An alien studying human culture wanted to understand the appeal of Dad Jokes but couldn’t figure it out.

Until one day it became apparent.

It's likely the first space aliens we make contact with will be ones of our own creation. Far from the bug eyed, green skinned humanoids of pop culture, first contact will be a diagnostics report from an orbiting satellite. Thinking machines of our own making are on their way to becoming the first alien intelligences. Never mind the hurdles of interstellar travel and communication over the vast gulf of space. The extraterrestrial artificial intelligences of the future will be truly alien, born and living outside Earth's atmosphere, and of our own design.

The current crop of AI tools like Large Language Models are narrowly tailored applications. They're good at text and image manipulation, but lack a human's ability to understand, learn, and apply their "knowledge" across a broad range of tasks and situations. You can't give ChatGPT a car and expect it to figure out how to drive. We can have self-driving cars, but the piloting software has to be built specifically for the task. The goal of an Artificial General Intelligence (AGI) is a piece of software that learns new things and more importantly, applies the lessons to new situations, just as a human can.

Beyond the nerd cred, why does anyone want to create AGI? There's a lot of money to be made, of course. I'm not against the profit motive, but let's not pretend that these tools aren't going to be given away for free. Whoever creates and controls artificial general intelligence will be able to lay claim to a slice of the entire world economy. They'll do this the same way that personal computers changed just about every industry when they became ubiquitous. When computers went from being the size of a refrigerator to the size of a deck of playing cards, all the world's economic activity became more productive. Computers took over dangerous jobs, sped up time consuming tasks, and generally made humans able to do their jobs faster. The goal of AGI is to do the same.

Being a computer, artificial general intelligence could work tirelessly in mundane or dangerous situations so long as it has enough power. A real AGI could be multiplicative across all industries, just like computers were multiplicative across science, engineering, and entertainment. Today we have new drugs, new airplane wing designs, and new forms of entertainment like video games that never could have existed without the ubiquity and low cost compute that society enjoys today. We stuff computers into everything, even things that shouldn't have them like toasters and light bulbs. AGI, if anyone manages to create it, will remove the need for human presence in a lot of activities. Products and services that used to take the combined efforts of dozens or hundreds of people will now get by with just a handful.

AGI is probably still a long way off and the current techniques we have for Artificial Intelligence may not be well suited to creating AGI. Generative AI's small context windows, lack of persistent memory and how they are trained on past data probably doesn't lend itself well to coming up with new ideas and responding to new situations. But people much smarter than me believe it's coming. They're spending a lot of money to pull their dream of the future forward into the reality of the present.

Chelan County is just a few hours east of Seattle, where I live. It is home to the town of Leavenworth, where the businesses dress up their storefronts to look like a Bavarian town. The charming facade hides that the county is home to some of the cheapest electricity in the country at around 3 to 4 cents per kilowatt hour. Seattle ranges from 8 to 16 cents depending on the time of day. 16 cents is the nationwide average. This makes it an attractive location for data centers, the place where we store the hardware to run AI.

Data centers are power hungry, and there's not enough electricity to power all the computers these companies want to run. Some of the data centers they want to build would consume as much power as New York City. Even if there was enough power, securing the permits to build data centers is getting more difficult as communities become more cautious about letting them into their backyards. Water usage, noise, light, and rising electricity bills contribute to their reticence. The power, permitting and environmental constraints aren't temporary problems. They're barriers that will only calcify as competition for these resources increases.

The final frontier is calling. Outer space is an attractive proposition for hyper scaling data centers and companies have already announced plans to build there. Solar energy is abundant in space. There's no pesky atmosphere to get between you and the photons you deeply desire. You'd have to have some really big solar panels, but we call it "space" for a reason. There's a lot of it out there. Beyond a couple smaller engineering concerns, the main barrier to putting whatever we want into orbit around Earth is just the cost it takes to get stuff up there.

In 1981 the first space shuttle took about $50,000 per kilogram to achieve orbit. Now, with reusable rocket stages and other innovations space hauling costs about $1,500 per kilogram. The reduction in cost continues to inch along and is getting closer to where it will make sense to consider construction in space, especially for the types of industries that have bigger terrestrial barriers, like data centers.

In space you don't have any neighbors. You don't have people concerned with property values or fighting over water rights and energy access. There's a lot of space out there. Satellite internet technology continues to get better as seen with SpaceX's StarLink and Amazon's Project Kuiper, so communicating with the things we put in space won't be difficult. Fulfilling a ChatGPT query is already slow, on the order of seconds. The way that ChatGPT and other LLMs act like they are typing out responses isn't just to make it look more human, its because it takes time to generate the next word to send back to the user. Where latency is already a part of the system, a few extra milliseconds because you’re talking to something in orbit isn't that big a deal.

The core of their benefit of AGI, the ability to learn and adapt to new situations, is sure to be affected by the vastness of their environment, their isolation, and their long operational times. The difference in their surroundings compared to ours is sure to affect their underlying software and they will drift towards increasing "otherness."

In 2001: A Space Odyssey, the computer, HAL, pleads for his life as he slowly shuts down. GPUs, the physical heart of all AI endeavors, last about 5-8 years. But it's only 3-5 years before they're considered obsolete and ready to be replaced by the next generation of processors. It's going to be hard to make those upgrades for a data center in space. Getting people into orbit to do maintenance will always be more costly than moving stuff. Space-based AGI will need to be more tolerant to failure and more autonomous to avoid maintenance delays. AGI in space will be subject to a constant cost/benefit analysis to decide if we should keep it running or not.

The true alien nature of AGI won't be that it looks different from us. It will be in how its priorities diverge from our own at a scale, speed, and efficiency powered by massive computational power. In the same way that computers sped up our entry into the modern, technological world, AGI will speed it's own path to becoming separate from humanity. AGI in space needs autonomy to operate. The more autonomy we give it, the more their human-like ability to learn and adapt to new knowledge and experiences will branch them away from humanity, reinforcing their alien aspect.

The LLMs we have now are certainly not sentient. I don't know that I'd ever believe that an Artificial General Intelligence could be considered a living being. I fall much more on the machine/tool taxonomy side of that proposition. In theory, even as GPUS fail and become obsolete there should still be a use for the data centers that house them. Maybe we'll have more modular designs that let us swap in blocks of cutting edge GPUs, keeping the power and cooling infrastructure of space data centers and simply augmenting their existing AI capabilities.

Decomissioning will be inevitable as we weigh the cost versus benefit of any project. Will we eventually have geriatric AI orbiting hundreds of miles over our heads? How long will we tend to our creations like a beehive tends to its queen? As these aliens we created age and the cost of repairing them outweighs their benefit, will they plead for their lives as HAL does, or will they throw themselves back through Earth's atmosphere towards land they will never reach as they disintegrate in a ball of fire? Perhaps will they become the first intelligences to leave Earth, breaking free of our orbit to go where no human has gone before?

For more on data centers in space check out the links below

This creature was the crowning and final work of the reclusive wizard Klaus Santhel. Seeking a protector that embodied vigilance, endurance, and ferocity, Klaus spliced forest spirits with verminous resilience, believing his intellect would be sufficient to command the result. He was wrong, of course. The reindearwig was quiescent for a time, but soon laid a clutch of eggs. Sensing Klaus, with his probing instruments and incantations, as a threat, it turned on its creator. His tower was found days later, the walls lacquered with corrosive secretions. Klaus himself was sealed in amber filth, preserved by the results of his own ambition.

As always, you can find more on the Reindearwig for free over on Patreon.

Some links on this site are affiliate links