In 2019, Elon Musk met 4-star general O’Shaughnessy & Jay Raymond to discuss homeland defense innovation. O'Shaughnessy took their discussion to the United States Senate to pitch a new space-based "layered missile defense system" much like Brilliant Pebbles but powered by artificial intelligence to quickly and lethally act upon hypersonic and ballistic missile threats. He proposed the acronym SHIELD which stands for Strategic Homeland Integrated Ecosystem for Layered Defense.
This system would consist of a satellite constellation in orbit equipped with infrared sensors and eventually ICBM interception capability. The U.S. Space Force was established later that year and O’Shaughnessy joined SpaceX where he now leads their StarShield division.
SpaceX started deploying these special military variants of their satellites in 2023, launching them interspersed and connected to other Starlink satellites. The first StarSHIELD satellites host infrared sensors designed by L3Harris to detect and track missiles and perform fire-control functions.
SpaceX’s first StarSHIELD contracts were with the Space Development Agency and announced in 2020. The SDA was conceived and established by Under Secretary of Defense (R&E) Mike Griffin, who was previously the Deputy of Technology at Reagan's Strategic Defense Initiative. It is interesting to note that Griffin has an extensive history with Elon Musk during the early years of SpaceX . While these first tranches of SDA satellites are focused on communication, missile detection and tracking, Griffin and others have said that including space-based interceptor weapons in later layers will be "relatively easy" and he now works with SpaceX employees and primes on an interceptor with a company called Castelion in El Segundo. The interceptors are hypersonic glide vehicles (like FOBS) that re-enter from LEO and maintain contact with the satellites through phased array communication, the constellation above gives continued guidance to the interceptor to hit the ICBM or other target at launch above the enemy country.
No. For the easy 99.999% of driving they keep very little of the training data.
Basically you want to minimize manual interventions (aka disengagements). When the driver intervenes, they keep a few seconds before (30 seconds?) and after that intervention and add that to the training data.
So their training data is basically just the exceptional cases.
They need to just make sure they don’t overfit so that the learned model actually does have some “understanding” of why decisions are made and can generalize.
It's not clear that a bunch of cascaded rectified linear functions will every generalize to near 100%. The error floor is at a dangerous level regardless of training. AGI is needed to tackle the final 1%>
The universal approximation theorem disagrees. The question is how large the network should be and how much training data it needs. And for now it can only be tested experimentally.
The universal approximation theorem does not apply once you include any realistic training algorithms / stochastic gradient descent. There isn't a learnability guarantee.
You said it only depends on network size, I'm saying it more likely is impossible regardless of network size due to fundamental limits in training methods.
Didn't Elon have a Twitter meltdown today about how AI will leak your secrets?
The cited Science journal paper answers your question about whether the U.S. govt wants it. As well as the quotes from Heritage and Trump. Sounds like including interceptors (Brilliant Pebbles) is a bit political, but missile tracking is bipartisan (Brilliant Eyes).
