Are We Teaching AI to Replace Us? What We Must Not Surrender in the Next Decade by Dr. Alvin Chan

Abstract

As artificial intelligence becomes part of education, work and public services, we need to ask a difficult question: are we using AI to strengthen human judgement, or are we quietly teaching ourselves to depend on it? This article examines three areas: schools, workplaces and government policy. It argues that AI should not be judged only by what it can do. It should also be judged by what people can still understand, access and challenge.

In schools, children should first learn to read, write, calculate, discuss and think for themselves. AI can support learning, but it should not replace the effort that builds understanding. In workplaces, employees may be asked to share knowledge that later helps automate parts of their jobs. Companies should be clear about how such knowledge will be used. Governments should ensure that people affected by AI systems have real ways to question decisions. The article also considers the author’s own use of AI in writing it.

Keywords: artificial intelligence, cognitive offloading, education policy, workforce automation, human oversight, AI governance

1. Introduction

As a professor teaching applied artificial intelligence and business, I believe people should learn to use AI. It can save time, explain difficult ideas and support creative work. But I am also worried by a different question: as we teach machines to do more of our work, are we allowing ourselves to think less, accept less control and become easier to replace?

Over the next decade, I worry about students who can produce polished answers they do not understand. I worry about workers who help automate the very tasks that give them value. I also worry about public institutions that follow algorithmic recommendations because questioning them takes time.

None of these outcomes is inevitable. That is why the choices we make now matter. Schools, employers and governments should not treat AI adoption as progress simply because it is fast or efficient. They should ask what people may lose in the process.

There is also an uncomfortable detail. I used AI to help write this article. Before asking schools, companies and governments to change, I must ask whether I am doing what I am warning others against.

This article has three parts. First, it examines what AI means for education. Second, it considers how workers’ knowledge may be used to build systems that could replace parts of their work. Third, it discusses what governments should do to protect people during this transition.

2. When Help Becomes Dependence

Imagine a business student in Singapore asking AI to write a business proposal for a neighbourhood café. Within minutes, the student may have a market analysis, a pricing strategy and a convincing conclusion. But can that student explain whether the café can afford its rent? Can the student explain why customers would pay the proposed prices?

The assignment may be finished, but the learning may not have happened. This distinction should concern parents, teachers and employers.

Cognitive offloading means passing some mental work to something outside ourselves. We already do this with notebooks, calculators and calendars. There is nothing wrong with using help. The danger is when we repeatedly hand over the very thinking we need to practise.

If AI frames every problem, compares every option and produces every conclusion, what are we still learning to do ourselves? If we never develop enough knowledge to recognise a weak answer, what does “checking the AI” really mean?

I am not saying that AI automatically makes people less capable. I am asking whether we are paying enough attention to the difference between performing better with a tool and becoming more capable ourselves.

3. When Should We Teach AI in Schools?

We should separate two questions: when should children learn about AI, and when should they use AI to complete schoolwork? These are not the same decision.

Children should learn, in age-appropriate ways, that machines can make convincing mistakes. They should also learn that personal information needs protection and that a chatbot is not a person. None of this requires giving young children unrestricted access to a tool that can do their homework.

Singapore’s Ministry of Education has taken a careful approach. Primary 1 to 3 pupils learn basic ideas about AI, but they are not required to use AI chatbots for schoolwork. From Primary 4, supervised use of purpose-built educational AI tools begins in selected classroom tasks (Ministry of Education, 2026).

This is a sensible starting point. However, a starting age should not end the discussion. Schools should keep asking what a child must be able to do independently before using AI for a particular task. They should also ask how they will know whether the tool is strengthening understanding or hiding its absence.

For younger pupils, I would protect reading, writing, number sense, conversation, play and hands-on exploration. These are not old-fashioned skills. They are the foundations on which later learning depends.

For older students, AI should be introduced through guided comparison and challenge. Students could examine a flawed AI answer, check its claims, identify what it leaves out and explain how they would improve it. They should learn to question a machine before being rewarded for producing more with it.

