AI’s Shadow Looms: A Controversial List of Jobs at Risk of Automation
A recent foray into the potential impact of artificial intelligence on the Australian workforce has ignited debate, after a prominent figure from OpenAI created and subsequently removed a list detailing jobs most vulnerable to AI automation. Andrej Karpathy, a co-founder of the company behind ChatGPT, initially compiled the list by leveraging AI to analyse extensive data from the US Bureau of Labor Statistics’ (BLS) Occupational Outlook Handbook. This comprehensive dataset encompasses approximately 143 million jobs across the American economy.
Each occupation was assigned a score ranging from 0 to 10, representing its “AI exposure.” A higher score indicated a greater likelihood of a job either being replaced by AI or significantly integrating AI into its daily tasks. The analysis unearthed a surprising correlation: jobs with higher remuneration tended to receive worse average scores, suggesting a greater susceptibility to AI’s influence. Conversely, individuals earning less than $35,000 annually showed the lowest exposure levels.
The data pointed to roles such as software developers, data scientists, and financial analysts as having the highest exposure scores. In stark contrast, occupations like construction workers, barbers, and nursing assistants were identified as having very low exposure.
Following the initial publication of his findings over the weekend, Mr. Karpathy withdrew them from his website. He cited a widespread misinterpretation of the threat posed by AI as the primary reason for their removal. In a post on the social media platform X, he explained his intention was to provide a useful tool for others to visually explore the BLS dataset, allowing for different interpretations and visualisations.
“It’s been wildly misinterpreted (which I should have anticipated even despite the readme docs) so I took it down,” Mr. Karpathy stated. He elaborated that the “exposure” was scored by a large language model (LLM) based on how “digital” a job is, and stressed that this metric has no direct bearing on the actual future of these occupations, which are influenced by a multitude of factors including demand elasticity and other complex economic drivers.
Despite their temporary removal, Mr. Karpathy’s findings have since been reinstated and, interestingly, echo conclusions drawn from earlier significant reports concerning AI’s influence on the job market.
Echoes of Previous Research: A Pattern Emerges
A 2023 study also originating from OpenAI highlighted a similar pattern, identifying comparable office-based roles as having high AI exposure, while those involving physical labour were ranked as having the lowest. This suggests a consistent trend in how AI’s potential impact is being assessed across different analyses.
More recently, a report from Anthropic, the creator of the AI model Claude, published just this month, further corroborated these findings. Their research indicated that AI exposure is most pronounced in higher-paying professions. However, the report also noted a scarcity of concrete real-world evidence to suggest that widespread AI-driven displacement is currently occurring for these highly exposed workers.
The Anthropic report observed, “We find no systematic increase in unemployment for highly exposed workers since late 2022, though we find suggestive evidence that hiring of younger workers has slowed in exposed occupations.” This suggests that while AI might be influencing hiring trends and potentially the career trajectories of younger workers, a mass exodus from traditional jobs due to AI hasn’t yet materialised on a significant scale.
Understanding “AI Exposure”: Nuance is Key
It’s crucial to understand what “AI exposure” signifies in this context. As Mr. Karpathy indicated, the scoring was based on the digital nature of a job, a metric that LLMs can readily assess. This digital footprint can include tasks that are repetitive, data-intensive, or involve processing large amounts of information, all areas where AI excels.
High Exposure Occupations: These are typically roles that involve:
- Significant amounts of computer-based work.
- Analysis of large datasets.
- Writing and coding.
- Information synthesis and summarisation.
- Customer interaction that can be standardised.
Low Exposure Occupations: These often involve:
- Physical dexterity and manual labour.
- Direct human interaction requiring empathy and complex social cues.
- Unpredictable environments and problem-solving in real-time.
- Skilled trades requiring hands-on experience and craftsmanship.
The controversy surrounding Mr. Karpathy’s list underscores the delicate balance between technological advancement and societal adaptation. While AI undoubtedly presents opportunities for increased efficiency and innovation, its potential to reshape the labour market necessitates careful consideration and open dialogue. The focus now shifts from merely identifying vulnerable jobs to understanding how these roles can evolve, how workers can be retrained, and how society can best navigate this transformative period. The insights, though initially controversial, serve as a valuable starting point for these important conversations.
