Amnon Shashua Discusses AI Infrastructure and doubleAI's Mathematical Breakthrough
Professor Amnon Shashua discussed AI infrastructure challenges, doubleAI's breakthrough solving a 38-year-old mathematical problem, and robotics at the EY Journey conference.
The artificial intelligence industry demands massive computing infrastructure investments unlike traditional high-tech, Professor Amnon Shashua, founder of Mobileye and doubleAI, said at the EY Journey conference. Shashua noted that while Israel excels in computer science innovation requiring only human capital, AI requires immense capital outlays. Discussing doubleAI's recent breakthrough in solving a 38-year-old graph theory problem for under $2,000, Shashua emphasized the importance of verification mechanisms alongside generative AI.
The Infrastructure Challenge and Computing Costs
Shashua pointed out that Israel is still deliberating whether to establish a national supercomputer or subsidize computing infrastructure. Without concrete steps, many Israeli entrepreneurs will continue establishing companies in the United States where major capital is concentrated. He stressed that Israel must formulate a smart resource allocation strategy keeping in mind the massive financial scale required for modern artificial intelligence development.
Solving a 38-Year-Old Mathematical Problem
doubleAI recently solved the 'longest path problem,' one of graph theory's top open problems, utilizing AI agents for brainstorming and conservative verification algorithms. The entire process cost less than $2,000 in token usage, achieving a fraction of OpenAI's typical expenses. Shashua explained that doubleAI's systems also rewrote an Nvidia CUDA software library, outperforming human-written code and increasing performance by 3.6 times on average.
Physical AI and Robotics Challenges
Addressing physical AI and humanoid robotics, Shashua noted that autonomous driving costs are decreasing, bringing consumer-ready autonomous vehicles closer to reality. However, humanoid robotics face generalization challenges—teaching a robot a task via a single demonstration and having it adapt to entirely new environments remains an open problem. Safety regulations and protocols will need careful adaptation as these robotic systems integrate into human environments.
