The former DARPA program leader and ex-Airbus CTO discusses how AI could design technologies humans have never built before.
Few engineers have moved as fluidly between defense research, Silicon Valley experimentation, and the upper ranks of the aerospace industry as Paul Eremenko ., the Ukrainian-born engineer has led ambitious technology programs at the Defense Advanced Research Projects Agency , helped launch experimental hardware initiatives at Google , and later became chief technology officer at Airbus and United Technologies.
Today, as co-founder of P‑1 AI, he is trying to solve a different challenge: whether artificial intelligence can automate the core cognitive work of engineering itself., Eremenko reflects on his path from Soviet-era Ukraine to Silicon Valley, the bottlenecks slowing aerospace progress, and why he believes AI may eventually design starships and Dyson-scale megastructures.. I have been drawn to aviation and space since childhood because I wanted to build things at the outer limits of our capabilities as a species.It was still the Soviet Union at the time. The contrast between the Soviet Union and the US in 1990 made me a lifelong America fanboy. I have a very visceral appreciation for freedom from government control, celebration of individualism, the rule of law, and the meritocracy of the free market.and visiting the cockpit. I started flight training at 14 and soloed in a plane before I soloed in a car. It’s just always been a passion.Studying aerospace engineering was just totally natural. I added a law degree later because I was interested in the US Constitution, national security law, and government contracts. Plus, I like reading history, and much of law school is about reading history.At DARPA, you led programs involving drones, robotics, spacecraft, and adaptive vehicle manufacturing. How did you first get involved with DARPA, and what did working there teach you?because I had a burning idea that could transform how we build the next fighter jet and other complex defense systems. Could we design and build an airplane 10x-100x faster by doing it the way we design and build microchips? Basically, designing with multiphysics simulation models and building programmable, rapidly-configurable factories.If we want to build an interstellar starship and it takes a century, how do we organize to make it happen? The most remarkable modern innovations have been done on a tight timeline: the Manhattan Project – 4 years, Apollo moon landing – 8 years, SR71 – 2 years, SpaceX Falcon 1 – 6 years. This was not always the case. Cathedrals could easily take 100 years or more to build, consuming non-trivial fractions of GDP. Reaching another star might take a concerted effort of that magnitude.Google taught me the ways of Silicon Valley. This was Google back in the day, and it really pioneered many concepts that Silicon Valley takes for granted today, like OKRs for alignment, giving researchers breathing room to pursue crazy ideas, continuous product experimentation, and more.To build a hardware equivalent to the Android app ecosystem. Developers would be able to create and sell hardware modules that could be integrated into a phone, essentially assembling a custom mobile device from modular components that could be fit into a family of frames.As the founding CEO of Airbus’ Silicon Valley innovation center, what was your mission? And how did it differ from working inside Airbus? The mission was to help an industrial behemoth like Airbus experiment at the pace of the Silicon Valley ecosystem and quickly adopt new technologies from outside the traditional aerospace supply chain. Basically, it turned into a mini-DARPA operating in When I moved to Airbus headquarters in Toulouse as CTO, my mandate was to transform Airbus research and engineering into a more nimble, faster-paced organization. We built a Skunk Works-like flight demonstrators organization and flew some great projects in autonomy and electric propulsion. My signature initiative, however, was called Digital Design, Manufacturing, and Services. DDMS was essentially the industrial-scale deployment of what we pioneered at DARPA-model-based engineering as a first-class design paradigm.Hydrogen is the ideal aviation fuel. It has the highest specific energy of any non-nuclear fuel, and aviation is the most weight-sensitive application. So it’s been a dream since the 1950s. What was new in 2020, as we were deciding to found Universal Hydrogen, was that hydrogen costs were projected to plummet, making hydrogen cheaper than jet fuel. The cleanest way to make hydrogen is by electrolysis of water using renewable electricity. And the theory was that as global renewable generation capacity was growing rapidly, off-peak renewable electricity which would otherwise be wasted would instead be used to produce cheap green hydrogen… and then the AI revolution happened. And data centers started to eat up all of the world’s excess electrical generation capacity. So the cost of hydrogen never fell. But we did fly the world’s largest fuel cell-powered airplane – a regional airliner – in record time and established