The Future of Intelligence
Year: 2025
Type: article
Abstract: The current trajectory of artificial intelligence, defined by the brute-force scaling of Transformer-based architectures, has brought remarkable capabilities but also fundamental limitations. This essay argues that achieving true artificial general intelligence requires a paradigm shift from statistical mimicry toward systems that genuinely understand the physical world. The paper identifies two critical barriers facing the current paradigm: The Embodiment Gap: The disconnect between digital prediction (next-token generation) and physical understanding, resulting in hallucinations and a lack of spatial intelligence. The Infrastructure Trap: The economic and energy unsustainability of current scaling laws, which contrast sharply with the 20-watt efficiency of biological intelligence. The essay advocates for a scientific re-alignment toward World Models (citing architectures like JEPA and Marble) and Spatial Intelligence, arguing that future systems must move from predicting pixels/tokens to predicting abstract physical concepts. It concludes with a framework for "Scientific Re-Alignment" and "Strategic Deceleration" to ensure safe and sustainable AGI development.
Author: Dourthe, Benjamin
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7119165081
FWCI:
Cites:
Cited by:
Related to: