Technology, Policy and You

In Grade 10 you met the old arguments Grade 10: Human and machine intelligence: Turing’s imitation game, the ELIZA effect, and Searle’s Chinese room. You also saw that a chatbot can now pass for a person in a short chat. At Grade 12 the job is harder: to debate these questions well, with each side at its strongest, and to see what follows for how people build and use AI.

Three different questions

Most arguments about “thinking machines” go wrong because they mix up three different questions:

The answers can differ. A calculator is far better than you at arithmetic and, as far as anyone can tell, feels nothing. A newborn baby can do very little, and certainly feels. And however clever a system is, the law today treats it as a tool: responsibility belongs to the people who build and use it.

What machines find easy, and what they find hard

Machines and people are clever in very different ways. A program beat the world chess champion in 1997, but a robot that can walk into an unfamiliar kitchen in Bhaktapur and make a cup of chiya is still very hard to build. In 1988 the roboticist Hans Moravec pointed this out: things that are hard for us, like long arithmetic or chess, are often easy for computers, while things a small child does without thinking, like seeing, walking and picking up a cup, are very hard for them. This is called Moravec’s paradox. One explanation: our seeing and moving were shaped over hundreds of millions of years of evolution, so they feel effortless; chess is a new skill, and we feel every step.

So “intelligence” is not one ladder that machines climb past us, rung by rung. AI researchers separate narrow AI, a system good at one kind of task (reading digits, translating, playing Go), from artificial general intelligence (AGI), a system that could learn and do almost any task a person can. Every system you have built in this course is narrow. Whether today’s large language models are a step towards AGI, and how far away it is, are among the most argued questions in computing, and the people who know the systems best disagree.

Consciousness: the hardest question

Scientists can explain more and more of what a brain does: how it recognizes a face, stores a memory, moves a hand. What is much harder to explain is why any of that is felt: why seeing red feels like something at all. In 1995 the philosopher David Chalmers called this the hard problem of consciousness. Nobody has solved it, which is why nobody can simply check whether a machine is conscious.

People who think hard about it hold, roughly, three positions:

• It depends on what the system does. If consciousness comes from the way information is organized and used, then a machine organized the right way could be conscious, whatever it is made of. This view is called functionalism.
• It needs a living body and brain. Searle and others argue that consciousness is produced by the biology of brains, as digestion is produced by stomachs; a simulation of a brain on a computer would no more feel than a simulated storm makes anything wet.
• We can’t know yet. Without an agreed theory of consciousness, any answer goes beyond the evidence, so we should be careful in both directions: neither declare machines conscious because they talk like us, nor rule it out forever.

Why does it matter? Because ethics treats two roles differently. A moral agent is someone who can be held responsible for what they do: an adult person. A moral patient is someone who can be wronged, whose suffering counts: a baby, a dog, and any adult too. If a machine could ever feel, then it could be a moral patient, and some ways of treating it would be wrong, even if it is not responsible for anything. Notice the order of the questions: what we owe a system depends on whether it feels, not on how clever it is.

Responsibility stays with people

Whatever is decided about consciousness, the law today is clear: an AI system can’t be fined, sacked or sent to prison, and it has no intentions to judge. In Grade 10 you saw a tribunal tell an airline that it was responsible for its chatbot’s promise. But who exactly, among all the people around a system, should answer for it? That is harder than it looks.

The case shows why responsibility for automated systems is hard. Many people made choices: the engineers who wrote the software, the managers who left emergency braking to one person watching for hours, the state that allowed the tests, the driver who looked away. The one person punished was the one closest to the crash, who had the least say over how the system was designed. The researcher Madeleine Clare Elish, in a paper published in 2019, called the human in this position a moral crumple zone: like the part of a car built to crumple in a crash and protect the rest, the nearest person absorbs the blame and protects the system and its makers. Others answer that the driver’s job was exactly to watch the road, and she chose not to. A fair debate weighs both.

Three ideas help you think about who should answer:

• Control. Responsibility should follow the power to make a difference. The people who decide how a system works, how it is tested and when it is released carry more than someone told to watch it for hours.
• Knowledge. Someone who knew, or should have known, of a risk carries more than someone who couldn’t have.
• A person for every decision. Every AI system that affects people should have a named person or office responsible for it, who can be asked to explain and to put things right. A system for which nobody answers leaves a responsibility gap.

