Research paths from the earlier book

The earlier Mirror and the Machine outline provides a research agenda beyond the entries already adapted. The items below are questions for investigation, not established conclusions.

Control and human judgment

Research Wiener’s warnings about literal goals, Weizenbaum’s distinction between deciding and choosing, and the 1980 NORAD false alarms. Establish what each source actually says before connecting it to modern alignment. The draft’s claim that a faulty chip proves automation removes a protective human buffer needs case-specific evidence.

Labor, agency, and inequality

Follow Ghost Work into evidence about data labeling, moderation, pay, and working conditions. Agentic Inequality now provides a source-linked entry into differences in access to autonomous delegates. Investigate whether those differences change bargaining power or international safety cooperation. The draft’s “necro-exportation” framing needs a source and a clearly attributed argument.

Mind and governance

Use ELIZA Effect as a starting point for studying projection and the draft’s digital-mirror hypothesis. Explore Rawlsian fairness, Treaty-Following AI, and institutional accountability alongside Society-in-the-Loop. Keep consciousness claims separate from claims about observed behavior. Stable Alignment introduces one simulated-social-feedback method, without equating simulated agreement with public legitimacy.

Training and deception

Inoculation Prompting and Emergent Misalignment now link to the reward-hacking study and distinguish its controlled training setup from deployed behavior. Further work should compare replications and mitigation limits. The earlier outline’s assertion that permission to cheat reveals deception as a defense mechanism remains an interpretation beyond the reported evidence. Scalable Oversight and AI Safety via Debate offer a separate path into evaluating difficult reasoning.