4. Schools Must Teach Thinking, Not Just Prompting

Imagine a Secondary 3 class discussing whether a hawker stall should join a delivery platform. Students could first develop their own arguments. They could then compare these arguments with an AI response and test both against information about costs, customer needs and delivery fees.

The purpose would not be to produce the smoothest paragraph. The purpose would be to improve judgement. Students should learn that a confident answer is not always a correct one.

I would ask students to keep their first attempt before using AI. They should disclose how they used the tool and explain what they accepted or rejected. Schools should also retain work completed without AI, including oral explanations and short written tasks.

A student should sometimes have to face a difficult question without an instant answer. This is not cruelty. It is practice. Learning often happens when we struggle, make mistakes and try again.

The education system should assess understanding without the tool, not only the quality of work produced with it. It should also give teachers time and training, protect students’ data, involve parents and ensure that learning does not depend on a family’s ability to buy a premium subscription.

Every classroom deployment should face an uncomfortable question: is this helping children become independent thinkers, or is it simply making completed work easier to produce? Schools should be prepared to say no when the answer is unclear.

5. Are Employees Training Their Own Replacements?

The concern that workers may be training their own replacements deserves a serious answer, not a comforting slogan about “working alongside AI”. There are real reasons to ask how workers’ knowledge is being turned into systems that may later perform parts of their jobs.

A discussion of the Dockers’ AI Toolkit argues that workers’ corrections, workflow adjustments and other forms of expertise can inform systems that may later repeat aspects of their work (Gallegos, 2026). The issue is not limited to technology companies.

Imagine an experienced customer-service officer in Singapore being asked to correct chatbot replies, explain unusual cases and document how difficult complaints are resolved. Those contributions could help build a useful support tool. They could also make it easier to automate some of that officer’s work.

Not every correction automatically trains an underlying model. Training a model, supplying a knowledge base and designing an automated workflow are different things. But from the worker’s point of view, the practical question remains: is my expertise being used to help me do my job, or to reduce the need for me?

Companies should answer that question plainly. Employees should not have to guess the purpose of a system they are required to improve.

6. Corporations Must Offer More Than Reassurance

If a company asks employees to transfer their knowledge into AI systems, it should explain what is being collected, how it will be used, who can access it and what changes to jobs are being considered. Calling a project “augmentation” is not enough if the business plan assumes fewer people.

Workers and their representatives should have a meaningful role before deployment. They should not merely be invited to comment after contracts have been signed. Employers should discuss limits on surveillance, recognition for training contributions, opportunities for redeployment and how productivity gains will be shared (Gallegos, 2026).

“Reskill yourself” cannot be the whole response. Consider a mid-career worker in Tampines with children, elderly parents and a housing loan. A course offered after working hours is not a credible transition plan unless there is time to learn and a realistic route to paid work.

Companies must also explain how they will develop junior employees if routine entry-level tasks disappear. Where will the next generation gain experience if the work once used to train them is now assigned to machines?

No employer can honestly promise that every job will remain unchanged. But employers can be transparent, fund training during paid time, plan transitions early and avoid treating people as disposable once their knowledge has been captured.

7. Governments Must Protect People, Not Just Promote Adoption

Governments should ask more than how many organisations have adopted AI. They should ask what has happened to wages, job quality, entry-level opportunities and people’s ability to challenge decisions.

For Singapore, government, employers and unions should make these questions central to the transition. Public support for AI adoption should come with credible plans for worker development and safeguards against harm, not only projected cost savings.

Governments should consider clear requirements for transparency when workplace data is used for model training or automation. They should also require appropriate testing, audits and routes of appeal where AI affects employment, education or access to essential services (Gallegos, 2026).

Meaningful human oversight requires more than a person clicking “approve”. That person must have the knowledge, time and authority to disagree with the system.

The same principle applies to a recruitment score or a prediction about a student’s prospects. If nobody feels able to challenge the output, human control may exist on paper while disappearing in practice.

