Melangkaui latihan semula: Pendidikan sepanjang hayat dalam era tenaga kerja luar biasa

At Bett Asia 2026, Professor Keith Lee from the Swiss Institute of Artificial Intelligence (SIAI) delivered a sobering yet constructive analysis of artificial intelligence, workforce displacement, and the future of human learning. Addressing the rise of superhuman labour—where a single adaptive professional using advanced AI tools accomplishes the workload previously requiring 10 to 20 workers—Professor Lee argued that traditional retraining models are fundamentally inadequate.

Rather than focusing on mechanical skill acquisition or ephemeral prompt engineering, educational institutions and corporate learning programmes must pivot towards deep cognitive training and domain-specific knowledge. By strategically upskilling the "middle tier" of the workforce, societies can mitigate impending job losses and achieve an optimal balance between technological adoption and social welfare.

The Reality of Work Displacement: Defining "Superhuman Labour"
The rapid evolution of generative artificial intelligence has fundamentally altered workplace economics. In 2022, early foundation models exhibited variable output quality, leading many organisations to underestimate their disruptive potential. However, massive capital investments have dramatically enhanced model capabilities, giving rise to superhuman labour. A superhuman worker leverages advanced AI tools to amplify output by an order of magnitude, completing tasks that previously demanded a team of 10 or 20 people.

Evidence from Academia and Industry
This structural shift is already observable across research labs and professional firms:

Academic Research Labs: Professor Lee highlighted his own experience managing a research lab. Having previously employed up to 10 student research assistants, he found that 3 AI-proficient assistants could handle the entire research workload. Consequently, scholarship funding for the remaining 7 students could no longer be extended, illustrating the immediate human cost of AI-driven efficiency.

Corporate Restructuring: Senior executives and Managing Directors in professional firms report identical patterns. Energetic early-career employees equipped with AI tools routinely complete the workload of entire cohorts. Facing budget constraints, corporate leadership shows little appetite for maintaining redundant headcounts when single workers paired with AI yield equivalent throughput.

Economic Modelling: AI Adoption vs. Social Welfare
To evaluate the macroeconomic impact, Professor Lee presented a quantitative simulation examining job loss relative to the speed of AI adoption (measured on a scale from 0 to 1.2).

Baseline Displacement: Under standard expert-aided adoption scenarios without educational intervention, the model predicts a baseline workforce reduction of approximately 15%.

The Social Welfare Function: Economists evaluate policy outcomes using social welfare functions that seek to maximise employment and economic stability. Professor Lee demonstrated that the relationship between AI adoption, human learning speed, and social welfare forms a concave curve with an identifiable optimal peak.

The Policy Imperative: Halting AI adoption is neither feasible nor desirable. Instead, the societal objective must be to accelerate human cognitive learning. By aligning human capabilities with technological progress, societies can shift the welfare curve outward, mitigating structural unemployment and preserving job stability.

The Core Framework: Mechanical Knowledge vs. Cognitive Knowledge
To design effective lifelong education programmes, educators must understand the critical distinction between two forms of knowledge:

1. Mechanical Knowledge
Mechanical knowledge encompasses routine problem-solving, standardised code writing, procedural calculations, and rapid information execution.

The AI Advantage: Generative AI tools perform mechanical tasks up to 20 times faster than human capabilities.
The Pedagogical Flaw: Historically, educational systems prioritised execution speed and mechanical accuracy. Today, competing with AI on speed is a losing proposition. Teaching mechanical skills offers rapidly diminishing educational returns.

2. Cognitive Knowledge and Domain Mastery
Cognitive knowledge involves deep critical thinking, contextual reasoning, problem formulation, system architecture, and domain-specific verification.

Model Mechanics and Probabilities: AI systems are probabilistic engines predicting word sequences based on underlying data distributions. If an unguided user provides flawed context or erroneous data, the model outputs inaccurate conclusions.

The Pitfall of Superficial Prompting: Many corporate training initiatives focus narrowly on static "prompt engineering." However, as foundation models iterate from older releases to newer architectures, rigid prompts rapidly become obsolete.

The Necessity of Domain Knowledge: Without deep domain expertise, users frequently misdirect AI tools. Professor Lee shared a personal example from software engineering: lacking formal computer science training, he spent an entire day attempting to resolve minor coding errors due to inefficient prompting. Domain mastery remains essential to verify whether an AI-generated answer is accurate and robust.

Strategic Upskilling: Targeting the "Fourth and Fifth Rung"
When evaluating workforce performance across an organisation, talent typically falls along an ability continuum:

• Top Tier (Rungs 1 to 3): Highly adaptive self-starters who naturally evolve into superhuman workers without formal intervention.
• Bottom Tier (Rungs 9 to 10): Workers who struggle to adapt and face the highest statistical risk of displacement.
• Middle Tier (Rungs 4 to 5): Capable professionals who possess foundational potential but lack the advanced cognitive framework to operate independently at a superhuman level.

The Middle-Tier Solution
Professor Lee argued that the primary focus of corporate upskilling and higher education must be this middle tier. By providing "one-point" targeted educational interventions—revisiting core analytical problem sets, diagnosing specific conceptual roadblocks, and reinforcing underlying domain principles—educators can elevate middle-tier workers to superhuman output levels. This strategy provides organisations with cost-effective, highly productive talent while protecting valuable professionals from systemic unemployment.

Re-Engineering Education: Moving Beyond Rote Learning
Achieving effective lifelong education requires a profound cultural and structural transformation:
Abandoning Exam Score Obsession: Educational cultures that prioritised exam scores over genuine comprehension produce graduates ill-equipped for AI-driven environments.

Emphasising Brain-Teasing Cognitive Rigour: Drawing on his transition from score-focused undergraduate studies in South Korea to analytical graduate training in London, Professor Lee emphasised the necessity of rigorous, problem-based learning that forces students to interrogate assumptions and think deeply.

Continuous Lifelong Coaching: Lifelong education cannot remain a one-time degree. It must function as an ongoing support structure where professionals receive continuous cognitive reinforcement as technological demands evolve.

A Pragmatic Path Forward
Professor Keith Lee's address at Bett Asia 2026 offers a realistic roadmap for the future of work and learning. While AI deployment will inevitably disrupt conventional employment structures, total job destruction is not foregone. By moving beyond superficial retraining, abandoning rote mechanical instruction, and investing heavily in cognitive domain mastery for middle-tier workers, educational leaders and policymakers can secure a productive, human-centred economic future.

Audio and written summaries powered by Gemini Notebook, a free generative AI tool grounded in your provided sources. Try Gemini Notebook for free. You can listen to the audio overview of the session here.