ĚÇĐÄVlog

SUTDLogoDark2x

Issue 1: Conventional Job Forecasting vs LLMs-Powered Models

DATE
30 Jun 2025

AI job disruption forecasting — Issue 1

 

There is a wide gulf of opinions on how Artificial Intelligence (AI) will disrupt the job market. In the United States, a found that 73% of AI experts believe AI will positively impact jobs, compared to just 23% of the general public. Meanwhile, China stands out globally for its techno-optimism, with showing it is the most confident country in AI’s potential to create new employment opportunities.

 

As AI continues to evolve at an accelerating pace, even experts struggle to stay current. Perplexity CEO Aravind Srinivas captured this urgency, noting that his company now plans in months rather than years due to the speed of technological change (Chang, 2025). This accelerating momentum exposes the limitations of traditional forecasting methods and underscores the need for more adaptive, real-time approaches to understanding job disruption.

 

The Future of Work research team at the Lee Kuan Yew Centre for Innovative Cities (LKYCIC) brings over a decade of experience collaborating with diverse organisations and has recently begun leveraging AI to advance its job forecasting methods.

 

Traditional job forecasting often requires extensive research cycles that can take two years or more to complete. Much of this analysis depends on prior market data and expert surveys, which may already be outdated by the time the study begins. As a result, by the time forecasts are published, new technologies—particularly in AI—may have already transformed industries and disrupted job markets. This time lag creates significant gaps between forecast predictions and real-world developments, limiting the accuracy and usefulness of conventional forecasts.

 

We leverage current Large Language Models (LLMs) to forecast the automation potential of a role in the education industry.

In the education industry
Role: Assistant Manager/Manager, Student Administration

 

Sample job responsibilities:

  • Coordinate and assist in managing the budget exercise for the Division
  • Monitor and tracking of expenditure against budgets
  • Assist in drafting and reviewing policies, procedures, and manuals for the Division to ensure compliance and efficiency in operations
  • Assist the Group Director in compiling Division plans and documenting outcomes from the departmental annual review

 

Methodology:

We begin by converting job descriptions into a standardised set of Intermediate Work Activities (IWA) based on O*NET classifications, utilising our proprietary universal task translator query system to ensure consistency and accuracy. Once the tasks are properly categorised, we leverage advanced AI models and platforms to conduct detailed forecasts using our proprietary methodology.

 

These models analyse task-level automation potential by drawing on the latest AI capabilities and knowledge sources.

Figure 1: DeepSeek models exhibit the highest automation potential among the models evaluated. The variance between different models suggests the importance of considering both conservative and aggressive scenarios when forecasting AI-driven job disruption.

Figure 2: There is considerable divergence in forecast uncertainty across models. DeepSeek models project a relatively narrow window (2025–2029), implying more confidence in the timing of AI-driven automation. In contrast, models like GPT-4o and GPT-o3 forecast much wider timelines, reflecting greater uncertainty about when AI technologies will reach sufficient maturity and adoption levels to significantly impact the job market.

Figure 3: While the potential for immediate automation is moderately low given the current technology, the high AGI Scenario Value shows that the job tasks can be highly automated in the event AGI is available in the near-term. This highlights an urgency for policymakers and organisational leaders alike to rethink jobs and make them more resilient.

Our approach offers significant advantages over traditional job forecasting methods. Most notably, it delivers results much faster than conventional forecasts, allowing it to keep up with the accelerating pace of AI advances. By harnessing the advanced reasoning capabilities and dynamic search functionality of current leading LLMs, our forecasts reflect the most up-to-date understanding of AI’s evolving impact on the workforce.

 

At the same time, it is important to acknowledge both the value and the limitations of viewing jobs as bundles of tasks. This framework is helpful in recognising that jobs consist of multiple components, some of which can be automated or enhanced by AI (LeCun, 2025). However, this perspective is incomplete. It risks overlooking the nuanced, harder-to-define aspects of work — the interactions, that often occur between or across tasks and are far more difficult to automate (Narayanan, 2025).

 

If you’re interested in learning how to interpret these task-based forecasts in a more holistic way for your organisation, city, or country, feel free to contact us — we would be glad to explore these insights with you.

 

 

References

  • Chang, E. (2025). How fast are AI companies evolving? Check this out. [online] Harvard Business School. Available at: (Accessed 25 June 2025).
  • LeCun, Y. (2025) Most jobs involve performing many tasks, only some of which can be enhanced… [LinkedIn]. Reshared 24 June. Available at: (Accessed: 27 June 2025).
  • Narayanan, A. (2025) I find the story of AI and radiology fascinating… [LinkedIn]. Posted 24 June. Available at:  (Accessed: 27 June 2025).