Ethics, Copyright & Licensing
A-Level Computer Science Topic 7 14:06 English narration · English + 中文 subtitles burned in
Chapters
Transcript
Imagine you build an A I tool that sorts job applications ten times faster.
设想你做出了一个人工智能工具，它筛选求职申请的速度快了十倍。
Your manager is delighted.
你的经理很满意。
But you notice something troubling: it quietly rejects more older applicants.
但你注意到一件让人不安的事：它悄悄地更多地拒掉年纪大的申请者。
Do you ship it, and keep your manager happy? Or hold it back, because it treats one group unfairly?
你是把它上线、让经理开心， 还是把它压下来，因为它对某一群人不公平？
This is an ethical decision.
这是一个伦理决定。
And as a computing professional, you will face many just like it.
而作为一名计算机专业人员， 你会遇到许许多多这样的决定。
Your work touches millions of people who cannot check it for themselves.
你的工作影响着千百万人，而他们无法自己去核查这些工作。
So today: why ethics matters, a professional code of conduct, copyright and patents, software licences, and the ethics of A I.
所以今天我们讲： 为什么伦理很重要、一份职业行为准则、版权与专利、软件许可证，以及人工智能的伦理。
Let's begin.
让我们开始吧。
Why does a profession need ethics at all?
一个行业为什么需要伦理？
Two reasons.
有两个原因。
First, trust: users and employers trust you to act in their interest, because they cannot judge the technical work themselves.
第一，信任：用户和雇主信任你会为他们的利益行事， 因为他们自己无法判断这些技术工作做得好不好。
Second, impact: your software may run a hospital machine, a bank, or a car — so careless or dishonest work can hurt real people.
第二，影响：你的软件可能在运行一台医院设备、 一家银行，或一辆汽车——所以粗心或不诚实的工作会伤害到真实的人。
That is why professional bodies publish shared codes of conduct for their members to follow.
这就是为什么专业团体会发布共同的行为准则，供会员遵守。
Ethics is not only about the code you write.
伦理不只关乎你写的代码。
Mass surveillance systems, like these walls of C C T V monitors, raise hard privacy questions: who watches, who stores the footage, and for how long?
大规模监控系统，比如这一整面的闭路电视监视器，会带来艰难的隐私问题： 谁在看、谁保存录像、保存多久？
A professional must weigh those privacy costs against any claimed security benefit.
专业人员必须把这些隐私代价，与所谓的安全收益放在一起权衡。
And computing has a physical cost too.
计算还有物理上的代价。
Discarded phones, laptops and servers become electronic waste — toxic if dumped, and a growing environmental burden.
被丢弃的手机、笔记本电脑和服务器会变成电子垃圾—— 若被随意倾倒就有毒，而且是日益沉重的环境负担。
When you design or ship a system, privacy and the environment sit alongside the technical goals.
当你设计或交付一个系统时， 隐私和环境应与技术目标并列考虑。
Put public wellbeing at the centre of every big decision.
把公众福祉放在每一个重大决定的中心。
Software can affect health and safety — a medical device or a self-driving car.
软件会影响健康与安全——医疗设备或自动驾驶汽车。
It can serve the public interest, or ignore public concerns.
它既可以服务公众利益，也可能无视公众关切。
Benefits such as better access to services must be balanced against those concerns.
更好的服务获取等好处，必须与这些关切相平衡。
Bodies like the B C S, A C M and I E E E publish codes for members whose work reaches far beyond the office.
英国计算机学会、美国计算机协会和电气电子工程师学会等团体为工作远超办公室的会员发布准则。
Ask whose welfare is at stake, not only whether the feature ships on time.
要问谁的福祉会受影响，而不只是功能能不能按时上线。
A code of conduct sets out clear principles.
行为准则列出了清晰的原则。
Put the public interest first: protect the safety of everyone affected.
把公众利益放在第一位：保护每一个受影响者的安全。
Be honest and competent: never claim skills you do not have.
诚实且有胜任力：绝不声称自己没有的技能。
Keep things confidential — confidentiality: guard your client's private information.
