An OpenAI co-founder has ignited significant discussion after compiling and subsequently removing a comprehensive list detailing which occupations face the greatest vulnerability to artificial intelligence. Andrej Karpathy, instrumental in founding the ChatGPT creator in 2015, generated the list by employing AI to meticulously analyse data from every occupation listed in the US Bureau of Labor Statistics' Outlook Handbook.
Analysing 143 Million US Jobs
The extensive analysis covered a staggering 143 million jobs across the entire US economy. Each professional role was assigned a specific score ranging from 0 to 10 to determine its "AI exposure." A higher score indicated a greater likelihood that the job could either be fully replaced by artificial intelligence or would substantially incorporate AI technologies into its standard workload.
Key Findings from the Data
The data revealed a clear and striking correlation: jobs commanding higher salaries generally exhibited worse average AI exposure scores. Conversely, workers earning less than $35,000 annually demonstrated the lowest levels of exposure to AI disruption.
Roles identified with the highest exposure scores included software developers, data scientists, and financial analysts. These positions, often involving significant digital and analytical components, were deemed most susceptible. In stark contrast, occupations such as construction workers, barbers, and nursing assistants received very low scores, suggesting their tasks are far less likely to be impacted by AI in the near term.
Controversy and Subsequent Deletion
After publishing his detailed findings online over a weekend, Mr. Karpathy made the decision to remove the list from his personal website. He explained that the public had largely misinterpreted the nature of the threat AI poses to these professions.
"I thought the code and data might be helpful to others to explore the BLS dataset visually, or colour it in different ways or with different prompts or add their own visualisations," he wrote in a post on the social media platform X. "It's been wildly misinterpreted (which I should have anticipated even despite the readme documentation) so I took it down."
Clarifying the 'Exposure' Metric
Karpathy provided crucial clarification regarding the methodology. "The 'exposure' was scored by a large language model based on how digital the job is," he stated. "This has no direct bearing on what actually happens to these occupations, which has to do with demand elasticity and a lot more complex economic factors."
His visualisation and findings, which have since been restored online, align with conclusions from several earlier reports examining artificial intelligence's impact on the workforce. A notable 2023 study from OpenAI itself listed similar office-based jobs as having high AI exposure, while roles involving physical labour consistently showed the lowest exposure levels.
Broader Context and Recent Research
A more recent report from Anthropic, the creator of the Claude AI assistant and published earlier this month, corroborated the finding that AI exposure tends to be highest in better-paid professions. However, this report also highlighted a critical nuance: there is presently little concrete, real-world evidence to suggest that AI-driven displacement is actively occurring for these higher-paid workers.
"We find no systematic increase in unemployment for highly exposed workers since late 2022," the Anthropic report noted. "Though we do find suggestive evidence that hiring of younger workers has slowed somewhat in occupations identified as highly exposed to AI." This indicates that while the theoretical risk is high for certain jobs, the practical, large-scale replacement of human workers by AI has not yet materialised as a dominant trend in the employment market.
