It Won’t Take 1 Million People for a City on Mars to be Self-Sustaining
The real number is likely far less
The key question for a Mars city, or an O’Neill Cylinder, the state of Hawaii, etc. is whether it can survive if the ships stop coming. I’d argue more important is whether the civilization can thrive and grow, not just preserve itself in some sad torpor.
Back around 2016, Elon threw out an order of magnitude estimate that it would take 1 million people on Mars to reach this threshold. This is presumably based on division of labor in a complex society, where keeping the lights on relies on many people with highly specialized skillsets. I can’t make most of the products I use daily, or provide the services (water, electricity) I make use of. Partly this comes down to knowledge: I don’t know how to make those goods, or how to make the machines that make them. It’s also about time: I suppose I could follow YouTube videos for some of these things, but this isn’t a good use of time when other folks already have the knowledge, and I can pay them to do and make things for me.
What’s changed dramatically since 2016 is the unexpected rate of AI progress:
When Elon gave his estimate, AI could only do things like recognizing handwriting and images at human levels. Today, LLMs like GPT4 (and soon 4.5 or 5) have high-level knowledge across basically every field. Broad and deep.
Many people focus on the economic impact of jobs now at risk of being automated away by AI. There are convincing counterarguments that this type of fear has never been realized in history because we always just find new jobs for people to do, but I suspect there may be a step function in the extent that post-AI jobs focus less on keeping the fabric of society together. More than ever, they’ll be in the fields of entertainment and indulgences.
If we sent a small group of humans to Mars with strong AI, could they become self-sufficient? Probably not, because even if they knew what they should do at each moment, their time would still be too limited. This is where the next emergent technology comes in: AI embedded in robotic systems. Just this week Figure showed off its AI robot that learned to make coffee not by having the routine programmed into it, but by watching humans perform the task.
Tesla is attempting the same thing with Optimus. Like the LLMs that precede them, these bots will be endlessly patient and will work around the clock with no complaints. They can work outdoors in space with no life support beyond a power supply.
AI robots can also surpass another hurdle: tacit knowledge. No LLM can tell you how to make an H100 GPU because that knowledge isn’t written down, but rather distributed in the working practice built up over decades at NVIDIA. But a robot with access to the lab could probably acquire that tacit knowledge by watching for long enough.
How far down can we drive the 1,000,000 figure with these capabilities? Surely to 100,000 and probably 10,000. But what about 1,000, or 100? Is there any reason for an individual space settlement to exceed Dunbar’s number, or the genetic Minimum Viable Population? Perhaps we can upgrade ourselves to get rid of these limitations, but I’m not sure there’s much benefit.
In any case, this bodes well for Mars settlement plans and space expansion in general. Logistics and governance for a given effort are far easier when you drop the number of people that need to breathe and eat by several orders of magnitude. Coincidently, below several hundred is around the cutoff where communal-type governance becomes viable, with real world evidence from kibbutzim. All you need is to gather some friends and robot helpers.