New AI Model Tackles Complex ‘Kissing Problem’ Puzzle

Rapid summary

• Kissing Number Contribution: Google DeepMind’s AI model, AlphaEvolve, increased the lower bound on the kissing number in 11 dimensions from 592 to 593. This represents a novel mathematical revelation.
• Wider mathematical Achievements: AlphaEvolve tackled various open mathematics problems. It replicated optimal solutions in 75% of cases, found new optimum solutions in 20%, and performed worse than known solutions in 5%.
• Algorithmic Advances: The AI developed an improved matrix multiplication algorithm that generalizes the best-known method previously discovered by DeepMind’s earlier model, AlphaTensor.
• Practical Applications at Google:

– Optimized data center orchestration for a 1% efficiency improvement.
– Improved tensor processing unit design.
– Reduced Gemini training time by finding kernel improvements (1% reduction).

• How It Works: AlphaEvolve uses genetic algorithms alongside large language models (LLMs) to iteratively refine and optimize solutions expressed as code verified by other programs.
• Next Steps for Expansion:

– Plans to apply it broadly across natural sciences via academic collaboration and simulators for verification challenges.
– Considerations for integrating an AI co-scientist system to improve hypothesis generation.!DeepMind’s practical achievements with AlphaEvolve

!AlphaEvolve optimization components diagram

Indian Opinion analysis

AlphaEvolve exemplifies how cutting-edge artificial intelligence can cross boundaries between abstract scientific discovery and practical industry applications. Its contributions to mathematics-such as refining the kissing number or improving matrix algorithms-highlight AI’s growing role as a collaborator in basic research rather than merely automating human tasks.

For India, advancements like AlphaEvolve present opportunities as well as challenges. With India’s ambition toward self-reliance in advanced technologies like machine learning and quantum computing, integrating similar systems could significantly bolster research initiatives across science, engineering, or even agriculture optimization.However, such breakthroughs also demand ethical foresight about reliance on proprietary foreign technologies that may shape domestic problem-solving capabilities.

Looking ahead, fostering collaborations with firms like DeepMind or investing heavily within homegrown talent frameworks will likely ensure India remains competitive globally while leveraging these tools responsibly within its socio-economic context.

Read More