Generative AI and the Workforce: What Are the Risks?
30 Pages Posted: 3 Oct 2023
Date Written: September 12, 2023
The unprecedented and rapid diffusion of generative AI represents a major disruption for the labour market, bringing new opportunities and risks within the same transformation. Our paper addresses two significant questions: What are the occupations exposed to generative AI in Australia? What are the risks associated with the diffusion of generative AI within occupations? In this paper, we quantify, map, and analyse worker’ exposure to generative AI and its risks, by measuring their likelihood to manifest within tasks. We provide the first evidence-based analysis of the occupational context of emerging AI risks for workers. This first mapping, within the Australian economy, of exposure of jobs and the workforce to generative AI and associated risks, can support risk mitigation strategies for regulators, employers, and workers. Our mapping uses an innovative research method that combines synthetic data on generative AI and its associated risks with administrative surveys (the Australian Skills Classification and the Australian Labour Force Survey).
Our findings show that the exposure of the Australian workforce to generative AI is widespread and massive. 38.9% of tasks within jobs are directly exposed to large language models, accounting for 36.7% of the time workers spend completing different tasks. When considering the occupational structure of the Australian labour market, 80% of the Australian workforce have 20% of their time allocated to tasks directly exposed to large language models. 20% of Australian workers have 60% of their time allocated to tasks directly exposed to LLMs, which could double to 40% of the workforce with more capable generative AI systems. This overview of Australian workforce exposure shows the potential productivity gains or job displacements that could occur with the diffusion of GenAI.
Our mapping of risk exposure shows that LLMs directly expose 12.4% of tasks to privacy risks, 13.7% to cybersecurity risks, 13.6% to breach in professional standards risks, 14.1% to unethical or harmful bias risks, 10.6% to misinformation and manipulation risks, 26.4% to safety and physical harm risks, 26% to liability and accountability risks and 9.8% to intellectual property risks. Within the Australian workforce, when considering all capabilities of GenAI (including capabilities with complementary software investments and image capabilities), around 20% of the time of workers is allocated to tasks exposed to privacy, cybersecurity, and ethical and bias risks. This exposure rate within the labour force reaches 47.8% for accountability and liability risks and 8.5% for misinformation risks. Our results also demonstrate a potential massive economic impact associated with evolution of competition and possible labour displacements.
Keywords: Generative AI, AI risks, Australian labour market, Synthetic data
JEL Classification: J24, O33, O56
Suggested Citation: Suggested Citation