Fireside Chat, Interview
Demis Hassabis: We're Three Quarters of the Way to AGI
- DeepMind anticipates the AI field will remain "exactly on track" to complete its 20-year mission timeline, originally predicted in 2010, with AGI projected for 2030.
- Applications of DeepMind's ideas to significant real-world problems are planned for when algorithms reach sufficient power and generality, a milestone Demis associates with the period roughly a decade ago following the AlphaGo victory.
- AI is expected to play a major role in material science, environmental, and energy sectors over the next few years.
- Isomorphic Labs intends to perform 99% of drug exploration work and time "in silico," reserving wet lab steps exclusively for validation.
- Current trajectories suggest drug discovery timelines could shrink from an average of 10 years to months, weeks, or potentially days within the next few years, revolutionizing the medical and drug discovery areas.
- The analysis and understanding of AI systems are predicted to evolve into a distinct engineering science.
- AI-driven simulations are envisioned to enable new sciences in economics and humanistic subjects by facilitating repeated controlled experiments currently impossible in the physical world.
- DeepMind has developed "WeatherNext," described as the world's most accurate weather simulator and significantly faster than those used by meteorologists, aiming to better understand weather through learned simulators before potentially extracting explicit equations.
- Future drug discovery breakthroughs are expected to enable personalized medicine and personalized variations of base medicines.
- Classical neural networks are predicted to model systems previously thought to require quantum systems, such as protein folding, suggesting many tasks currently believed to need quantum computing may be modulated on classical systems.
- The development path involves creating highly intelligent and precise tools, progressing through an "agent era" of increasing autonomy before addressing whether systems possess agency or consciousness.
- A gap in experiential consciousness is expected to remain uncloseable in AI systems due to a lack of substrate equivalence with biological brains.
- Post-AGI, methods may emerge to determine if systems possess experiential consciousness, though Demis notes this is currently out of scope.
- Upon achieving AGI, it will be used to answer questions posed in David Deutsch's The Fabric of Reality.