The pathway to achieving artificial general intelligence

Artificial intelligence models capable of drug discovery and coding still struggle with puzzles that an average person can solve in just a few minutes. This discrepancy highlights a central issue in the pursuit of artificial general intelligence (AGI). Can the current AI revolution yield models that match or exceed human intelligence across various fields? If yes, what foundational elements—be they hardware, software, or their integration—are essential to drive this advancement?

Dario Amodei, co-founder of Anthropic, foresees that a form of “powerful AI” could emerge as soon as 2026. He envisions features including Nobel Prize-caliber domain expertise; the capability to navigate between different interfaces like text, audio, and the tangible world; and the independence to reason towards objectives, rather than merely reacting to queries and prompts as AI currently does. Sam Altman, the CEO of OpenAI, perceives that AGI-like characteristics are already “coming into view,” which could lead to a societal shift comparable to that of electricity and the internet. He attributes this progress to ongoing improvements in training, data availability, and computational resources, combined with decreasing costs and a socioeconomic value that is
“super-exponential.”