Ece Kocabıçak
Senior Lecturer in Sociology, The Open University, UK
Youngjin Yoo’s approach to artificial intelligence (AI) appears to be informed by the classical political economic perspective when addressing questions regarding the origin of value and the achievement of sustainable development. Therefore, I believe that Yoo’s work is useful for analysing the implications of workplace AI for social inequalities in the labour market. In this piece, I will provide a brief overview of his work.
Yoo begins from a dissatisfaction with the dominant public debate. He does not deny the importance of concerns about safety, alignment, surveillance, disinformation or job loss. However, he argues that these concerns do not go far enough. For Yoo, the deeper risk is not simply that AI may replace human labour, but that it may prevent human beings from growing. His approach therefore shifts the question from what AI will do to jobs, firms or markets to what kind of people, organisations and forms of value will emerge from sustained interaction with generative technology.
Is unemployment the biggest risk posed by AI?
Yoo’s answer is that this fear is not wrong, but incomplete. Job loss matters because work is connected to income, dignity, social status and everyday security. Yet if the debate focuses only on employment, it risks treating human beings as workers who must either be protected, retrained or compensated. Yoo’s concern is broader: whether AI expands or constrains what people can imagine, attempt and become.
This is why he introduces the idea of ‘Ghost GDP’ or ‘phantom GDP’. This describes a scenario in which AI increases productivity and economic output while ordinary people’s purchasing power, capabilities and lived opportunities decline. In such a future, national accounts may show growth, but that growth does not translate into human flourishing. Yoo’s argument sharpens this point. If generative AI is used only to produce more outputs at lower cost, abundance may become socially empty. The real question is whether GenAI creates new human possibilities or merely accelerates the replacement of human contribution.
The difference between ‘being loved’ and ‘being lovely’
Yoo uses Adam Smith’s distinction between ‘being loved’ and ‘being lovely’ to express the moral core of his argument. Much of today’s AI helps people become ‘loved’: it helps them produce the email that gets a reply, the CV that gets attention, or the post that receives approval. It optimises people for external response. But becoming ‘lovely’ is different. It means becoming someone whose work, character and life are genuinely worthy of love, respect and appreciation. For Yoo, AI should not simply make people more visible, efficient or appealing to others. It should help them become more capable, imaginative and worthy of the futures they seek to build.
The visual is created by Ece Kocabıçak using ChatGPT (June 2026)
A key distinction in Yoo’s approach is between ‘vacuum-cleaner AI’ and ‘scaffolding AI’. Vacuum-cleaner AI is designed to remove work. It performs tasks that humans previously carried out: writing emails, summarising documents, producing reports, analysing information or generating routine outputs. This kind of AI may be efficient, but it does not necessarily make the human user more capable. Once the task has been automated, the person may be no more skilled than before. Indeed, the value of their time may fall.
Scaffolding AI works differently. It does not simply replace what a person used to do. Instead, it helps them do something they could not previously do. It expands the frontier of human action rather than compressing it. In Yoo’s account, this is the kind of AI that matters most for human growth. It turns technology into a support structure through which people can learn, experiment, solve new problems and develop new capabilities. The value of AI then lies not in producing the same thing more cheaply, but in generating situated support that helps people move beyond their current capacities.
Yoo uses the example of Paul Conyngham, a data scientist in Sydney whose rescue dog Rosie developed cancer. Conyngham was not a medical researcher, but he used ChatGPT to learn molecular biology and AlphaFold to model protein structures. He then worked with a university lab to help produce a personalised vaccine. The significance of this example, for Yoo, is not only that Rosie’s condition improved. It is that Conyngham became capable of doing something previously beyond his reach.
This example illustrates Yoo’s broader argument that AI can enable people to approach problems they would previously have regarded as inaccessible. When people solve urgent personal problems in this way, they may also create knowledge, tools or possibilities that others can use. Yoo calls this a ‘generative externality’: one person’s growth can create opportunities for others.
The key principles of a better AI design
For Yoo, access is necessary but insufficient. Training and AI literacy are also important, but they are not enough. The deeper issue is the design of the technology itself. AI should not merely give people tools; it should help them grow through using those tools. This requires asking what AI does to the person over time. Does it reveal new capacities? Does it help the user understand their strengths and weaknesses? Does it support their aspirations? Or does it simply optimise their output for the objectives of a platform, employer or market?
Yoo’s argument adds another layer. Since generative AI produces outputs in relation to context, the quality of the context matters. Human-growing AI must know something about the person using it: their history, records, work, choices, aspirations and unfinished possibilities. But this raises a crucial architectural question: who owns and controls the model of the person? If the model of the user is held by a platform whose objective is commercial extraction, then the technology may not serve the user’s growth. It may instead turn the user into an input for someone else’s optimisation system.
Yoo suggests three principles for better AI design. First, people should control their own AI models and personal data. Second, their data should not be locked into one platform; they should be able to use it across different tools. Third, AI should support different forms of human growth, rather than pushing everyone towards the same idea of success.
To conclude, Yoo’s approach reframes the politics, ethics and economics of AI. The central question is not only whether AI is safe, aligned or productive. It is whether AI helps human beings become more than they were before. The runtime revolution makes this question even more urgent. If value is increasingly generated through abundance, context and variation, then societies must decide whether this abundance will serve human growth or merely automate human relevance away. The future Yoo advocates is one in which AI does not simply replace human effort, but expands human imagination, capability and collective possibility.