In Open Skill Genome Project’s October webinar, Wharton Professor Daniel Rock and University of Pittsburgh Professor Morgan Frank discussed groundbreaking research on how Large Language Models (LLMs) are transforming the workplace. In this research, recently published in Science, Professor Rock and his team, comprising Tyna Eloundou (OpenAI), Sam Manning (Centre for the Governance of AI, Oxford, UK), and Pamela Mishkin (OpenAI), uncovered surprising findings that challenge our understanding of AI’s impact on jobs, particularly which workers face the most significant changes.
Unlike previous waves of automation that primarily affected lower-wage jobs, LLMs are poised to impact highly educated, well-paid knowledge workers the most. Through analyzing 20,000 workplace tasks, his team found that 14-15% of tasks could be immediately impacted by current LLM capabilities, with this number jumping to 47-56% when considering potential system integrations.
Who’s most exposed? Researchers, scientists, engineers, and surprisingly, mathematicians top the list. But before anyone panics about job security, Prof. Rock emphasizes an important point: exposure doesn't necessarily mean replacement. Instead, exposure here means these tasks show promise for potential transformation and augmentation by LLMs.
Prof. Rock classifies LLMs as a General Purpose Technology (GPT) – putting them in the same category as electricity, the printing press, and computers, which have the potential to significantly transform how humans work. However, he cautions that full integration could take 20-30 years, similar to previous technological revolutions. This is not an overnight transformation, but rather a gradual reshaping of our working environment.
Perhaps the most practical takeaway is the need for new ways to communicate skills between workers and employers. As traditional job skills become less relevant, we need more dynamic systems to match evolving workplace needs with worker capabilities. He offered an interesting perspective for students and workers wondering about adaptation strategies: success will increasingly be measured by originality rather than just accuracy. “You don't get an A for being right anymore. You get an A if you're right and original, where originality is defined as a difference from whatever the LLM output would be.”
Want to learn more about this research? Access this article in Science Magazine here.
This webinar was hosted by Open Skill Genome Project, a joint initiative between xLab at Case Western Reserve University, University of Pittsburgh, and MIT Digital Credential Consortium, funded by the Walmart Foundation. The conversation continues next week with a session on Personal Data Ownership in the Age of AI, featuring Prof. Alex ‘Sandy’ Pentland (MIT Media Lab and HAI Fellow at Stanford University, and one of the architects of the EU’s GDPR) and Dr. Thomas Hardjono (CTO of MIT Connection Science, and one of the architects of the MIT Kerberos authentication protocol), moderated by Prof. Youngjin Yoo (xLab at Case Western Reserve University).
