How to Build the Future: Demis Hassabis

Artificial General Intelligence (AGI) requires moving beyond current large-scale pre-training to incorporate continual learning, long-term reasoning, and sophisticated memory systems. While existing architectures provide a strong foundation, achieving AGI by 2030 necessitates developing autonomous agents capable of active problem-solving and goal-oriented planning. AI acts as a transformative "root node" for scientific discovery, as demonstrated by AlphaFold’s success in protein structure prediction, which now accelerates drug discovery globally. Future breakthroughs will likely emerge from interdisciplinary deep-tech teams that combine AI with domain-specific expertise to tackle massive combinatorial search spaces. Developers should build systems that anticipate the integration of AGI, focusing on high-impact, defensible applications—such as material science and cellular modeling—rather than simple API wrappers. Ultimately, the most significant advancements will come from systems that can move beyond pattern matching to genuine scientific hypothesis generation.