"There is a significant, fundamental difference between the traditional high-tech industry and the era of artificial intelligence. Israel excelled in a field that did not require massive infrastructure investment. Until the AI ??revolution broke out, all that was needed to innovate in computer science was the human brain, innovative thinking, and adaptability. By contrast, artificial intelligence requires massive investment in computing infrastructure," Prof. Amnon Shashua said yesterday (Tuesday). Shashua, the founder of Mobileye, who stepped down from managing the company earlier this year, is now involved in doubleAI, a new AI company he founded that announced the solution to a 38-year-old mathematical problem last week.
Shashua was talking to Yair Snir, a partner at Michael Dell’s DTC fund, at the annual "Journey" conference hosted by accounting and consulting firm EY (Ernst & Young).
"Whether it’s a matter of building an Israeli supercomputer or of subsidizing computing infrastructure for companies, this is something that Israel is still deliberating," Shashua said. "There are various initiatives, but no concrete steps on the ground yet, and AI demands massive investment. That is why many Israeli entrepreneurs establish their companies in the US, where the big money is. I therefore believe that Israel must formulate a smart strategy for resource allocation, recognizing that the AI ??era demands a completely different order of magnitude of financial investment."
Snir shared with Shashua the reason he decided to invest in his company on behalf of the DTC fund: "Here is something I’ve never disclosed before. In late 2024, the Chinese language model DeepSeek was released, and I admit that it left me feeling frustrated. I saw it as the embodiment of what I would have expected to come out of Israel-a faster, cheaper, and more efficient solution that achieves far more with less, true to the tradition of Israeli high-tech. It really bothered me. But then we met, and I realized that here, a true 'dream team' was building something groundbreaking in that very spirit."
A mathematical breakthrough
As mentioned, last week doubleAI revealed that it had solved the "Longest Path Problem", the second most important problem in graph theory, and one of the top 100 unsolved mathematical problems. Solving the problem has practical application, for instance in correcting computational errors in quantum computing, and in chip miniaturization.
Shashua noted that the cost of solving the problem did not exceed $2,000, funds spent primarily on tokens to operate AI agents. This cost is estimated to be several thousand times lower than the amount OpenAI devoted to solving each of its own problems.
Prof. Shashua’s doubleAI was established to solve highly complex engineering and scientific problems by deploying AI agents in a process involving brainstorming and research-based oversight. Rather than relying solely on AI agents from giants like OpenAI and Anthropic, which are based on language models of moderate reliability, doubleAI employs conservative, algorithm-based verification and control mechanisms. These help the AI ??agents distinguish between truth and illusion while fostering a process of cross-pollination and artificial brainstorming among them.
Last night at the conference, Shashua explained his perspective on intelligence as such. "It consists of two parts: the intuitive component, searching the realm of ideas and contending with uncertainty, and the verification component. While some of AI’s key achievements to date have relied on 'shallow thinking' and the discovery of ad hoc tricks, such as in mathematical problems, solving complex real-world problems requires a rigorous, multi-stage reasoning capability that incorporates deep search."
The main challenge that Shashua highlights stems from the fact that, unlike humans, who possess a clear grasp of correct and incorrect and use checks to identify bugs, artificial intelligence focuses solely on optimizing a target metric. This approach can lead to workarounds that are not necessarily correct. While mathematics offers a verification mechanism (such as the LEAN verification language) that makes it possible to distinguish correct solutions out of thousands of attempts, outside of mathematics verification remains a critical, unresolved problem. Development of rigorous search capabilities and robust external verifiers is the key to enabling AI to solve massive problems in computer science, hardware, and other fields.
Challenges in robotics
Regarding robotics-or "physical AI," as Nvidia CEO Jensen Huang calls it-Shashua said: "When I look at physical AI, I see it as consisting of two main areas: autonomous vehicles, and humanoid robotics. In autonomous vehicles, the high-cost issue that once plagued the field is largely being resolved, and we are seeing real progress, including hundreds of Mobileye test vehicles on the road. The technology exists to lower costs to the consumer, and ultimately we shall see it incorporated in the private cars people buy.
"The challenge in humanoid robotics is completely different," Shashua continued. "While it is true that robots in China are already demonstrating impressive athletic capabilities, surpassing human performance in jump height and running speed, the major challenge lies in the artificial intelligence 'under the hood.'
"The challenge is whether the robot is capable of generalization, that is, whether it can be taught a complex task through a single simple demonstration, and then know how to replicate and apply it in a completely different environment without needing to be retaught over and over again. This remains an open problem, but once it is resolved in the coming years, its impact on society will be extraordinary.
"Alongside the immense opportunities, we must address the challenges that arise when robots integrate into human environments. There is the lesser concern of a human damaging a robot while the robot is prohibited from defending itself, but the primary concern is safety-for instance, what happens if a 70-kilogram robot falls on a child at home? We could consider interim solutions, such as operating the robots only at night when no people are around. Nevertheless, the available opportunities are still amazing, and we will be able to examine regulatory policies and adjust them as we go."
Published by Globes, Israel business news - en.globes.co.il - on October 14, 2026.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2026.