帮助，但它不是帮助的全部。
很多时候，一个人遇到的困难，并不需要别人付出金钱。
可能只是看不清一张通知。
可能是听不懂一段语音。
也可能是面对手机上密密麻麻的按钮，不知道下一步该按哪里。
这些问题很小，小到很难成为一个吸引人的公益故事。
但它们又足够具体，足够频繁。
尤其是对于老年人、残障人士，或者不熟悉数字设备的人来说，我们习以为常的操作，并不总是理所当然。
这让我想起之前做适老化产品时的一些思考。
技术发展得越来越快，界面越来越精致，功能越来越丰富，可为什么还有人正在被这些变化悄悄落下？
我们常常把效率当作技术进步的重要标准，却很少追问：
对于那些原本就跟不上数字世界的人来说，技术的进步意味着什么？
一个功能可能让熟练用户节省几十秒，却也可能让另一群人多出好几个不知道如何跨越的步骤。
于是，我们逐渐把方向从「如何让捐款变得有趣」，转向了另一个问题：
如果我们不急着向用户索取什么，而是邀请他们花一点时间，把一件小事做得更容易呢？
如果 AI 公益不一定要从宏大的社会议题切入，而是先解决一些日常生活中真实存在、却容易被忽视的小麻烦呢？
这个方向，后来成为 CyberCare 更重要的部分。
03｜从一张看不清的海报开始
在构思产品时，我发现一个很常见、却容易被忽略的现象。
许多活动通知、社区公告和公益招募，仍然以图片海报的形式传播。
对于大部分人来说，点开图片、放大、阅读，似乎并不困难。
可是，如果图片里的文字太小呢？
如果一个人依赖读屏软件，而海报没有提供可以直接读取的文字呢？
如果时间、地点和报名方式散落在复杂的视觉设计里，连普通人都需要仔细寻找呢？
这些信息明明是公开的，却不代表每个人都能轻松获取。
于是，我们决定从一张海报入手。
让用户上传海报，尝试利用 AI 提取其中的活动信息，再将其整理成更加清晰、便于阅读的结构化内容。
这个功能并不复杂。
图片文字识别早已不是新鲜技术，大模型也可以快速完成信息整理。
但当我们真正开始思考使用流程时，新的问题出现了。
如果 AI 把日期识别错了怎么办？
如果活动地点被截掉了一部分怎么办？
如果海报本身就没有提供完整信息呢？
AI 很擅长生成一个看起来完整的答案。
可对于真实用户来说，一个错误的时间，可能比没有时间更糟糕。
所以，我们在原型中加入了人工核对的环节。
识别结果不是最终答案，而是一份等待确认的草稿。
对于不确定的信息，系统应该明确标记；对于原图没有提供的内容，也不必强行补全。
用户可以对照原图修改和确认，再导出方便阅读的版本。
当然，这并不意味着人工核对之后的信息就一定准确。
它只是让我们尝试建立一个更可靠的过程。
我越来越觉得，一个好的 AI 产品，不只是能够回答问题，也应该知道什么时候需要承认自己不知道。
这种能力听起来没那么酷。
但在一些真实场景里，它可能比生成答案的速度更加重要。
04｜那个会磕头的虚拟人，后来怎么样了？
它没有完全消失。
在 CyberCare 的概念原型里，我们依然保留了一个略显荒诞的虚拟角色。
它可以与用户进行一些简单互动，也可以承担部分讲解和引导的任务。
但它不再是最初设想中的捐款致谢机器。
这个变化让我重新思考了幽默在公益产品中的位置。
公益一定要严肃吗？
我觉得未必。
如果一段有趣的互动，可以让原本对公益不感兴趣的人愿意多停留三十秒，那么它可能就有价值。
但如果所有人的注意力都停留在虚拟人的表演上，真正需要解决的问题反而被忽略，那么这个设计也就偏离了初衷。
更重要的是，幽默不能建立在被帮助者的尴尬或者弱势处境之上。
我们可以让技术表现得荒诞一点，却不应该把真实的人变成玩笑。
所以，我更希望这个角色是在替一个项目吸引注意，而不是替任何真实的人请求怜悯。
用户也不应该被迫接受那些不需要的声音、动画和互动。
幽默可以是入口，但不能成为使用产品的前提。
我挺喜欢这个原则。
一个公益产品可以不那么一本正经，但它对人的态度应该是认真的。
05｜那些没有想象中那么酷的工作
随着想法逐渐展开，我们又尝试加入了短音频转写、截图操作指导等功能。
它们的共同目标，是让一些原本不方便阅读、理解或者操作的信息，变得更容易处理。
但在整理产品流程的过程中，我发现，我们花费精力思考的很多事情，并不是那些特别有科技感的部分。
反而是一些琐碎的问题。
识别失败了怎么办？
用户上传的图片包含个人隐私怎么办？
AI 整理过的信息与原始内容不一致怎么办？
如果自动检测通过了，是否就意味着这份内容真的容易阅读？
用户不愿意上传图片时，是否还有别的解决方式？
这些问题很难成为演示视频里最吸引人的部分。
但它们恰恰决定着一个工具究竟是在提供帮助，还是制造新的麻烦。
以前我做项目时，很容易被「功能实现」带来的成就感吸引。
一个按钮可以点击了。
一个接口成功调用了。
一个原本只存在于脑海里的想法，终于变成了可以运行的页面。
这些时刻确实让人兴奋。
可后来我逐渐发现，功能跑通只是很小的一步。
