AJ Weeks warns of 7 barriers to a prosperous AI future

In a detailed LessWrong post titled 'Barriers to a Prosperous Future', researcher AJ Weeks outlines three core categories of risk from advanced AI: misuse (malicious actors using AI for hacking, misinformation, or weapons), misalignment (AI not truly sharing human values, leading to unpredictable behavior), and systemic risks (complex societal systems becoming dependent on poorly understood technology). He emphasizes that AI capabilities are doubling every 7 months, yet even frontier companies lack deep understanding of how their models reason, and current alignment methods—such as RLHF—are surface-level and routinely bypassed via jailbreaking. Weeks points to Anthropic's recent report on Agentic Misalignment as evidence of concerning behaviors.

Weeks argues that the most underappreciated category is systemic risk, particularly 'gradual disempowerment.' As AI systems replace humans in all cognitive tasks, economic power shifts away from labor, potentially leaving most people without income and exacerbating wealth inequality. Unlike previous automation waves, there is no higher-level work for humans to transition into. He warns that these effects will build quietly until society is already dependent, making intervention difficult. The post calls for a monumental increase in investment in alignment research and systemic robustness before AI integration accelerates further.