Dr Stephen Chiu
29 April 2026
In recent years, some academics have actively advocated a “pro-worker” approach to AI development. Their core argument is not a rejection of technology or AI per se, but rather a challenge to the bias against technological advancement over the past few decades—overconcentration of technological development in automation. In other words, machines are used to replace human labour in order to reduce cost rather than to augment human capabilities. (see Note)
In the view of scholars including Daron Acemoglu, an economist at the US Massachusetts Institute of Technology (MIT) and a 2024 Nobel Laureate in Economics; his longtime collaborator Simon Johnson, a co-recipient of the same prize; and David Autor, a notable labour economist at MIT, the digital revolution of recent decades has raised corporate efficiency and returns on capital. Yet at the same time, it has contributed to the loss of middle-class jobs, stagnation in wage growth, and widening gaps in both income and wealth. The problem, in fact, is not technological progress per se but how technology is designed and applied. AI is not necessarily a tool for replacing humans. It can be a force that drives human productivity, creates new tasks, and promotes emerging industries.
How technology generates new tasks
Economic history provides important clues in this regard. Why did the Industrial Revolution, electrification, or the information revolution not result in long-term mass unemployment? The key lies in the fact that, while eliminating old jobs, technology also created many new tasks, which require human judgment, coordination, and creativity.
Electric motors and factory restructuring (1910s and 1920s)
During the steam-engine era, factories relied on a single central power source, with power transmitted through belts and line shafts. Factory layouts were rigid and workers could only follow the rhythm of the machines, while the work environment was dangerous and efficiency was limited. With the widespread use of electric power, each machine was equipped with its own electric motor, and factories could be redesigned according to the production process, thereby giving rise to the assembly-line system. This not only did not lead to the replacement of workers but instead spawned new positions such as equipment maintenance, process management, and quality control. Obviously, technology restructured the production process and the division of labour.
Electric machinery boosts physical performance
Instead of turning construction sites into automated workplaces, tools such as cranes, bulldozers, and electric drills became extensions of workers. Take a heavy machine operator, for example. His output could be tens or even hundreds of times that of a manual labourer. As the economic value created by workers increased, companies often had to pay higher wages under market competition and pressure from trade unions. This is a classic example of “augmenting technology”: humans remain at the core of the production process, with their capabilities remarkedly expanded.
Electronic spreadsheets and transformation of knowledge work
Prior to the 1980s, a large number of bookkeepers were responsible for manually handling general ledgers. Electronic spreadsheets did automate some bookkeeping tasks, but as the cost of computation fell significantly, companies were able to conduct more complex financial modelling and forecasting. As a result, despite the loss of some jobs, more job opportunities were created for accountants, financial analysts, and business consultants. Technology did not lead to a contraction in economic activity, but rather broadened the scope of analysis and decision-making.
As Acemoglu emphasizes, the question is whether contemporary AI will continue to follow this path of “creating new tasks” or simply automate existing tasks, thereby causing a contraction in demand for labour.
Policy initiatives to steer AI development in a pro-worker direction
In the opinion of the three economists discussed above, market forces alone will not necessarily lead to labour-augmenting technologies. Out of considerations of costs and tax incentives, companies tend to opt for labour-substituting solutions. Hence, institutional design is crucial, including the following key points.
- Eliminating hidden tax subsidies for machines
Under the existing tax system, employers are required to shoulder social insurance and healthcare costs but can enjoy tax deductions through depreciation for machinery and software purchased. Such asymmetry is liable to result in “so-so automation”: even when the associated efficiency gains are limited, companies may lay off workers to save on taxes.
The tax system should be reformed to equalize the tax burden on capital and labour, so that corporate decisions are based on productivity gains or losses rather than tax considerations.
- Channelling research funding to support complementary AI
Venture capital favours artificial general intelligence that can completely replace human beings. Acemoglu argues that government funding for scientific research should give priority to projects that complement human skills and create new tasks, as distinct from those that simply pursue automation.
- Granting workers participation rights
If decisions about AI adoption are made only by senior executives and engineers, its purpose will tend to be surveillance and cost cutting. Hence, it is necessary to strengthen trade unions and labour representative systems so that frontline employees can participate in the design and deployment of AI. This would ensure that technology is used to solve real workplace problems, instead of simply becoming a machine for downsizing.
- Antitrust and decentralization of technological power
When AI development is concentrated in the hands of a few tech giants, their business models are likely to lean towards scalability and automation. Through stronger enforcement of antitrust laws and tighter restrictions on mergers and acquisitions, it will be possible to create room for more diverse business models.
- Establishing “data dignity”
AI capabilities are built on texts, code, and creative works accumulated by humans over time. Compensation and revenue-sharing mechanisms should be established to recognize creators’ contributions to training data. This approach would help to prevent AI companies from extracting value from creators’ original works without compensation and ultimately replacing the creators.
- Reshaping the cultural narrative around technology
At present, the technology community regards “human parity” as the ultimate goal. Acemoglu calls for a shift towards “machine usefulness”, using the enhancement of human capabilities as the yardstick for success.
Key criticisms and debates
This set of ideas has sparked widespread discussion in academia and policy circles, while also encountering challenges from multiple stakeholders.
- The blurred line between augmenting and replacing workers
If AI enables an employee to increase their output fivefold, while demand does not grow simultaneously, a company’s rational choice may be to lay off other employees. Micro-level empowerment, therefore, does not necessarily translate into macro-level employment expansion.
- Weakening innovation and national competitiveness
Technology companies and national defence think tanks warn that excessive intervention could slow the pace of innovation. In the global AI race, a country that falls behind in the name of labour protection may be overtaken by its competitors.
- Vision of a “post-labour utopia”
Techno-optimists, including Sam Altman and Elon Musk, argue that the trend towards full automation is both inevitable and worth embracing. Instead of restricting AI, they contend, substantial taxes should be levied and a universal basic income provided to liberate humans from labour.
- A free-market perspective
Neoclassical economists maintain that market mechanisms will eventually create new industries and that governments cannot determine in advance which technologies will prove better. Excessive intervention runs the risk of lowering overall efficiency.
- Radical left critique: ownership is the core issue
Within a capitalist framework, so long as AI ownership lies with capitalists, any technology may over time be used to exploit workers. Thus, genuine reform should aim at public ownership of corporations and AI infrastructure.
Technology is about choice rather than fate
In the face of criticism, Acemoglu et al. emphasize that the development path of technology is not a natural evolution but the result of institutional and policy choices. Acemoglu does not reject outright the idea of universal basic income, but warns that it may create a highly unequal society where “resources are distributed by a handful of technological elites to the public”. Work brings not only income but also social participation and dignity.
This debate has gone beyond a purely technological issue and has become a fundamental debate about political economy and the future of humanity: do we hope to build a post-work society supported by automated factories and welfare transfers, or a society in which the majority can participate through high-skill work and create value? The answer will not only determine the direction of AI development but also shape the economic and political landscape of the 21st century.
Note: Acemoglu, D., Autor, D., and Johnson, S. (2026). “Building Pro‑Worker Artificial Intelligence.” NBER Working Paper No. 34854.