保守机密——保密性：守护客户的私人信息。
Avoid conflicts of interest, where your own gain clashes with the client's.
避免利益冲突——也就是你自己的利益与客户的利益相冲突的情形。
And respect intellectual property and privacy.
并尊重知识产权和隐私。
When a decision is hard, check the code and the law, and put users before short-term convenience.
当决定很难做时，查一查准则和法律，把用户放在短期便利之前。
Codes also ask you to keep skills current, treat colleagues fairly, and hold personal data in trust.
准则还要求你保持技能更新、公平对待同事，并把个人数据当作受托保管。
Acting ethically protects users, strengthens reputation, reduces legal risk, and builds trust.
合乎伦理地行事会保护用户、强化声誉、降低法律风险并建立信任。
Acting unethically — skipping testing, hiding bugs, misusing data — can harm users, lead to dismissal or legal action, and erode trust in technology.
不合伦理——跳过测试、隐瞒缺陷、滥用数据——会伤害用户、招致解雇或法律行动， 并侵蚀对技术的信任。
On a borderline decision: identify whose interests are affected, check the code and the law, weigh consequences, ask a senior if you can, and protect users above short-term convenience.
面对模棱两可的决定时：弄清谁的利益受影响，查准则和法律， 权衡后果，如有可能请教前辈，并把用户放在短期便利之上。
Back to the hiring tool.
回到招聘工具。
It sorts C Vs ten times faster, but rejects more older applicants.
它筛选简历快了十倍，却更多地拒掉年纪大的申请者。
Shipping pleases the manager; holding back costs time.
上线让经理满意；压下来费时间。
The ethical choice is to hold it back until the bias is fixed — public interest and fairness before short-term convenience.
伦理选择是压下来，直到偏见修好—— 公众利益与公平优先于短期便利。
Report the finding, fix the data or model, and release only when groups are treated fairly.
汇报发现，修正数据或模型， 只有当各群体被公平对待时才发布。
Now, protecting creative work.
接下来，保护创造性的成果。
Copyright is the creator's legal right to control how their work is copied, distributed, modified and performed.
版权是创作者控制其作品如何被复制、分发、修改和表演的法定权利。
It is automatic — the moment you write code, you own its copyright, with no need to register.
它是自动生效的——你一写下代码，就拥有了它的版权，无需登记。
It lasts a long time, often seventy years after the creator dies.
它持续很长时间， 常常是创作者去世后七十年。
But copyright protects only the expression — the actual code or text — not the idea behind it.
但版权只保护表达——也就是实际的代码或文字——而不保护背后的想法。
A new idea or algorithm needs a patent: you must file for it, and it lasts about twenty years.
一个新的想法或算法需要专利：你必须去申请，它大约持续二十年。
Copyright covers more than code.
版权不只覆盖代码。
It applies automatically to source code, software, documents, images, audio and video — any original expression fixed in a form.
它自动适用于源代码、软件、文档、图像、音频和视频—— 任何固定下来的原创表达。
Without copyright, anyone could copy software freely, the developer would not be paid, and plagiarism would be legal.
没有版权，任何人都可以自由复制软件，开发者得不到报酬， 抄袭也会变成合法。
With copyright, developers can earn from their work, which encourages more software; users know who made it; and re-use happens on the developer's terms through a licence.
有了版权，开发者可以从作品中获得收入，从而鼓励更多软件出现； 用户知道是谁做的；并且再利用必须通过许可证，按开发者的条件进行。
General ideas and algorithms are not covered by copyright.
一般性的想法和算法不受版权保护。
Those may need a patent instead: you must file, it lasts about twenty years, and it protects the invention itself, not just one way of writing it.
那些可能需要专利：你必须去申请，大约持续二十年， 保护的是发明本身，而不只是某一种写法。
A software licence is a contract that says how you may use software.
软件许可证是一份合同，规定你可以如何使用软件。
Four kinds are common.
常见的有四种。
Commercial software — proprietary — you pay for, and the source code stays hidden — you cannot change or share it.
