← Back

Computation

Our exploration of the computational universe is still in its infancy, constrained by traditional von Neumann architectures and energy-intensive processing. Emerging approaches—ranging from low-energy and neuromorphic systems to reversible and thermodynamic computing—promise to radically improve efficiency. At the same time, advanced AI safety, interpretability, and robust software synthesis are needed to ensure trustworthy and broadly capable computational systems.

R&D Gaps (7)

Silicon Compute is Massively Energy Intensive Compared to Biological Brains

Computation
Modern deep learning and general computation demand enormous energy, limiting scalability and sustainability. Addressing energy efficiency is critical for the next generation of computing platforms, though it also supports potential proliferation of advanced AI and should be advanced alongside AI sa...
Computation
49

Proving Math Theorems is Challenging for Both Humans and AI

Computation
Both human mathematicians and current AI systems struggle with proving complex math theorems. Enhancing theorem proving through interactive and automated methods could push the boundaries of mathematical reasoning.
Computation
50

AI Could Go Rogue

Computation
The potential for AI systems to behave unpredictably or dangerously (“go rogue”) is a critical concern. Ensuring safe and controllable AI architectures is essential for reliable operation. See also:  • https://www.lesswrong.com/posts/fAW6RXLKTLHC3WXkS/shallow-review-of-technical-ai-safety-2024 • h...
Computation
51

AI Could Be Misused

Computation
The risk of AI being misused—whether through malicious intent or unintended consequences—necessitates robust safeguards and countermeasures.
Computation
52

Our Compute Stack is Insecure but Fundamentally Doesn’t Have To Be

Computation
Insecure software can lead to vulnerabilities that undermine the reliability and safety of computational systems. Formal methods and rigorous verification are needed to synthesize secure software.
Computation
53

AI is Still Narrow in its Reasoning and Planning

Computation
Current AI systems exhibit narrow reasoning and planning capabilities compared to human cognition. Broadening AI training methods to include holistic, brain-inspired architectures and cognitive frameworks can advance general intelligence (flagging that there is an AI safety risk here).
Computation
54

Biological Life is Our Only Working Example of Complex Evolved Computation

Computation
Biological systems are the sole example we have of complex, evolved computation. Replicating this level of complexity in digital systems could unlock entirely new computational paradigms.
Computation
55