Singapore says AI layoffs may happen. The transition promise needs a worker ledger

THE SIGNAL IN ONE SENTENCE

There is a phrase governments love when technology starts chewing through a workplace: nobody will be left behind. It sounds humane. It is also very difficult to audit. Singapore's Prime Minister Lawrence Wong offered a more useful version on October 12. Speaking at the DBS Asia Future Forward Conference, he said the country wants artificial intelligence to benefit all Singaporeans, with safety, human responsibility and broad-based gains treated as non-negotiable. Then he said the part polished technology speeches usually hide. Singapore is not promising zero retrenchments. Wong said some job losses may happen when companies automate. His argument was that the transition should be managed through fair treatment, advance notice, union involvement, retraining, job matching and government support. He also warned that a rush toward mass unemployment would create a backlash against AI. That is a serious statement because it refuses the fairy tale in which every efficiency gain becomes a better job for the same person by Friday. The plain signal is that worker-first AI policy cannot be measured by the number of training seats purchased. It has to be measured by what happens to workers after their jobs change or disappear. Singapore already has more machinery for that job than many countries. Its tripartite model brings government, employers and unions into the same transition system. The Ministry of Manpower, National Trades Union Congress and Singapore National Employers Federation announced a Tripartite Jobs Council in April to coordinate AI-related training, job redesign and transition support. Government programmes offer AI courses, readiness tools and support for companies redesigning work. The country's retrenchment rules also create a basic reporting channel. Employers registered in Singapore with at least 10 employees must notify the Ministry of Manpower within five working days after telling any employee about a retrenchment. The ministry says those notices help public agencies and tripartite partners connect affected workers with employment and training support. That is infrastructure. It is not yet proof. A notice filed after a worker receives bad news does not tell the public whether the worker found another job, whether the new job paid less, whether a training course mattered or whether the company tried to redesign the role before removing it. This is where Singapore can make its worker-first promise unusually concrete. Build a worker-transition ledger. Not a public list of people's names. Not a surveillance file on somebody already having a miserable week. A privacy-protected record of what companies, agencies and support systems did, followed by aggregate outcomes people can inspect. The first line in that ledger should be the reason for the job change. Companies rarely write AI in the little box marked cause. A role is restructured. A team is consolidated. A workflow is redesigned. Productivity improved. The language turns soft just as the consequence turns sharp. An honest system needs categories that can overlap. Was the position removed because an AI tool directly performed tasks? Did automation reduce the number of people needed? Did the company reorganize around a new product? Was the business already shrinking? Did management make a conventional cost-cutting decision and hang an AI ribbon on it? There will be judgment involved. That is fine. The answer should be reviewable rather than magically precise. The second line should show what happened before retrenchment. Did the employer test redeployment? Was the job redesigned? Was the worker offered paid training while still employed? Did the union receive useful notice before the decision became irreversible? Did the company use public subsidies, grants or tax support for its AI transformation? This matters because transition policy can accidentally subsidize both sides of the same event. Public money helps a company automate. More public money helps the displaced worker recover. That may still be sensible, but the full transaction should be visible. If a company receives help to redesign jobs, the public should be able to see whether jobs were redesigned. The third line should follow the worker after the announcement. Training completion is a weak finish line. Re-employment is better. Re-employment at similar or higher pay is better still. A durable job with predictable hours, benefits and room to grow is the result that matters. The ledger should track time to a new role, wage change, occupation change, contract type, hours, job tenure and whether the worker used the skill taught by the programme. Results should be grouped by age, income band, occupation, disability status and other relevant categories while protecting privacy. Those slices matter because average success can hide a bad distribution. A young professional may move from routine analysis into AI operations after one short course. An older administrative worker may complete the same course and still find that every local vacancy asks for experience they have not had a chance to earn. Both count as trained. Only one made the transition. Singapore's own September parliamentary discussion acknowledged the harder cases, including older workers, lower-wage workers