Navigating the Future of Work with Machine Learning

As machine learning (ML) becomes more integrated into various sectors, new job roles are emerging at the intersection of human and ML systems. These roles include incanters, who specialize in optimizing inputs for large language models (LLMs) to achieve desired outputs. They understand the nuances of interacting with LLMs, such as the impact of input length and context contamination. Process engineers are crucial for maintaining quality control, especially in fields like law, where AI-generated errors can have significant consequences. They design workflows to catch and correct errors before documents are finalized.

Statistical engineers focus on measuring and managing the variability in ML outputs. They work to understand how different factors, such as the order of options presented to an LLM, can influence its decisions. This role requires deep domain knowledge and is akin to psychometrics, where complex human behaviors are modeled statistically. Model trainers face the challenge of training LLMs with high-quality data amidst a sea of misinformation. They might use uncontaminated pre-2023 data or employ experts to guide the training process and evaluate model outputs.

The concept of 'meat shields' refers to individuals who bear accountability for ML system failures. These roles are essential because, unlike LLMs, humans can be held legally and socially responsible. This accountability can be internal, like reviewing automated moderation decisions, or external, such as legal penalties for submitting AI-generated falsehoods. Haruspices are tasked with diagnosing why ML systems fail, whether it's a drone misfiring or a healthcare model's bias. They analyze inputs, outputs, and internal states to provide explanations for these failures.

These emerging roles highlight the complex interplay between technology and accountability, as well as the need for human oversight in an increasingly automated world. The article also touches on broader societal implications, questioning how different political systems might handle these technological advancements.