Artificial Intelligence and Job Loss: What the Evidence Actually Shows
By Aaditya Nair
The increasing use of artificial intelligence (AI) in labour markets leads to two main problems that threaten job security and reduce workforce numbers. Research studies about AI’s effects on job loss show more complex results than what public discussions about AI present through their alternating positive and negative views. The current evidence indicates that AI technology will not create worldwide unemployment, but it has already started transforming the industry by reorganising tasks rather than eliminating entire occupations. So, instead of replacing jobs wholesale, AI systems automate specific tasks within jobs, particularly routine, predictable, and data-driven activities such as scheduling, data entry and customer support. Which, in turn, has caused job losses and social inequalities between different job categories.
A study by Frey and Osborne (2017) estimates that 47% of jobs in the United States are at high risk of automation, particularly those involving routine and repetitive tasks such as administrative work, basic service functions, and transportation. However, subsequent research argues that automation does not typically eliminate entire occupations; instead, AI systems primarily automate specific tasks within jobs, a process known as task automation (Autor & Salomons, 2018). While many occupations continue to exist, automation of key tasks reduces labour demand in those roles, leading to job displacement, deskilling, or wage pressure rather than immediate extinction. Advances in machine learning combined with natural language processing, often referred to as cognitive automation, have extended this process to white-collar occupations that were previously considered resistant to automation, including clerical, legal, and professional services. As a result, automation restructures how work is organised and performed, and although it does not always eliminate jobs directly, it can still lead to job losses and increased inequality by reducing the number of workers required and favouring those with advanced, complementary skills.
Research conducted on a worldwide scale supports these findings. The Organisation for Economic Co-operation and Development (OECD) analysed labour markets across 32 countries and found that around 14 per cent of existing jobs are at high risk of complete automation, meaning that most of the tasks within these occupations could be fully performed by machines (OECD, 2019). However, the OECD also estimates that AI and related technologies will transform, rather than replace, approximately 32 per cent of existing work tasks. This modification involves automating specific components of jobs, such as data processing, scheduling, or routine decision-making, while other tasks continue to be performed by human workers.
The implementation of automation, therefore, produces uneven social effects across different groups within the workforce. Workers who lack digital competencies or access to retraining are more vulnerable to job displacement or downgraded roles, whereas workers with advanced skills can work alongside AI systems, increasing their productivity and earning higher wages. As a result, automation contributes to growing labour market polarisation and widening income inequalities. These effects are further intensified by regional differences in economic development and access to education, leading to increased social separation between workers and rising income disparities between populations.
AI technology also creates new employment opportunities, particularly in data science, software engineering, and AI governance. However, these roles are often difficult to access because they require specialised skills and advanced training, limiting occupational mobility for displaced workers (Acemoglu & Restrepo, 2020). In the absence of effective reskilling policies, firms that adopt AI systems may experience permanent job losses rather than broader job creation, reinforcing existing inequalities rather than offsetting them.
Thus, the evidence suggests that the employment effects of AI are shaped primarily by policy choices rather than technological inevitability. While AI is projected to displace certain jobs, it also has the potential to generate new employment opportunities if supported by appropriate institutional frameworks. For example, the World Economic Forum (2020) estimates that while automation could displace approximately 85 million jobs globally by 2025, it could also create around 97 million new roles, particularly in areas such as data analysis, AI development, cybersecurity, and digital governance. Whether these new jobs offset losses depends on how effectively governments and firms invest in workforce transitions.
The ability of countries to manage AI-related change, therefore, depends on their commitment to lifelong learning, the implementation of active labour market policies such as retraining and job-matching programmes, and the strength of worker protection systems that support displaced employees during transitions. The primary challenge is not preventing technological adoption, but ensuring that the productivity gains from AI are widely accessible rather than concentrated among highly skilled workers. Without inclusive education and reskilling policies, AI risks deepening existing class-based inequalities; with them, it can enhance economic participation and reduce long-term labour market insecurity.
