Demis Hassabis has recently opined that we are nowhere near AGI.
Has dear colleague Gary Marcus been vindicated?
It would be fairer to state that LLM-based AI has hit a wall, if not deep learning itself. The methodology is simply inadequate to claim the throne. There might, however, be more powerful methodologies that may have indeed worked, such as that which is pursued by our groundbreaking neurosymbolic AGI architecture, which is a mathematically rigorous manner of AI unification.
That’s not just next-gen AI. Please think of it as next-next-next-gen. It is true AGI.
Coming soon to a machine near you.
My own stance on the matter is that deep learning has the potential to reach AGI, however, with appropriate extensions and architectural capabilities that will make it possible for an AI system to match and surpass human-level performance properly.
Ad astra per astera.
References:
- The Foundations of Deep Learning with a Path Towards General Intelligence E Özkural, International Conference on Artificial General Intelligence, 162-173 (see arXiv preprint here)
- Deep Learning: A Critical Appraisal Gary Marcus (arXiv preprint)
