From Perplexity to Drones

The Artificial Intelligence (AI) field is so vast and diverse yet so interconnected that sometimes you start to inquire about something and you end up at a totally different place.

I have a long-standing interest in the AI-military connection, but that was not on my mind when I began to look into the Perplexity plagiarism case. If you are reading this substack, you are probably aware that this AI-powered (charitably labeled) news aggregator was caught repeating almost verbatim a Forbes scoop on Eric Schmidt’s secret drone project. And when Wired reported on that, calling it a “bullshit machine,” Perplexity duly plagiarized that too, inadvertently showing that, besides intelligence, LLMs also lack a sense of irony.

I am old enough to know who Eric Schmidt was long before he served as the CEO of Google. Remember Novell? Or DR DOS?

So, when I saw his name, I promptly clicked the link. It was an amazing story with roots in a controversial 2017 Pentagon project called Algorithmic Warfare Cross-Function Team, also known as Project Maven.

Project Maven

The project was “an effort to use machine learning (a form of AI) for analyzing massive datasets consisting of surveillance images taken by drones in the Middle East and other locations. (pdf)” It was awarded to six tech companies, including Google, the home of DeepMind and Google Brain, and two defense contractors. However, when many Google staff members signed a petition urging the senior management to drop the contract, Google withdrew from the project.

As with all such projects, this was not the end of Project Maven: it just moved on to other contractors, notably Peter Thiel’s Palantir. Last May, Palantir landed a $480 million extension to the Maven Smart Systems to be powered by LLMs.

What fueled the Pentagon’s interest in drones was the realization that the unchallenged American military air superiority was about to disappear. The doctrine that relied on expensive aircraft, precise weapon systems, and highly trained pilots was about to become outmoded in a world increasingly dominated by cheap and versatile drones.

War on the Rocks quotes defense analyst T.X. Hammes, who estimated that for the total cost of one F35A ($460 million), one could buy 16,000 Chinese Sunflower suicide drones ($30,000). “As Hammes concluded, “Which do you think creates more problems for air defense?”

While the American military-industrial complex pursued the development of expensive Unmanned Combat Aerial Vehicles like the MQ-9 Reaper, countries like China, Turkey, and Iran opted for cheaper and sometimes disposable drones. That created an ironic economic equation: when the US Navy was protecting ships against Houthi attacks in the Red Sea, they were  “shooting down a $50,000 one-way drone with a $3 million missile,” as Pentagon’s chief procurement officer put it recently. (Actually, Iran’s Shahed 101 or Shahed 131 drones cost about $20,000.)

Game of Drones

The game changer was the Ukraine war. It is estimated that Ukraine has been losing about 10,000 drones a month. Even the Shahed or Sunflower drones would be too expensive at that rate. As a result, over 200 local drone manufacturers popped up, producing cheap kamikaze drones. There is an incredible array of them, and they cost as little as $350. Small drones are put together with off-the-shelf Chinese parts, front-line soldiers add the payload, and off they go.

Home made drones

White Stork

After he left Google in 2016, Eric Schmidt became the first Chair of the Defense Innovation Board, established by the Pentagon. During his tenure, he became such an insider that, when he stepped down in 2020, he offered himself as the “prime liaison between Silicon Valley and the military-industrial complex” to companies that might need such an intermediary.

Through Project Maven, he was aware that the last century’s aircraft-centered doctrine was about to be replaced with drone warfare. It was evident that established companies like Lockheed Martin would not (and could not) drop their very expensive fighter jet projects to pivot into cheap and disposable drone production.

Schmidt founded the White Stork last year in Estonia with the goal of mass-producing $400 swarm attack drones. He poached talented engineers from Apple, Google and SpaceX to lead the project. He wants to create fleets of autonomous swarm drones that can handle precise saturation attacks.

What is intriguing is the AI behind the concept of the Third Age of Drone Warfare. Early drones were remotely operated by trained specialists. But it is not possible for humans to control hundreds of swarm drones. They need to be able to complete their mission by themselves. The question is, how do you find the data to train the AI that will turn them into formidable weapons? The obvious answer is Ukraine, as the country was labeled "the laboratory in the world of drones" by Schmidt. The new drones are tested in his California property in Hillspire and Ukraine.

These drones will be able to spot obstacles and alter their itinerary; they will communicate with each other and adapt to changing battle circumstances, and they will overwhelm the enemy’s air defenses.

Remember the 2017 Slaughterbots ALTER clip?

This is the ultimate aim.

Incidentally, Schmidt is not a lonely figure. According to a Brown University research paper, “From 2021 through 2023, venture capital firms reportedly pumped nearly $100 billion into defense tech startup companies — an amount 40 percent higher than the previous seven years combined.” This might be the harbinger of the emergence of the military-technological complex.

I will now post this and wait for Perplexity to plagiarize it.

Good times.