In Grade 9 you learned the difference between a law and a policy, judged laws with five questions (the problem, who gains and who pays, whether it is clear, enforced and proportionate), and compared how the European Union, China, the United States and India govern AI Grade 9: AI rules around the world. In Grade 10 you wrote a proposal in seven parts Grade 10: Proposing a change to a law. This section builds on both: you will read Nepal’s own AI policy, compare two ways of governing AI as evidence, and write a proposal that changes it.

What Nepal’s AI policy plans

Nepal’s Council of Ministers approved the National Artificial Intelligence Policy, 2082 in August 2025, Nepal’s first policy made for AI. It was drafted by the Ministry of Communication and Information Technology. News reports of the policy describe its aims as making AI develop and spread safely in Nepal; setting up the institutions, laws and regulations to govern it; making its use ethical, transparent and inclusive in every sector; protecting people’s rights, privacy and security; and helping Nepali IT companies sell AI services abroad. Among its plans, as reported:

• Institutions: an AI Regulation Council, chaired by the communications minister with secretaries from other ministries, to develop standards and rules, and a National AI Centre in the ministry.
• Rules to come: standards for data, algorithms, transparency and accountability; the handling of risky, false or misleading AI-made content, including deepfakes; laws on data protection; and a review of the policy every two years, since AI changes fast.
• People and skills: AI in education from school to university, awareness programs, and training AI specialists (a target of 5,000, reports say).
• Infrastructure and business: data centers (using the cold of the mountains for cooling), faster internet, research, and support for AI startups.

Remember what a policy is (Grade 9): the government’s plan. It binds nobody and punishes nobody. Until the laws it promises are made, a harm done with AI in Nepal falls under the laws that already exist: the Privacy Act, 2075, the Electronic Transactions Act, 2063, and others.

Commentators welcomed the policy and raised the same worries: it names no budget; its council and center need staff who are independent and expert, not only officials; it says little about how its aims will be measured; and Nepal’s data protection is weak (Grade 9 compared the Privacy Act with the GDPR: no independent office, small fines, no duty to report a leak). A policy is a start. What makes it work is the next step: the laws, budgets, offices and checks that put it into practice. That is where a proposal from citizens, and from students, can make a difference.

The tools a policy can use

A policy can choose from many tools, from a light touch to a heavy hand (Figure 8.2). The lighter tools cost little and block nothing, but protect little; the heavier ones protect more, but cost more to follow and enforce, and can block good uses as well as bad. A good policy uses each tool where it fits: light tools for most uses of AI, heavy ones only where the harm could be great.

Two of the tools in the middle are worth knowing by name, because they work well for AI:

• An impact assessment is a set of questions a body must answer before it uses a system: what it decides, about whom, how bad a mistake would be, what data it uses, how it was tested, and who can overturn it. The answers set the level of risk, and the level sets the safeguards. Canada began asking this of its federal government in 2019: before a department uses an automated system to make or help make decisions, it fills in an Algorithmic Impact Assessment, which places the system on one of four levels, from little impact to very high impact, and the safeguards (testing, peer review, notice to the people affected, a human in the decision, explanation) grow with the level. The results are published.
• A public register lists every algorithm a public body uses, what it is for, what data it uses, how people act on what it says, and whom to contact. Amsterdam and Helsinki opened the first city AI registers in September 2020, so that anyone could see where algorithms were used to decide things about them.

Both are cheap compared with a ban, and both make the next step possible: once you know where AI is used and how risky each use is, you can put the heavy tools only where they are needed.

Two ways to govern AI

To propose a change for Nepal, look for evidence of what has worked elsewhere. Compare two of the approaches you met in Grade 9:

Nepal is a small country with few regulators and few AI companies, most of the AI its people use made abroad. That suggests borrowing the ideas that cost little and protect much (levels of risk, impact assessments for the government’s own use, labels, a channel for complaints) rather than copying a whole law written for 27 rich countries.

Writing a proposal that changes a policy

A proposal to change an existing policy or law starts from what it already says. Read it, find the gap, and propose the exact words to add or change. Then use Grade 10’s seven parts, and add two more that a committee will ask about: the evidence from elsewhere, and what it costs and who pays.

A law says what you must do. But most choices at work are inside the law: whether to tell a client about a limit of your system, whether to ship code you haven’t tested, whether to build a feature that tricks people. Professions have their own rules for these choices. Doctors have had them since ancient times; engineers since their bridges and boilers began to kill people. Computing, whose systems now touch every life, has them too.

