Are We Really on the Path to AGI?

The author, a former political scientist, explores the social and ethical challenges of early engagement with Artificial Intelligence by examining the assumptions within the field regarding the development of large language models and agentic systems. The central line of inquiry questions the pervasive assumption that AI progress necessarily leads toward Artificial General Intelligence (AGI). The author posits that while current AI demonstrates impressive fluency across many domains due to absorbing vast human knowledge, the mechanisms of progress—absorptive learning versus recursive learning—suggest different trajectories. Absorptive learning involved feeding models massive datasets of human expression, breaking down language barriers through unscripted computing, and allowing models to internalize implicit patterns. Recursive learning, involving testing outputs against external standards, is presented as the likely path forward for further progress, particularly in verifiable domains like coding and formal sciences. The article concludes that AI will advance unevenly across domains, with specialized progress in formalized areas rather than simultaneous generalization, which has significant implications for the white-collar world and the necessity of retaining human informal knowledge.

The narrative establishes a tension between the apparent breadth of current AI competence (fluency) and the structural limitations imposed by the mode of learning. The shift from absorptive learning, which allowed models to internalize the *substance* of human expression through statistical pattern recognition, to recursive learning frames the central challenge: how to insert necessary structure—scripting or verification signals—into the learning process without sacrificing emergent novelty. The argument about specialization versus generalization highlights a critical disconnect between empirical capability and cognitive breadth; AI excels at mastering the *medium* (language) but struggles with the underlying *substance* (implicit human knowledge). This implies that current progress is heavily biased toward formalized, verifiable domains like coding, where recursive methods are most effective, rather than true general intelligence. The implication for the white-collar sphere—where tacit, informal knowledge sustains expertise—is that replacing novices with systems risks dissolving the continuity of practical wisdom. The underlying pattern suggests a systemic resistance to acknowledging non-formalized knowledge as valuable, favoring highly measurable, quantifiable progress, which ultimately channels future development into specialized, formal regimes rather than true cognitive generalization.