真正困难的事情，往往发生在这之后。
当用户不知道该怎么办时，产品能不能提供清楚的指引？
当 AI 出错时，用户有没有办法发现并纠正？
当系统处理不了某个问题时，它能不能坦诚地说明自己的边界？
这些都不是单纯增加功能就能解决的。
尤其是当一个产品试图帮助更容易遇到数字障碍的人时，我们更需要谨慎。
不能因为出发点是善意，就忽略设计本身可能产生的新问题。
善意不能代替验证，技术也不能代替理解。
这大概是 CyberCare 带给我的一个重要认识。
06｜做出了一个原型，不等于解决了问题
后来，CyberCare 有了一个可以访问的网页原型。
它包含海报信息整理与核对、音频转写、截图操作指导等尝试，也保留了那个最初来自玩笑的虚拟角色。
从一个随口提出的想法，到真正做出可以交互的页面，这个过程本身让我很开心。
但我不想把它描述成一个已经成熟的无障碍产品。
我们还有很多事情没有完成。
例如，真实用户究竟会不会使用？
他们在什么场景下需要这样的帮助？
我们的交互是不是足够清晰？
识别错误会不会给他们带来更大的麻烦？
这套工具是否真的比他们已经使用的解决方式更方便？
这些问题都需要进一步验证。
而且，公益项目有一个很容易让人陷入的误区。
当我们说自己在帮助某个群体时，很容易默认这个项目已经具有了某种社会价值。
但实际上，想要帮助别人，和真正帮助到了别人，是两件不同的事情。
一个项目的社会价值，不应该只由开发者的初衷来证明。
还需要真实的使用反馈、可以观察的改善，以及对潜在负面影响的持续关注。
至少在目前这个阶段，CyberCare 更接近一个关于 AI 公益的产品实验，而不是一个已经得到验证的解决方案。
承认这一点，并不会让我觉得它失去了意义。
相反，这意味着我们还有机会去发现真正的问题。
而不只是不断证明最初的想法是正确的。
07｜磕头只是开场
回头再看，那个最初关于虚拟人磕头的创意，和现在的 CyberCare 已经相差很远。
一个原本围绕捐款反馈的趣味互动，慢慢变成了一次关于信息可及性、AI 辅助和产品责任的探索。
这个变化并不是因为最初的想法不够好。
相反，我仍然很喜欢它的荒诞感。
因为有些问题，就是从一个不那么正经的念头开始的。
我们经常习惯先评价一个想法是否专业、是否合理、是否具备商业价值。
但我越来越觉得，也许在急着判断之前，可以先允许它存在一会儿。
问一问，它到底为什么有趣？
它触碰到了什么真实情绪？
它为什么会让人愿意停下来？
如果把其中最夸张的部分拿掉，还剩下什么值得认真对待的东西？
当然，并不是每一个奇怪的想法都值得做成产品。
也不是每一份善意，都一定能产生真正的帮助。
但如果一个玩笑，最终让我们认真研究了一件原本容易被忽略的小事，那么这个玩笑至少没有白开。
这也是我现在越来越喜欢做产品的一个原因。
它允许一个想法从很不成熟的地方开始，却要求我们在真正面对问题时，逐渐变得认真。
我们可以有天马行空的想象，可以做一些奇怪的尝试，可以允许自己偶尔不那么正经。
但当这些想法真正涉及某个人的生活，我们就需要对它们负责。
至于 CyberCare 未来会变成什么样，现在还没有答案。
也许我们会找到真正需要它的人，继续完善某个具体场景。
也许经过验证后，会发现现有方案并不合适，需要彻底改变方向。
也可能这个项目最终只停留在一次短暂的实验里。
我觉得都可以接受。
毕竟，一次尝试的价值，不一定只能通过它最后变成了什么来衡量。
有时候，它也在于我们因为它而注意到了什么，理解了什么，以及开始认真对待什么。
虚拟人可以先磕一个。
真正的事情，还是得由我们站起来做。
2026 年 10 月 · CyberCare 项目手记
这是一篇关于 AI 公益概念原型的个人记录。项目仍处于探索与验证阶段，文中涉及的产品设计不代表已获得正式无障碍认证或完成真实用户效果验证。
01 | An AI-for-good project that began with a kowtow
At first, this project wasn’t particularly serious.
A friend had an amusing idea: build a website where, after someone donated, a virtual character on screen would kowtow to thank them — kneeling and bowing its head to the ground.
An electronic kowtow machine in cyberspace.
It sounded absurd, with the kind of surreal humor that seems to belong to the internet.
I thought it was funny.
