Exclusive: What if AI could design a reaction engine, or even a star ship? Google Deepmind and Airbus veterans have just raised $ 23 million with an eye in this future

When dreaming the day that artificial intelligence gets a human capacity, previous Airbus CTO Paul Erenko says he has always done so in the context of the building of real world machines. “I want a Superintelligence of the AI ​​that I can build -Dyson’s Star Naves and Spheres,” he said Fortune—The last is a hypothetical megastructure of science fiction that would take advantage of the energy of a star.

While his dream is still far away, Elemenko lays the foundations. Has united strength with the first Google Deep Researcher Aleksa Gordic and Adam Nagel, leader in engineering in Acubed, Airbus’s Innovation Center. Together, they have founded P-1 AI, which has emerged from a sneaky today with a round of $ 23 million, directed by radical ventures. Other investors include Village Global, Schematic Ventures, and Lerer Hippeau, along with notable angels such as the Google Deepmind’s CAP scientist, Jeff Dean and the Vice President of New Welinder’s New Product Explorations.

P-1, called after P-1 adolescenceA 1977 science fiction novel by Thomas Joseph Ryan on a sensitive IA is developing an engineering assistant fueled by Ai called Archie. Similar to other AS attendees as the Devin who encodes the AI ​​of Cognition and, the idea is to incorporate Archie as a younger member of all engineering teams, to manage repetitive tasks but who are shaking how to interpret requirements, generate concepts of early design and check compliance with the regulations. It is an early step towards a much more ambitious vision: to use AI to design the complex machines of the future.

Erenko said he was surprised that no one was already working on this, but he quickly found out why. As with self-conduct cars and robots, teaching AI to build machines requires a large amount of training data. The key, he explained, is to simulate realistic engineering systems through the creation of virtual models of real world components, such as engines, pipes and axes. Then these physics -based simulations are combined in various configurations to generate data, which are used to form AI models that help automate engineering design.

According to Gordic, it is similar to how Google Deepmind used games to help coach Alphago, the AI ​​that overcome human champions in Go, a famous strategy board game. “Alphago initially trained to imitate real human players,” he said Fortune. Now, it will form and adjust the models of great language (LLMS) and other IA systems to understand and modify complex engineering designs in rich in physics systems such as the refrigeration of the data center or air conditioning systems.

In order to go beyond the “glorified autocomplete” capabilities of the LLM as Chatgpt, he explained, the models must be useful for engineering tasks. Alas, therefore, must understand the orders and follow the instructions. The powerful combination of AI models that are trained in synthetic data based on physics simulations and which can then understand and act on this data make automated engineering assistance a reality. “We train at Archie in synthetic data to make it a type of engineer at a university school,” he continued Erenko. But after the deployment, Archie can learn from human feedback and data from the real world of companies that use the AI.

The P-1 investors, Said Erenko, are interested in the best-term startup plans, but are especially excited about the future. “Many of us in the world of engineering and alas, we grew up in science fiction and science fiction promised us a super intelligence that will build star ships,” he said.

Great headlines such as Autodesk, Siemens and IBM Working on elements of using and engineering, but they are not creating a new class of AS attendees of generalist engineering, nor after the same great view of the machines built by AI.

However, Erenko and Gordic insist that his is a very realistic and focused path, and it is not purely a research project with an indefinite period of time. “We will not be a ten -year moon,” said Erenko. “This is a very pragmatic launch and a path to the market.”

This story originally presented to Fortune.com

Source link