We should also be honest about uncertainty. Nobody can reliably promise how many jobs AI will remove or create over the next decade. Governments should prepare for difficult transitions rather than assume that new opportunities will arrive in the right places, at the right time, for everyone affected.

8. Am I the Pot Calling the Kettle Black?

I cannot write about cognitive offloading as if I stand outside it. AI helped me draft and substantially revise this article, including its structure, wording and examples.

So the criticism is fair. If I warn students against handing over their thinking while accepting AI-written prose myself, I owe them more than a convenient claim that my use is somehow different.

There is a difference between asking for help expressing an argument and allowing a system to decide the argument. But that difference can become an excuse. When AI offers a persuasive sentence, am I accepting it because it reflects what I believe, or because it saves me the effort of deciding?

Even if I can defend the final text, I must ask what I have not practised by using the tool. Writing is not only a way of reporting thought. It is also a way of discovering what we think.

My disclosure does not settle the problem. It simply makes the assistance visible and leaves readers free to judge whether the article meets the standard it asks of others.

I should be expected to explain its claims, check its evidence, reject misleading language and accept responsibility for errors. I should also remain willing to write without AI, rather than allow convenience to become an inability to do so.

Using AI does not automatically invalidate a criticism of AI dependence. But it makes that criticism a personal obligation, not just advice for other people.

9. Progress Should Leave Us with More Control

I do not want schools to produce children who can prompt but cannot reason. I do not want companies to extract workers’ knowledge without giving them a fair say in what follows. Nor do I want governments to mistake rapid adoption for shared progress.

The next decade should not be judged only by what machines can do. We should ask what people can still understand, what work they can still access and which decisions they can still challenge.

AI may become much more capable. That does not remove our responsibility for the institutions we build around it.

Before calling a change progress, we should ask who gains, who bears the cost and who can say no. As an author who has used AI to help make this argument, I must keep asking those questions of myself too.

10. Conclusion

This article has argued that AI integration in education, workplaces and governance requires deliberate attention to what we may surrender if we are not careful. In education, schools should distinguish between learning about AI and using AI to complete work. They should protect foundational skills and emphasise critical thinking.

In workplaces, employees’ knowledge contributions to AI systems raise legitimate concerns about job displacement. Companies should provide transparency and meaningful worker voice. In governance, human oversight must be real rather than symbolic. People need the authority to challenge algorithmic decisions.

The author’s own use of AI in drafting this article shows the tensions in this debate. Using powerful tools does not automatically invalidate concerns about dependence. But it makes those concerns a personal obligation rather than advice for other people.

The next decade of AI adoption should be judged not only by technological capability. It should be judged by what people can still understand, access and contest. Before calling a change progress, we should ask who gains, who bears the cost and who can say no.

References

Gallegos, J. L. (2026, January 21). How can unions negotiate AI in the workplace? Lessons from the port sector. Cornell University ILR School. https://www.ilr.cornell.edu/public-impact/collective-representation-and-worker-voice/how-can-unions-negotiate-ai-workplace-lessons-port-sector/

Ministry of Education. (2026, July 30). What you need to know about preparing our students for an AI-enabled world. https://www.moe.gov.sg/news/edtalks/what-you-need-to-know-about-preparing-our-students-for-an-ai-enabled-world

Appendix A: Summary of Key Recommendations

Domain	Recommendation
Education	Separate “learning about AI” from “using AI for schoolwork”. Protect foundational skills for younger pupils. Use guided comparison and challenge for older students. Assess understanding without the tool.
Workplace	Provide transparency about data collection and use. Give workers meaningful voice before deployment. Fund training during paid time. Plan transitions early. Explain how junior employees will develop.
Government	Require transparency for workplace data used in AI training. Mandate testing, audits and appeal routes. Ensure human oversight includes genuine authority to challenge. Prepare for difficult transitions.

Appendix B: Author Disclosure Statement

Dr. Alvin Chan used AI assistance to help draft and substantially revise this article, including its structure, language and examples. The author remains responsible for the final text. This assistance, and the tensions it raises, are discussed openly in Section 8 of the article.