Charting Multiple Courses to Artificial General Intelligence

Artificial intelligence (AI) and the potential emergence of artificial general intelligence (AGI) have important national security implications for the United States, particularly with regard to its competition with China. Existing AI technology—in the form of large language models (LLMs)—has shown great promise, and many in the AI technology and policy worlds argue that LLMs may scale up to AGI in the near future. This paper is intended to complicate that position, explaining why there are barriers to LLMs hyperscaling to AGI, and why AGI may instead emerge from a suite of complementary, if not alternative, algorithmic and computing technologies. The goal of this paper is to provide U.S. policymakers a clear, nontechnical introduction to the issue of LLM hyperscaling and alternative pathways to AGI. The authors argue that there may be multiple courses to AGI and thus recommend that policy around AI avoid over-optimizing for a given possible future (e.g., hyperscaling LLMs), even while that policy addresses the possible near-term emergence of AGI in the hyperscaling paradigm.