What a code of ethics is

A professional code of ethics is a statement, written by a profession’s association, of how its members should behave towards the public, their clients and employers, and each other. It is not a law: breaking it isn’t a crime, though it can cost you membership of the association, and in some professions your license. It is not an algorithm either: it doesn’t give an answer to every case, but the principles to weigh. Its greatest use is to give you words, and backing, when you must say “no” to something at work: “this goes against our profession’s code” is harder to brush aside than “I don’t feel good about it.”

The ACM Code

The ACM (Association for Computing Machinery), the world’s largest society of computing professionals, adopted its current Code of Ethics and Professional Conduct on 22 June 2018. It says that it applies to everyone who uses computing in a way that affects others, students included, and that the public good is always the main consideration. Its first section, the general principles, says that a computing professional should:

Section 2 adds professional duties: high-quality work, keeping up your skills, knowing the rules of your work, reviewing and being reviewed, evaluating systems and their risks thoroughly, working only where you are competent (2.6), and building systems that are secure. Section 3 is for leaders: making sure the public good is the central concern of all their organization’s work (3.1), and taking special care of systems that become part of the infrastructure of society, such as payments, health and government (3.7). Section 4 asks members to uphold the Code.

Two passages are worth knowing word for word. Principle 1.2: “A computing professional has an additional obligation to report any signs of system risks that might result in harm. If leaders do not act to curtail or mitigate such risks, it may be necessary to ‘blow the whistle’ to reduce potential harm.” And principle 2.5, written with AI in mind: “Extraordinary care should be taken to identify and mitigate potential risks in machine learning systems.” A whistleblower is someone who reports wrongdoing or danger inside an organization to people outside it, such as a regulator or a journalist, when the people inside won’t act. The Code also warns that careless reporting can itself do harm, so it asks you to assess the situation first.

Three codes compared, and Nepal’s engineers

The two other great codes in computing come from the IEEE, the world’s largest association of electrical and electronic engineers, and the BCS, The Chartered Institute for IT in the United Kingdom. In Nepal, engineers registered with the Nepal Engineering Council, set up under the Nepal Engineering Council Act, 2055, follow its professional code of conduct.

The codes were written on different continents, but they agree on the core: put the public first, be honest, be fair, work within your competence, and speak up about danger. That agreement is itself evidence that these are not one country’s values but the profession’s.

Judging a real case: the cars that cheated

In Grade 10 you met emerging technologies and asked who gains and who is left out Grade 10: The Future of Computing. Now you will design with one: not build it, but plan a solution to a real problem well enough to judge its benefits and harms before anyone spends money. A conceptual design is such a plan: what the solution is, how it works in outline, who it serves, and what could go wrong.

A method

1. The problem, and whose it is. Describe it from the point of view of the people who have it, with evidence: how many, how often, how bad.
2. Why this technology, or none? Compare it with simpler ways to solve the problem. A new technology must earn its place: if a phone call or a better timetable would do, choose that.
3. How it would work, as a drawing of its parts and how they connect: devices, data, people, power, network.
4. The stakeholders: everyone it affects, including people who never use it.
5. Benefits and harms for each stakeholder, now and in ten years: money, jobs, safety, privacy, the environment, who depends on whom.
6. Choices and safeguards: which design choices shift harms and benefits, and which you would choose, and why.
7. A small test: how you would try it small, with what measure of success, before anyone builds it big.

A real design: drones on Everest

The Khumbu Icefall, on the route from Everest’s base camp to Camp 1, is a slope of moving ice, cut by deep cracks, that climbers and Sherpa workers cross on ladders. It is one of the deadliest parts of the mountain, and every ladder, rope, oxygen bottle and bag of rubbish was carried across it on people’s backs.

In April 2024 a Chinese drone maker tested a heavy cargo drone between base camp and Camp 1: it carried about 234 kg an hour, a job for some 14 porters over six hours. It could lift 30 kg from Camp 1, but only 18 kg from Camp 2, higher up, where the air is thinner. In August 2024 the Sagarmatha Pollution Control Committee, Khumbu Pasang Lhamu Rural Municipality and a Nepali drone company, Airlift Technology, agreed to use drones to bring rubbish down. After a trial on Ama Dablam that brought down 641 kg, drones flew about 300 kg of rubbish down from Everest’s Camp 1 in the spring season of 2025, and carried ladders, ropes and oxygen up.