商业软件——专有软件——要付费购买，源代码是隐藏的——你不能修改或分享它。
Open-source gives you the code, free to read, change, and share, though some licences ask you to share your changes too.
开源软件把代码给你，可以自由地阅读、修改和分享，不过有些许可证也要求你把改动一并分享出来。
Freeware costs nothing, but stays closed.
免费软件不收钱，但仍是封闭的。
And shareware is free for a trial, then you pay to keep it.
而共享软件在试用期内免费，之后要付费才能继续用。
Match the licence to the developer's goal: revenue, reach, or community.
让许可证匹配开发者的目标：收入、覆盖面，还是社区。
Look closer at open-source.
再细看开源。
The source is public: users may read, modify and redistribute it.
源代码是公开的：用户可以阅读、修改并再分发。
Permissive licences, such as M I T and B S D, allow almost any use, even inside a closed commercial product.
宽松型许可证几乎允许任何用途， 甚至放进封闭的商业产品。
Copyleft licences, such as the G P L, require modified versions to use the same licence — share-alike.
著佐权许可证要求修改版使用同一许可证——相同方式共享。
The Free Software Foundation and the Open Source Initiative promote these licences.
自由软件基金会和开源促进会推广这些许可证。
Examples: Linux, Python, Apache.
例子：某主流操作系统内核、某编程语言、某网络服务器。
Choose open source when you want the software widely used and improved by the community.
当你希望软件被社区广泛使用并改进时，选择开源。
Commercial software is different: you buy a licence, get no source, and cannot modify or redistribute — used when the developer wants revenue and control, as with office suites, photo editors, and most games.
商业软件则不同：你买许可证、拿不到源码、 不能修改或再分发——开发者要收入和控制时用它，比如办公套件、图像编辑和大多数游戏。
Freeware costs nothing, often may be redistributed, but gives no source and no right to modify — a free document reader or messaging app.
免费软件不收费，往往可再分发，但无源码、不可修改——比如免费文档阅读器或即时通讯应用。
Shareware is free only for a trial, then you pay; source stays closed.
共享软件只在试用期免费，之后付费；源码仍封闭。
Compare the four: commercial is paid with no source; open-source is free with source and, often, modify rights; freeware is free with no source; shareware is trial then paid.
比较四种：商业付费无源码；开源免费有源码， 往往可修改；免费软件免费无源码；共享软件先试用再付费。
To choose: sell or keep control means commercial; share source means open-source; free forever without source means freeware; free trial then pay means shareware.
选择时：要卖或控代码用商业； 分享源码用开源；永远免费且无源码用免费软件；先试用再付费用共享软件。
Link the choice to revenue, reach, or community, and to the user's needs.
把选择与收入、覆盖面或社区，以及用户需求联系起来。
Artificial intelligence builds systems that do tasks once thought to need human intelligence — recognising speech and images, translating languages, playing games, and even driving.
人工智能构建的系统，能完成曾经被认为需要人类智能的任务——识别语音和图像、翻译语言、 下棋，甚至驾驶。
Most modern A I uses machine learning: algorithms that improve at a task by learning patterns from large amounts of data, instead of being programmed step by step.
现代人工智能大多使用机器学习：算法通过从海量数据中学习模式来改进任务表现， 而不是被一步一步编写。
Deep learning is the leading approach today.
深度学习是当今的主流方法。
It uses neural networks with many layers, so the model can discover complex features on its own.
它使用带有许多层的神经网络， 让模型自己发现复杂特征。
Machine learning sits inside A I; deep learning sits inside machine learning.
机器学习在人工智能之内；深度学习又在机器学习之内。
They are nested, not three separate things.
它们是层层嵌套的，不是三个并列的东西。
Everyday A I tasks split into two kinds: understanding input, and producing output or decisions.
日常的人工智能任务分成两类：理解输入，以及产生输出或决策。
Understanding input includes speech recognition — turning spoken words into text for a voice assistant — and image recognition — finding objects, faces or text in a picture.