and people with disabilities who may not move easily into new occupations even after training. The worker ledger should make those gaps impossible to bury inside a national average. The fourth line should measure advance notice honestly. The current mandatory notification rule requires covered employers to notify the ministry within five working days after affected employees are informed. Employers are encouraged to send an earlier alert while planning a retrenchment, but that early alert is voluntary. Those are two different clocks. One clock helps government count a decision that has already reached the worker. The other may give unions and support agencies time to intervene before the landing. If advance notice is part of the worker-first promise, Singapore should publish how often employers provide it, how much lead time workers receive and whether earlier intervention improves re-employment outcomes. Good intentions become policy when a clock starts. The fifth line should record the quality of training. The Tripartite Jobs Council says SkillsFuture Singapore will identify courses suited to different AI-readiness levels and highlight programmes with evidence of training outcomes and employer demand. That is the right instinct. A course catalogue is not a labor-market strategy. Useful training should connect to a specific role, a real hiring need and a chance to practice the skill. It should publish completion rates, assessment methods and job outcomes. Workers should know whether an employer helped design the course, whether that employer is hiring and what entry requirements stand between the certificate and the job. There is a slightly rude but useful question for every reskilling programme: reskilled for what? If nobody can name the job, the course may be comfort disguised as preparation. The sixth line should track the companies too. Public debate often asks whether workers are ready for AI. Companies should face the mirror. Did managers map tasks before buying tools? Did they consult workers who know the workflow? Did they test whether the system actually saved time? Did the saved time improve service, reduce overload or create new work? Did the company share productivity gains through wages, shorter hours or better staffing? An AI project can be technically successful and organizationally lazy. Buying a model is easier than redesigning a job. Removing a position is easier than rebuilding a career path. The ledger should distinguish companies that invest in transitions from companies that treat the public training system as a cleanup crew. This is not an argument against automation. Singapore is a small economy with an aging population, limited labor supply and a long history of using technology to stay competitive. Automation can relieve dangerous, repetitive or understaffed work. It can help firms grow without finding workers who do not exist. It can also produce new services and occupations. The problem is not that work changes. The problem is when a society celebrates productivity in real time and measures the human cost three years later with a survey nobody reads. Wong suggested that global AI adoption may need to unfold over more than a decade so people and institutions have time to adjust. He contrasted that with the roughly 30 years it took factories to shift from steam power to electricity. The historical analogy should be handled carefully. AI will not enter every occupation at the same speed. A customer-support team can change software in a quarter. A hospital, port or factory may need years of procurement, integration, safety testing and training. Small firms may move later than large banks but leave workers with fewer internal paths when they do. The transition is not one wave. It is weather moving across different neighborhoods. A public ledger helps people see where it is raining. It can also improve policy while there is still time to change it. If one training pathway repeatedly leads to lower pay, redesign it. If early union involvement shortens unemployment, make earlier notice stronger. If a grant helps companies retain and redeploy workers, expand it. If a subsidized tool produces layoffs without measurable productivity or service gains, stop applauding the invoice. The ledger should not wait for perfect data. Start with quarterly, aggregated reporting from mandatory retrenchment notices, early alerts, training programmes and job-matching services. Publish definitions. Mark missing fields. Show the share of workers whose outcomes are unknown. Invite unions, employers, researchers and affected workers to challenge the measures. Then connect incentives to evidence. Companies receiving public AI-transformation support should report task changes, headcount effects, redeployment attempts and wage outcomes. Training providers should report completion and employment results. Agencies should report how quickly support reaches people and where the handoffs fail. None of this guarantees that every worker lands well. That is exactly why the record matters. Singapore's candour is useful because it removes one excuse. The government has already said retrenchments may happen. The next question is no longer whether AI can cost jobs. It is whether the institutions promising a fair transition can show their work. Nobody left behind is a slogan. A worker ledger is a way to find out who was.