But after laughing, I kept thinking about it.
Why would an exaggerated gesture from a virtual character attract more attention than an ordinary thank-you message?
Why are we willing to watch a strange interaction, yet scroll past a charitable cause that actually needs help?
If a slightly absurd design brought attention to a cause that otherwise went unnoticed, would that be a worthwhile experiment?
I began to realize that the interesting part of the idea might have very little to do with kowtowing.
It was the question behind it: on an internet where attention is increasingly scarce, how do we first get people to notice something worth caring about?
That became the starting point for CyberCare.
02 | What could we do without asking for money?
Following the original idea, it was easy to imagine a complete product flow.
A user donates, the character thanks them, a shareable card is generated, and social sharing brings in more participants.
As a product design, it seemed to make sense.
There was an emotional reward, an interaction, and some potential for spreading the word.
But I gradually started questioning one part of it.
Why did we have to begin with a donation?
Donating is one way to help, but it isn’t the only way.
Often, the difficulty someone faces doesn’t require another person to spend money.
Perhaps they simply can’t read a notice clearly.
Perhaps they can’t make out a voice message.
Or perhaps they’re looking at a phone screen crowded with buttons and have no idea what to tap next.
These problems are small — too small to make an attention-grabbing charity story.
Yet they are concrete, and they happen often enough to matter.
For older adults, people with disabilities, or people unfamiliar with digital devices, the actions we take for granted aren’t always straightforward.
This brought me back to questions I’d had while working on products for older adults.
Technology keeps moving faster. Interfaces look more polished and offer more features. Why, then, are some people quietly being left behind by these changes?
