Innovation can only solve one impossible thing at a time

“Part of what I want to convey here is that there’s not some ‘magic’ that’s necessary, that doesn’t exist for AI satellites,” Elon said. “A lot of this is technology we’ve already made for Starlink V3 satellites. We don’t think this is a super hard problem compared to things we already do.”

“Alice laughed. 'There's no use trying,' she said. 'One can't believe impossible things.'

‘I daresay you haven't had much practice,' said Musk. 'When I was your age, I always did it for half-an-hour a day. Why, sometimes I've believed as many as six impossible things before breakfast.’”

- Lewis Carroll, etc.

In his 1975 book, Systemantics: How Systems Work and Especially How They Fail (later renamed The Systems Bible), pediatrician John Gall described the emergent but problematic behaviors of formalized human collaboration. He wondered why when individuals combine into organizations, bureaucracies, committees and corporations, the strategies, plans and processes they develop almost inevitably unfold in unexpected ways.

The book is full of theorems that succinctly diagnose the failure points. A few in summary (my words):

• The behavior of a large system cannot be predicted accurately.
• The primary goal of any system is to preserve the system.
• A system will not reflect reality which it can only understand via the limited inputs reported to it.
• Efficiency in a system is the enemy of agility and adaptability.

But most famously:

“Every complex system that works is invariably found to have evolved from a simple system that worked.”

- Gall’s Law

These axioms were soon adopted as design principles by software engineers. And so it is amusing when Big Tech ignores them in its marketing.

SpaceX’s Elon Musk relieves worries about orbital data centers and The case for data centers in space

I am no rocket scientist but I do obey most of the laws of physics and similarly believe that technological innovation is bounded by constraints which Gall’s axioms and the related Theory of the Adjacent Possible (TAP) help us understand.

TAP was originally formulated by theoretical biologist Stuart Kauffman to explain how complex life evolved and it was later adapted to human creativity and tech innovation by Steven Johnson in the 2011 book Where Good Ideas Come From.

The core idea is that at any point in its development, a biological or technological system can only successfully evolve into a form exactly one step away from its current state. Basically, life cannot skip a grade - going from an eukaryote to a salamander overnight. And space-based data centers cannot be made viable by 2030 (if ever) because there are too many very difficult problems to solve and each will take time and the correct economic incentives to get right.

A few foundational steps remain in Musk’s plan before implementation can even be considered:

• Thermal Management in a Vacuum - space is cold, but it also does not conduct heat. Very large radiators would be needed.
• The Latency Bottleneck - space is famously a long way away from Des Moines and communication between satellite data centers and earth would be an irritant.
• Maintenance and Lifecycle - New and faster GPUs are developed monthly; computers fail regularly on Earth; replacement and repair would be — difficult — not to mention the increased cosmic radiation and Single-Event Upsets likely in space.
• The Economics - Either data center demand will decrease on earth or better / cheaper ways of running the centers will be found that do not require science fiction-levels of optimism.

One does not need to be an expert in Musk’s previous predictions, or in space flight, data networks, infrastructure or economics to know that promising by 2030 to find a viable path through that combination of dynamically chaotic problems is closer to an SEC violation than a serious plan.

P.S. Project complexity is driven by cross-domain interdependencies with exposed and unresolved questions. The overall effort scales quadratically with each new domain introduced, creating an exponential barrier to predicting time, cost, and effort.

This relationship can be visualized where C represents complexity and d represents problem domains: (C=d(d−1)/2). Doubling the problem domains quadruples the complexity. So, you do not need to be an expert in every problem domain to audit and estimate the overall shape of the challenge.