理解输入包括语音识别—— 把说的话变成文字，供语音助手使用——以及图像识别——在画面中找出物体、人脸或文字。
Producing output or decisions includes machine translation between languages, recommendation systems that suggest products, videos or music, and autonomous vehicles and robots that act in the real world.
产生输出或决策包括语言之间的机器翻译、推荐产品、视频或音乐的推荐系统， 以及在真实世界中行动的自动驾驶汽车和机器人。
Exams often mix several of these in one scenario, so name each skill clearly.
考试常把几种技能混在一个情境里， 所以要清楚地给每一种技能命名。
A classic exam scenario chains three A I skills.
一个经典考题情境把三种人工智能技能串起来。
A program reads a foreign product label with a camera.
程序用摄像头读取一张外国产品标签。
First, optical character recognition finds the printed words in the image.
首先，光学字符识别在图像中找出印刷文字。
Next, machine translation converts those words into the user's language.
接着，机器翻译把这些文字转换成用户的语言。
Finally, text-to-speech reads the translation aloud.
最后，文本转语音把译文读出来。
Each stage is a different A I task: vision to text, language to language, then text to audio.
每一阶段都是不同的人工智能任务：视觉到文字、语言到语言， 再从文字到声音。
Name the three stages in order, and you score the marks.
按顺序说出这三个阶段，就能拿下分数。
Which brings us back to that hiring tool.
这就把我们带回了那个招聘工具。
Modern A I learns patterns from huge amounts of data — so if the data is biased, the A I learns the bias, and makes unfair decisions.
现代人工智能从海量数据中学习模式——所以如果数据有偏见， 人工智能就学到那个偏见，做出不公平的决定。
And there are more concerns: A I uses vast personal data, so privacy suffers; large models are black boxes, hard to explain; and when A I is wrong, who is to blame?
还有更多的担忧：人工智能使用大量个人数据， 隐私因此受损；大型模型是"黑箱"，难以解释；而当人工智能出错时，该由谁负责？
A I also brings real benefits — accessibility, productivity, spotting patterns a doctor might miss.
人工智能也带来实实在在的好处——无障碍、生产力，以及发现医生可能错过的规律。
But a professional must know its limits, warn users, and reduce the harm.
但专业人员必须了解它的局限，提醒用户，并减少危害。
Bias enters A I through a simple pipeline.
偏见通过一条简单的流水线进入人工智能。
Biased training data — for example, years of hiring that favoured one age group — teaches the model that bias as if it were a useful pattern.
有偏见的训练数据——例如多年招聘偏向某一年龄段—— 会把那个偏见教给模型，仿佛它是有用的规律。
The trained model then produces unfair decisions in hiring, lending, policing or medicine.
训练好的模型随后在招聘、借贷、治安或医疗中 做出不公平的决定。
The fix is not just "be nicer": audit the data, rebalance examples, test outcomes across groups, and refuse to ship until the unfair gap shrinks.
修复不只是"态度好一点"：审计数据、重新平衡样本、 按群体测试结果，并且在不公平差距缩小之前拒绝上线。
A professional owns that process, not only the accuracy score on average.
专业人员要为这一过程负责， 而不只看平均准确率分数。
Benefits worth naming: accessibility — speech and image A I help users with impairments, and translation helps non-native speakers; productivity — automating repetitive work; decision support — spotting patterns in huge datasets for diagnosis or fraud; always available and personalised.
值得点名的好处：无障碍——语音与图像人工智能帮助有障碍的用户，翻译帮助非母语者； 生产力——把重复工作自动化；决策支持——在海量数据中发现诊断或欺诈的规律； 始终可用且个性化。
Concerns go beyond bias: job displacement, privacy from training on personal data, black-box transparency, accountability when A I is wrong — developer, user, or operator — and misuse such as deepfakes, misinformation and surveillance.
担忧不止偏见：工作替代、用个人数据训练带来的隐私问题、 黑箱透明度、出错时的问责——开发者、用户还是运营者——以及深度伪造、虚假信息和监控等滥用。
Professionals must understand the limits, inform users, and reduce harm.