We often use efficiency as a measure of technological progress, but rarely stop to ask:
What does technological progress mean for people who already struggle to keep up with the digital world?
A feature might save an experienced user a few dozen seconds while adding several steps that someone else has no idea how to get through.
So we gradually moved from asking “How can we make donating fun?” to a different question:
What if, instead of rushing to ask users for something, we invited them to spend a little time making one small task easier?
What if AI for good didn’t have to start with a grand social issue, but could begin with small, real difficulties in everyday life that are easy to overlook?
That direction became a more important part of CyberCare.
03 | Starting with a poster that’s hard to read
While thinking through the product, I noticed something common but easily overlooked.
Many event notices, community announcements, and volunteer recruitment messages are still shared as poster images.
For most people, opening an image, zooming in, and reading it doesn’t seem difficult.
But what if the text is too small?
What if someone relies on a screen reader, and the poster provides no readable text?
What if the time, location, and registration details are scattered across a busy design, so that even people without particular access needs have to search carefully?
Information can be public without being easy for everyone to access.
We decided to start with a single poster.
Users would upload a poster, and we’d try using AI to extract the event details and organize them into a clearer, more readable structure.
The feature itself wasn’t especially complex.
Recognizing text in images is hardly new, and language models can organize information quickly.
But when we started thinking through how someone would actually use it, new questions appeared.
What if the AI got the date wrong?
What if part of the venue name had been cropped out?
What if the poster never included all the necessary information in the first place?
AI is very good at producing an answer that looks complete.
For someone trying to attend an event, though, the wrong time could be worse than no time at all.
So we added a human review step to the prototype.
The extracted result would be a draft awaiting confirmation, rather than a final answer.
The system should flag uncertainty clearly, and it shouldn’t fill in details missing from the original image just to make the result look complete.
Users could compare the draft with the original, make corrections, confirm the details, and then export a readable version.
Of course, human review doesn’t guarantee that the information is accurate either.
It is simply our attempt to build a more reliable process.
I increasingly feel that a good AI product should be able to recognize when it needs to admit it doesn’t know, as well as answer questions.
That ability doesn’t sound particularly cool.
But in some real situations, it may matter more than how quickly an answer appears.
04 | What happened to the kowtowing character?
It didn’t disappear completely.
We kept a slightly absurd virtual character in CyberCare’s concept prototype.
It can offer some simple interactions and take on part of the explanation and guidance.
But it is no longer the donation thank-you machine we first imagined.
That change made me rethink the place of humor in a product intended to help people.
Does work for a good cause always have to be serious?
I don’t think so.
If an amusing interaction persuades someone who wouldn’t normally care about a cause to stay for another thirty seconds, it might have value.
But if everyone’s attention stays on the character’s performance while the actual problem is overlooked, the design has lost its purpose.
More importantly, the humor mustn’t come at the expense of the people we’re trying to help, their embarrassment, or their vulnerable circumstances.
We can let the technology behave absurdly. We shouldn’t turn real people into the joke.
I’d like the character to draw attention to the project, rather than ask for pity on anyone’s behalf.
Users shouldn’t have to put up with sounds, animation, or interactions they don’t want, either.
Humor can be an invitation, but it mustn’t be a requirement for using the product.
I like that principle.
A product for a good cause can be playful while still taking people seriously.
05 | The work that wasn’t as cool as we’d imagined
As the idea developed, we also experimented with short-audio transcription and guidance based on screenshots.
They share a goal: making information easier to deal with when it is difficult to read, understand, or act on.
But while working through the product flow, I found that much of what occupied our attention wasn’t especially futuristic.
It was a collection of small, practical questions.
What happens when recognition fails?
What if an uploaded image contains private information?
What if the AI’s organized version disagrees with the original?
If an automated check passes, does that mean the result is actually easy to read?
If someone doesn’t want to upload an image, is there another way to help?
These questions rarely make the most exciting part of a demo video.
