UK looks to nature to train AI at 0.1% of the cost

Key Points:

  • ARIA’s Scaling Compute program aims to find economic alternatives for training AI models.
  • The program seeks inspiration from nature and biology to improve computing infrastructure.
  • The ambition is to reduce the costs for training AI by a factor of 1,000 to enable globally accessible AI.


The UK’s Advanced Research and Invention Agency (ARIA) has launched the Scaling Compute program with a commitment of £42 million to address the high costs associated with training AI models on large datasets. This initiative aims to explore more cost-effective alternatives to the energy-intensive hardware currently used in the AI industry.


The program director, Suraj Bramhavar, believes that the current trend of increasing computing costs is reaching physical limits, impacting the broader societal and geopolitical landscape. Bramhavar highlights the need to explore innovative approaches inspired by nature, biological materials, and the human brain to revolutionize AI computing infrastructure.


ARIA’s ambition is to reduce the costs of training AI by a significant factor, potentially making globally accessible, safe, and transformative AI a reality. By exploring alternative hardware materials and biological substitutes for traditional silicone semiconductors, the program aims to enhance computing efficiency while minimizing economic and environmental impacts.


In an effort to democratize the benefits of AI technology, the program welcomes concept paper submissions from scientists, engineers, startups, and established companies across various fields. ARIA, funded by the Department for Science, Innovation, and Technology, operates with a total funding amount of £800 million and does not claim intellectual property rights or equity in the projects it supports.


Applicants for the Scaling Compute program can submit concept papers by March 27 and full proposals by May 7. The initiative seeks to unlock groundbreaking innovations in AI technology, empowering diverse stakeholders to participate in advancing the next frontier of computing power and efficiency.



Prompt Engineering Guides



©2024 The Horizon