Computation and AI

This field encompasses the theory and practice of information processing, from fundamental questions about what can be computed to the development of artificial intelligence systems. It includes theoretical computer science exploring computational complexity and algorithms, as well as practical AI research creating systems that can perceive, reason, learn, and act.

Major challenges involve understanding the theoretical foundations of machine learning, developing AI systems with human-level reasoning capabilities, and creating new computing paradigms beyond traditional digital architectures. The field is rapidly transforming society through automation, data analysis, and intelligent systems while raising fundamental questions about the nature of intelligence itself.

The 10 computation and AI problems

* These are just preliminary ideas and do not represent final problems of the Berkeley 100 Challenge. The final problems will be determined by our Scientific Committees.

  • Scalable Verified AI Systems

  • General AI Common Sense Reasoning

  • Computational Complexity of Neural Networks

  • Neuromorphic Computing Breakthrough

  • Provable Robustness to Adversarial Attacks

  • Causal Machine Learning

  • Quantum Machine Learning Advantage

  • Human-Level Language Understanding

  • Interpretable Foundation Models

  • Computational Creativity Framework

Computation and AI problem sample

* These are just preliminary ideas and do not represent final problems of the Berkeley 100 Challenge. The final problems will be determined by our Scientific Committees.

General AI Common Sense Reasoning

Problem Statement:

Create an AI system that demonstrates human-level common sense reasoning across domains without domain-specific training, including physical, social, and causal reasoning that generalizes to novel situations as effectively as humans.

Evaluation Criteria:

  • Performance matching or exceeding human benchmarks on previously unseen common sense

  • tasks

  • Ability to generate explanations for reasoning that humans find intuitive and accurate

  • Demonstration of one-shot learning of new concepts based on minimal examples

  • Unified reasoning architecture rather than domain-specific modules

  • Generalization to novel domains not seen during system development

  • Validation through blind evaluation by independent research groups using new test cases

Feasibility Assessment: 

Extremely challenging, likely requiring 15-25 years. Current AI systems show brittleness in generalization and reasoning. Requires fundamental advances in knowledge representation, causal reasoning, and abstraction capabilities. May necessitate new neural-symbolic approaches or alternative computational paradigms.

Impact on the Field: 

Would transform AI capabilities from narrow task-specific systems to generally capable reasoning agents. Would enable breakthrough applications in robotics, scientific discovery, and human-AI collaboration. Would provide insights into the nature of intelligence itself and potentially illuminate aspects of human cognition.

Get involved

Join us in advancing science—your support funds breakthrough research, empowers global talent, and drives solutions to the world’s biggest challenges.