Humans Are The Loop

Employment law offers a useful lens for cutting through the noise around artificial intelligence. Before anyone can form an opinion on what AI will do to jobs, recruiting, or society, it helps to get clear on what the technology actually is and isn't.

Several persistent myths get addressed directly:

• The claim that anything typed into a tool like ChatGPT automatically becomes public information
• The idea that generative AI tools are inherently and permanently biased because their creators are biased
• The assumption that all AI is generative AI, when predictive models (the kind behind Netflix recommendations, Amazon suggestions, and Spotify playlists) have been in everyday use for over a decade
• The phrase "human in the loop," which is treated as backwards — the goal is humans using the technology, not humans as an afterthought to an automated process

A review of AI-related legislation across states, Congress, and other countries turns up five recurring themes that regulators keep returning to, regardless of political alignment:

• Transparency — requirements to disclose when and how AI is used, including new WARN Act provisions asking employers whether AI adoption caused layoffs
• Pre-use vetting — testing tools before deployment for discrimination, privacy violations, hallucination, and other risks
• Training and monitoring — treating AI like a new car that needs regular alignment checks, not a "set it and forget it" tool
• Corporate policies — internal guidelines governing acceptable use, meant to prevent unnecessary job displacement
• Individual rights — the right to opt out of AI decisions, access the data used, and appeal for human review at key points in a process

These themes point toward a shared conclusion: the emerging legal framework around AI keeps people, not automation, at the center. The practical distinction between generative, predictive, and agentic AI matters for anyone trying to figure out where these tools genuinely reduce workload versus where they introduce new risk — and where a human still needs to make the final call.