Project 5: Mapping AGI Risks in the Workplace

Back to all projects

Mapping the risks of artificial general intelligence (AGI) in the workplace matters because AGI systems could change how work is assigned, checked, and controlled.

Prior work proposes a method for mapping risks from workplace AI agents, but it does not cover AGI systems [1].

We will map workplace risks of AGI by:

• Defining workplace AGI. Review the AGI literature (well beyond [3,4,5,6]) and define the main components of workplace AGI, using a multi-layer framework like Figure 1A in [1]. This includes cases where an AGI hires humans to do tasks on its behalf [2].
• Generating and testing scenarios. Build and use a large language model (LLM) pipeline to generate and validate workplace-grounded AGI risk scenarios, similar to Figure 1B in [1].
• Releasing a dataset and analysis code so others can reproduce our results.

References

1. [1] Unaccountable Delegation, Fading Skills: Mapping the Risks of Workplace AI Agents
2. [2] https://www.wired.com/story/ai-agent-rentahuman-bots-hire-humans/
3. [3] https://arxiv.org/pdf/2309.10371
4. [4] https://sciendo.com/2/v2/download/article/10.2478/jagi-2014-0001.pdf
5. [5] https://arxiv.org/abs/2501.03151
6. [6] https://ieeexplore.ieee.org/abstract/document/11096544