Weigh it with the method. Benefits: fewer trips across the icefall for the people who carry the loads; a cleaner mountain; faster delivery of gear. Harms and risks: porters’ income, since every load a drone carries is a load nobody is paid to carry, which the municipality answered by planning to train Sherpas as drone pilots; dependence on one foreign maker for drones and spare parts; batteries that struggle in the cold; and noise and crashes in a national park. It is a real case of a design whose benefits and costs fall on different people.

In seven years of this course you have learned blocks, Python, the web, databases, networks, AI. Within ten years, some of the tools you learned will have changed, and some new ones will be everywhere. The most valuable skill this course can leave you with is the one that doesn’t date: learning a new tool, language or idea by yourself. Grade 10 gave you a method for learning any new program Grade 10: Learning any new program; here is how to go further.

Read the official documentation

Tutorials and videos are a good start, but the official documentation, written by the makers, is where the exact answers are. Here is a surprise that only the documentation explains:

print(round(2.5), round(3.5), round(-0.5))print(round(2.675, 2))

2 4 02.67

You may have learned at school to round a half up, so 2.5 should be 3, and 2.675 to two places should be 2.68. Python’s built-in help (help(round)) doesn’t explain it either. But the full documentation of round at docs.python.org does: when a number is exactly halfway, Python rounds to the even choice (so round(2.5) is 2 and round(3.5) is 4), which avoids pushing totals upwards when many halves are rounded. And it warns that round(2.675, 2) gives 2.67 “is not a bug”: 2.675 can’t be stored exactly in binary (Grade 11 showed why), and the number really stored is a hair below 2.675. Someone who reads only tutorials meets this as a mystery; someone who reads the documentation finds it explained in two paragraphs.

Official documentation usually has several parts: a tutorial (to start), a reference (every function, exactly), how-to guides (recipes for common jobs), and release notes or a changelog (what changed in each version). When a tool updates, read its release notes: they tell you what breaks and what is new.

A way to keep learning

• Build something slightly too hard for you. You learn most at the edge of what you can do: a project you can finish without learning anything teaches nothing, and one far beyond you stops you. Pick one that needs one or two new things.
• Read code, not only write it. Open-source projects are free textbooks of how professionals write. Pick a small project you use, read how it does one thing, and you will learn its language’s habits.
• Explain it. Write a short note, a blog post or a help answer about what you learned. If you can’t explain it simply, you have found what you don’t understand yet.
• Use AI as a tutor, not as a substitute. Ask it to explain, to quiz you, to find your bug’s cause; then check what it says against the documentation (Grade 10), and write the code yourself until you could do it without help.
• Join others. A club, a local meetup, an online community. Ask good questions: what you tried, what you expected, what happened (Grade 10).

The open-source projects you met in Unit 3 Grade 12: Open source are a school after school as well: a first bug report, a translation into Nepali, a small fix, each reviewed by experienced people for free.

A portfolio

A degree or a mark sheet says what you studied; a portfolio shows what you can do. It is a collection of your best work, each piece presented so that a stranger can understand it in a minute: a university, an employer, a client abroad (Grade 10’s Working online from Nepal).

For each project in it:

• What problem, for whom: one or two sentences.
• What you did, and in a team, which part was yours.
• How it works: a picture or a short video, and the code, with a README (Unit 3).
• Evidence it works: tests, users’ feedback, a measured result.
• What you would do next, and what you learned. Honesty about limits impresses more than claims.
• Help you had, including AI, and the licenses of what you used.

You already have the pieces. This course asked you for a project in every unit: a page about your place (Grade 8), a model built, tested and described (Grade 9), a database designed for a library and a product built in a team (Grade 10), an app people use and a neural network judged (Grade 11), an analysis of open data from Nepal and a model with its model card (Grade 12), and your capstone project still to come. Choose four to six that show different skills, improve them, and present them well.

What you can do, and what you want

One standard for this unit asks you to connect what you have learned to your own goals. Start from the threads of this course, and what each prepares you for:

Computing also leads into fields that are not called computing: a doctor who analyzes data, a farmer with sensors, a lawyer who understands AI, a civil engineer who models floods, a teacher of computing. Many of the most useful people in the next decades will be those who know another field well and can bring computing to it.

Paths after Grade 12

After Grade 12, the main paths are a bachelor’s degree in Nepal; a degree abroad; a diploma or short course leading straight to work; and learning on your own while working. Each can lead to a good career in computing, and many people combine them. Choose by what you want to do, what your family can afford, and what each path really leads to: ask graduates, not only advertisements.