Hong Kong built an AI talent ladder. The missing rung is outcome evidence

THE SIGNAL IN ONE SENTENCE

Hong Kong has drawn one of the clearer government diagrams for how ordinary people are supposed to move into an artificial intelligence economy. It starts with public literacy. It climbs through workplace training and university programs. It adds internships, research support and computing infrastructure. At the top sit new companies, better jobs and a city that wants to be an international technology center. That is a proper ladder, at least on paper. The missing rung is evidence that people who step onto it actually reach the next level. On October 5, Hong Kong's Permanent Secretary for Innovation, Technology and Industry, Kevin Choi, used a youth summit speech to assemble several commitments from the 2026 Policy Address into one talent strategy. The starting point was unusually blunt. Choi said routine work that once filled many graduate jobs, including basic coding and first-line customer service, is precisely where AI is already capable. The government's answer is not to protect every old task. It is to make sure young people can still find a first job and that the work becomes better. That is the right problem to name. Entry-level work is not just cheap labor with a nicer title. It is where people learn how an organization actually functions. Junior staff discover which questions matter, how customers behave, what experienced colleagues notice and how apparently tidy processes fall apart on a Tuesday afternoon. If AI removes enough of those starter tasks, a labor market can develop a nasty little trap. Employers want experienced workers, but the jobs that used to create experience have thinned out. Hong Kong's proposed ladder tries to bridge that gap at several levels. The broadest step is the AI for All Inclusive Programme. The Policy Address says the government will roll out more than 200 training courses and activities by the first quarter of 2028 for the public, university students, young researchers and working professionals. The stated goal is responsible use and practical application, not simply a citywide festival of prompt tricks. The next step is workplace training. An upgraded Upskill Hong Kong program is expected to begin an 18-month campaign in the first half of 2027. It will include free online training focused on workplace uses of AI, with advanced short courses available to people who complete the first course. The government expects the initiative to benefit around 40,000 employees. Then comes formal education. Universities funded by the University Grants Committee are introducing 27 undergraduate STEM programs focused on AI, creative technologies and data science. Hong Kong Science and Technology Parks Corporation's expanded Talent Foundry program is meant to offer about 500 training opportunities in AI and emerging technologies each year. Then comes the first-job bridge. The government and business sector plan a two-year program covering 30,000 youth employment and internship opportunities. Within that effort, Hong Kong says it will expand the STEM Internship Scheme to provide around 10,000 internship quotas over two years. Then the ladder reaches research and infrastructure. The Hong Kong Artificial Intelligence Research and Development Institute is expected to begin operating later in 2026. The government is also backing university commercialization programs and exploring a center that would match public and private sector problems with technologies emerging from research parks and institutes. The Sandy Ridge Data Facility Cluster is expected to start phased operation in mid-2027. The government says its computing power should reach 180,000 peta-floating point operations per second by 2032, which it describes as 36 times Hong Kong's current level. Finally, the Policy Address adds a governance layer. Hong Kong plans to create a Commissioner for AI inside the Digital Policy Office. Its risk strategy covers seven areas, including criminal misuse, protection of minors, ethical governance, application safety, liability for AI-caused harm, management of AI agents and the effects of widespread AI use on employment. This is a more complete policy chain than the usual announcement that everybody should learn AI immediately, followed by a link to a chatbot and a ceremonial photograph. The chain still needs a ledger. More than 200 courses is an activity count. Forty thousand employees is a reach target. Twenty-seven degree programs is an institutional commitment. Ten thousand internship quotas is an opportunity count. None of those numbers, by itself, shows that a person gained a useful skill, found a durable job or earned more money. Training programs often look strongest at the entrance. They count registrations because registrations are easy. Completions are harder. Demonstrated skill gains are harder still. Employment outcomes take months to appear, require data sharing across institutions and can expose uncomfortable differences between neighborhoods, income groups, education levels, disabilities, languages and previous work experience. That is exactly why the harder measures matter. For the public courses, Hong Kong should publish enrollment, attendance and completion rates by program and participant group. It should test whether learners can perform useful tasks before and after training, while also recognizing unsafe requests, fabricated answers and privacy risks. For the employee program, the government should report which industries participate, whether small firms can spare workers for training and whether the new skills change actual job responsibilities. A course that helps an employee automate two hours of paperwork each week is useful. A course completed in another browser tab while nothing at work changes is mostly a statistic. For the 27 university programs, the useful evidence begins with seats, applications and graduation. It then follows students into internships, first jobs, wages, job quality and retention. The programs should also show how much teaching involves live systems, responsible-use practice, domain knowledge and projects with real users rather than model demonstrations designed to behave nicely. For internships, conversion matters. How many positions are paid? How many last long enough for meaningful work? How many become permanent roles? Which employers receive repeated subsidies? Are young people learning alongside experienced staff, or cleaning datasets for three months and being thanked for their innovation? The same scrutiny belongs on infrastructure. A large computing target sounds impressive, but students and small companies need allocation rules, prices, queue times, support and predictable access. Compute that technically exists but is reserved for a narrow collection of institutions does not widen the talent ladder very far. Hong Kong also needs to connect its workforce policy with its risk policy. If government departments and employers use AI to screen applicants, assess workers or assign internships, those systems should be tested for error and discrimination. People need notice when automation affects them, a route to challenge a decision and a human with the authority to correct it. The planned Commissioner for AI could make the talent strategy more credible by publishing a common outcome dashboard and a minimum set of safeguards across programs. That office should not merely count how many agencies adopted AI. It should show whether the adoption improved service, reduced drudgery, created new opportunities or shifted risk onto the least powerful person in the workflow. There is a broader lesson here for governments outside Hong Kong. AI workforce policy cannot stop at telling people to adapt. The public cannot retrain its way around missing jobs, expensive education, weak employer demand or inaccessible computing infrastructure. A credible plan connects learning to paid experience, research support, business adoption, compute and rules for the systems changing the labor market. Hong Kong has connected those pieces in policy language. Now it needs to publish the movement between them. How many people started? How many finished? Who was excluded? Which skills lasted? Which internships became jobs? Which jobs paid enough to build a life? Which employers changed work rather than simply cutting headcount? Which public investments produced tools, companies and services that people actually use? The plain signal is simple. Hong Kong has built the outline of an AI talent ladder. The city should judge success by who reaches stable, better work, not by how many feet touched the first rung.