Yet they determine whether a tool is helping or creating new trouble.
In earlier projects, I was easily drawn to the satisfaction of getting a feature to work.
A button became clickable.
An API call succeeded.
An idea that had existed only in my head finally became a working page.
Those moments are exciting.
But I gradually found that getting a feature to run is only a small step.
The harder work often comes afterward.
Can the product offer clear guidance when someone doesn’t know what to do?
When the AI makes a mistake, can the user spot it and correct it?
When the system can’t handle a problem, can it explain its limits honestly?
Adding more features alone won’t answer these questions.
We need to be especially careful when a product is intended for people who are more likely to encounter digital barriers.
Good intentions don’t excuse us from noticing the new problems our design might create.
Good intentions cannot replace validation, and technology cannot replace understanding.
That is one of the lessons CyberCare has brought home to me.
06 | Building a prototype doesn’t mean we’ve solved the problem
Eventually, CyberCare had a web prototype people could visit.
It includes experiments in extracting and reviewing poster information, transcribing audio, and guiding actions from screenshots, along with the character that began as a joke.
Taking a casual suggestion and turning it into an interactive page made me happy.
But I don’t want to describe it as a mature accessibility product.
There is still a lot we haven’t done.
Would real users actually use it?
In what situations would they need this kind of help?
Are the interactions clear enough?
Could recognition errors create bigger problems for them?
Is the tool actually more convenient than the solutions they already use?
All of this needs further validation.
There is also a trap that projects for good causes can easily fall into.
When we say we’re helping a group of people, it’s easy to assume the project already has social value.
But wanting to help someone and actually helping them are two different things.
A project’s social value shouldn’t be established solely by its developers’ intentions.
It also needs real feedback, observable improvements, and continued attention to possible negative effects.
At this stage, CyberCare is closer to a product experiment in AI for good than a validated solution.
Acknowledging that doesn’t make the project feel meaningless to me.
It means we still have a chance to discover the real problem.
Instead of spending all our time proving that the first idea was right.
07 | The kowtow was just the opening act
Looking back, the original idea of a kowtowing character is quite far from what CyberCare has become.
A playful interaction built around thanking donors gradually turned into an exploration of information accessibility, AI assistance, and responsibility in product design.
That change doesn’t mean the first idea wasn’t good enough.
I still like its absurdity.
Some questions begin with a thought that isn’t very serious.
We’re often quick to judge whether an idea sounds professional, reasonable, or commercially useful.
But I increasingly think it might be worth letting an idea exist for a while before rushing to judge it.
Why is it funny?
What real feeling does it touch?
Why does it make someone stop and pay attention?
If we took away its most exaggerated part, what would remain worth taking seriously?
Of course, not every strange idea deserves to become a product.
And not every good intention leads to real help.
But if a joke leads us to look carefully at a small problem we might otherwise have missed, it hasn’t been wasted.
That is one reason I enjoy making products more and more.
An idea can begin somewhere immature, but as we face the actual problem, we have to become more thoughtful about it.
We can imagine freely, try strange things, and allow ourselves to be a little unserious now and then.
When those ideas start touching someone’s life, though, we need to take responsibility for them.
There is no answer yet to what CyberCare will become.
Perhaps we’ll find people who need it and keep improving one specific use case.
Perhaps testing will show that the current approach doesn’t fit, and we’ll need to change direction completely.
Or perhaps the project will remain a brief experiment.
I can accept any of those outcomes.
The value of an attempt doesn’t have to be measured only by what it eventually becomes.
Sometimes it also lies in what we noticed because of it, what we came to understand, and what we began to take seriously.
The virtual character can go ahead and kowtow.
The real work still needs us to get up and do it.
October 2026 · CyberCare project notes
This is a personal account of an AI-for-good concept prototype. The project is still being explored and evaluated. The designs described here do not imply formal accessibility certification or completed evaluation of outcomes with real users.