专业人员必须了解局限、告知用户并减少危害。
Sort the impacts the way the syllabus does, because a question will ask for one of each.
按考纲的方式给影响分类，因为题目会要求各举一个。
Social: a label-reading program helps people with a visual impairment, people who cannot read the language and people with reading difficulties, and facial recognition at an airport speeds up identity checks — but facial recognition can misidentify people, and it works less well on some groups than others.
社会层面：读标签的程序帮助视力障碍者、看不懂这门语言的人和阅读困难的人， 机场的人脸识别加快了身份核验—— 但人脸识别可能认错人，而且对某些人群的效果不如对另一些人群。
Economic: an A I fault-diagnosis module in a garage diagnoses faults faster and more accurately, so more vehicles are repaired per day and costs fall — but fewer skilled mechanics may be needed, so jobs are lost, and the module itself has to be bought and maintained.
经济层面：修车行里的人工智能故障诊断模块诊断得更快也更准， 于是每天能修更多车、成本下降—— 但可能需要的熟练技工变少，因而有人失业， 而且这个模块本身也要花钱购买和维护。
Three marks to lock in.
三个要拿稳的分。
First, answer ethics questions against a professional code — public interest, honesty, competence — not your personal opinion.
第一，回答伦理题要依据职业准则——公众利益、诚实、胜任力——而不是你个人的看法。
Second, keep copyright and patents apart: copyright protects the expression, a patent protects the invention.
第二，把版权和专利分清楚：版权保护的是表达，专利保护的是发明。
Third, be ready to compare the licence types — commercial, open-source, freeware, and shareware.
第三，要能比较各种许可证类型——商业、开源、免费和共享软件。
Get these right, and this topic is yours.
把这些做对，这个专题就是你的了。
The fixed-wording definitions.
固定措辞的定义。
Ethics: the moral principles that govern how a person behaves — in a profession, the standards set out in its code of conduct.
伦理：支配一个人行为的道德原则——在一个行业里，就是其行为准则所规定的标准。
A code of conduct: the rules an organisation or professional body sets for how its members must behave.
行为准则：一个组织或者专业团体为其成员的行为所设定的规则。
Copyright: the legal right of the creator of an original work to control how it is copied, distributed and modified — and it applies automatically to the work as written, so writing that it must be registered is wrong.
版权：原创作品的创作者控制其被复制、发行和修改的法定权利—— 而且它在作品写成时自动生效，所以写「必须注册」是错的。
A software licence: the legal agreement stating how software may be used, copied and distributed.
软件许可：说明软件可以怎样被使用、复制和发行的法律协议。
Free software: users are free to run, study, change and redistribute it, so the source is available.
自由软件：用户可以自由地运行、研究、修改和再发行，所以源代码是公开的。
Open source: the source is available and may be modified and redistributed under its licence — which does not mean free of charge, and a fee may still be charged.
开源：源代码公开，并且可以在其许可下修改和再发行—— 这并不意味着免费，仍然可以收费。
Shareware: free for a trial period or with limited features, then paid for.
共享软件：在试用期内或者以功能受限的方式免费，之后需要付费。
Freeware: free of charge to use and copy, but the source is not released and may not be modified.
免费软件：使用和复制不收费，但源代码不公开，也不允许修改。
Free software and freeware are not the same thing, and confusing them is the commonest error here.
自由软件和免费软件不是一回事，把这两者混淆是这里最常见的错误。
Two more traps.
还有两个陷阱。
Give a reason with a consequence, not an opinion: "faulty software could harm users, so it must be tested" scores where "it is wrong" does not.
要给出带后果的理由，而不是个人看法： 「有缺陷的软件可能伤害用户，所以必须测试」能得分，「这样做不对」不能。
And name an impact WITH its consequence — job losses scores when it is tied to why, that the AI does the diagnosis so fewer mechanics are needed.
另外，说影响时要连着它的后果—— 「失业」只有在和原因绑在一起时才得分，也就是人工智能做了诊断，所以需要的技工变少了。