a certification pathway for hydrogen flight with the FAA.The biggest impact of AI will be on the physical world around us. And yet, very few people are working on AI for designing and building things. I asked why, and the answer was that there was no training data. The modern AI paradigm is deep learning – learning patterns from large amounts of data. So if you want an AI aerospace engineer, for instance, you need to show it a million airplane designs, and it will learn the underlying physics and engineering principles. The problem is that there aren’t a million airplane designs. Since the Wright Brothers, there’ve been only a few thousand designs, and most aren’t available or accessible for training. So I thought: if you need a million airplane designs, why not use the model-based engineering techniques we pioneered at DARPA and that my co-founders and I have spent the last 15 years deploying across various industrial settings to create a synthetic data set for Although hypothetical designs, they need to be good ones: interesting, realistic, and realizable. And so that was the founding idea behind P-1 AI.for physical products. For now, it’s a mechanical engineer focusing, somewhat ironically, on data center cooling and critical power systems. Eventually, they will be an aerospace engineer. Our focus is on automating the cognitive tasks a human engineer performs. This, I think, is critical to gaining rapid widespread acceptance. We are shipping a product to large industrial companies in a form factor they are very familiar with: a junior engineer who joins a team, initially does all the mundane, repetitive work no one else wants to do, and learns and improves over time. We are selling work, not software. Our customers don’t have to change a thing about how they run their businesses, because change management in large industrial firms is difficult and slow. So what does a human engineer do? We do two things: we have a quantitative intuition over the product design space, and we know when and how to use engineering tools. The first is the ability to do very broad, very versatile multiphysics reasoning, but at relatively low fidelity. For instance, I can design an airplane on a yellow pad. It will be roughly correct, and I know where the issues will pop up at the intersection of aerodynamics, structures, thermal, vibrations, electromagnetic interference, etc. And I know which tool to use and how to do deeper, higher-fidelity analysis when and where needed. That’s the focus: quantitative design intuition and complex engineering tool use.In the near and medium term, it will make engineering organizations more efficient and expand their bandwidth. In the medium to long term, AI can help us design and build things that we don’t know how to make today. Ultimately, I want to build the AI that gets us starships andThe biggest challenge is engineering throughput. We don’t lack ideas. We lack the engineering bandwidth to design, test, certify, and manufacture at speed.My 17- and 18-year-old self was quite interested in AI: training simple neural networks, genetic algorithms, etc. By 20, I was pretty disenchanted and shifted my focus to other topics, like deterministic autonomy. Obviously, with the benefit of hindsight, I’d love to tell my 20-year-old self to stick with it!Georgina Jedikovska Based in Skopje, North Macedonia. Her work has appeared in Daily Mail, Mirror, Daily Star, Yahoo, NationalWorld, Newsweek, Press Gazette and others. She covers stories on batteries, wind energy, sustainable shipping and new discoveries. When she's not chasing the next big science story, she's traveling, exploring new cultures, or enjoying good food with even better wine.AI and RoboticsBeyond EarthInterviews
Aerospace Engineering AI Airbus Aviation DARPA Defense &Amp Military Google Innovation Interview Paul Eremenko Sustainability
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Paul Rudd & Nick Jonas Movie Power Ballad Gets Hilarious TrailerThe Power Ballad trailer has dropped, teasing what to expect from Paul Rudd and Nick Jonas' upcoming musical comedy film.
Read more »
Rob Gronkowski fires back at Logan Paul's callout: 'I'll whoop his ass'Another celebrity boxing match that no one is asking for could be in the works.
Read more »
Paul Rudd, Nick Jonas Get Violent as Songwriting Goes Wrong in ‘Power Ballad’ TrailerLionsgate releases John Carney's comedy feature later this year after it screens at SXSW.
Read more »
Trump predicts Jake Paul will run for office in ‘not too distant future’President Donald Trump gave podcaster Jake Paul his approval to run for office while visiting Kentucky despite no show of considering a run.
Read more »
YouTuber Jake Paul joins Trump in Kentucky midterm rallyYoutuber Jake Paul also joined Trump onstage as the president visited Rep. Thomas Massie's Kentucky district hoping to see him defeated in an upcoming GOP primary.
Read more »
President Donald Trump Endorses Boxer Jake Paul for OfficePresident Donald Trump endorsed boxer and influencer Jake Paul ‘for political office’ during a speech in Hebron, Kentucky, on Wednesday, March 11
Read more »