AI models disagree on which jobs face the biggest automation risks, raising questions about how reliably these tools can predict the future of work.
May 26, 2026 / 11:12 IST
Artificial intelligence
- Study finds AI models disagree on job vulnerability to automation
- Experts warn against relying on AI for career choices
- AI exposure scores vary due to model biases and training data
Did our AI summary help?
A new study has raised questions about how reliable artificial intelligence tools are when predicting which jobs may be most vulnerable to automation. According to a report cited by The Wall Street Journal, researchers found that leading AI models often produced very different answers when asked to assess the impact of AI on the workforce.
The study, published as a working paper by researchers from Northwestern University and American University, examined responses from OpenAI’s ChatGPT-5, Google DeepMind’s Gemini 2.5, and Anthropic’s Claude 4.5. The economists asked the models to rank occupations based on how exposed they are to AI-driven disruption.
The results showed significant disagreements between the systems. In some cases, one model ranked a profession as highly vulnerable while another suggested the same job faced much lower risk. Accountants, advertising managers, and chief executives were among the occupations where the AI systems delivered notably different assessments.
Researchers said the findings highlight the uncertainty surrounding AI-based “exposure scores,” which are increasingly being used by policymakers, businesses, and analysts to estimate how automation may reshape employment. These scores are generally created by evaluating how many tasks within a profession could potentially be handled faster or more efficiently by AI systems.
The paper noted that economists currently rely on three broad approaches to measure AI exposure. One method involves human experts manually rating job tasks. Another uses surveys from workers already using AI tools in their daily work. The third relies on AI models themselves to evaluate which occupations are most exposed.
According to the researchers, each method has limitations. Human evaluations can be subjective, while worker surveys may only represent users of a particular platform. AI-generated assessments introduce another challenge because the models may reflect biases from their own training data.
The study also suggested that professions already using AI extensively may appear more vulnerable simply because those industries generate larger amounts of AI-related training data. Financial analysts, for example, are early adopters of AI tools, which may influence how models judge the future of that profession.
The economists behind the paper said AI-generated predictions should not be treated as definitive guidance for career decisions. Instead, they recommend comparing results across multiple models and combining them with real-world surveys of how AI is actually being used in workplaces.
The findings arrive at a time when governments, employers, students, and workers are increasingly looking for clearer answers about how AI could reshape the global job market in the years ahead.
Invite your friends and family to sign up for MC Tech 3, our daily newsletter that breaks down the biggest tech and startup stories of the day
MC Tech Desk Read the latest and trending tech news—stay updated on AI, gadgets, cybersecurity, software updates, smartphones, blockchain, space tech, and the future of innovation.
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!