AGI in 5 Years? Ben Goertzel on Superintelligence

A Machine Learning Street Talk interview that opens with Goertzel’s disagreement with Ray Kurzweil: both say human-level AGI around 2029, but Kurzweil’s 2045 singularity is, in Goertzel’s words, ‘overfit to having humans do the invention’ — once the AGI invents, the exponent changes, and superintelligence follows within a few years. He argues language models’ emergent reasoning is real but limited, makes the case for neuro-symbolic hybrids, and describes a design principle from his OpenCog Hyperon system: an intelligence must spend energy on ambient, goal-free processing, because a mind that only pursues its goals overfits to how those goals were phrased. The conversation closes on regulation, which he fears will be both overreaching and too slow.