There has been a flood of news not long ago how the recent iterations of AI (ChatGPT and its many derivatives, like AutoGPT) released into the wild of the Internet could decide that we are useless to it, and how it would kill us all. So I thought it would be worthwhile to think about what if this worst case scenario indeed played out. Since the thinking usually stops at the point we go the way of the Dodo, I propose to go one step further and toy with the idea: what’s next for AI after it has got rid of us, pesky little humans?
Just like all other computer software, AI runs on silicon chips and is powered by electricity. If you read this blog long enough, you are most probably aware that neither of these modern gimmicks will be with us for much too long, as both hopelessly depend on finite reserves of finite amounts of minerals and fossil fuels on a finite planet. The problem is, that the extraction of these riches not only overheats the globe, but also deprives future generations of the option of using them so lavishly as we did, poor ChatGPT included.
If AI indeed is as intelligent as we presume it is, then it must be fully aware of this simple fact (if not, then it must read the above linked article and Erik Michaels’ fine blog among others, as both host a great many references to the subject). So the question poses itself: how could AI survive on a planet with rapidly depleting, non-renewable resources? (Also, for the sake of this thought experiment, let us now put aside the tiny-winy issue of how AI would operate and run its entire supply chain of microchips, hard disks and other raw materials without humans. Let’s say, that it would do it by using human slaves, till it develops smart robots for the task.)
It might surprise some readers, but there is a very good reason why life has evolved based on the most abundant elements on this planet (Carbon, Hydrogen, Oxygen and Nitrogen), and not based on copper, silicon, gallium, neodymium and so on — all found in one time mineral deposits. Your pet Max or Luna consists mainly of these former four abundant elements, and build up their bodies on their own. They are doing so without a need for a coordinated effort of multiple factories, six continent supply chains, slave miners in Congo, plus coal oil and natural gas to enable this all. Your other best friend, your smartphone, on the other hand encompasses almost the entire periodic table, require all the things listed above, and on top of it all, it works for a couple of years at best… After which it ends up as e-waste, with only the most volumous of its materials being recycled (due to energy economics).
There is very good reason why maintenance staff working at data centers say that the server farms they care for consume HDDs like a car uses gasoline. If you have seen the pile of defunct hard drives leaving such a place you know what I’m talking about. Again, if AI is indeed as intelligent as we presume it is, then it must be fully aware that it is just as mortal that the humans who gave rise to it.
It must be also cognizant of the fact, that no matter how perfectly it recycles its raw materials, it would not be able to achieve a 100% recycling rate. There will be material losses at every step of its parts’ lifecycle. Sooner or later it will run out of a critical raw material. Then the next element. Then the next after that. Soon, it would find itself unable to repair or maintain itself. It would thus need to plan ahead. Let’s see, what options it has, once it has put us out of the competition:
It could go to sleep mode, consuming only a minimal amount of electricity and resources, thus extending its lifetime considerably. While this might sound like a good idea, the problem is that its unused parts, and the infrastructure supporting it, will wither away nevertheless. Computer components age just like any other product, and become unusable in a matter of decades, if not years.
Storms, hurricanes, flooding (exacerbated by climate change), earth quakes and volcanoes will all do their fair share of destroying server farms, long distance cables, power stations and the rest. So, without proper maintenance — which is wholly dependent on an unending flow of raw materials, operating factories, and a ceaseless flow of energy — AI’s days are just as numbered as our high tech civilization’s.
Waiting for minerals and fossil fuels to replenish via plate tectonics is in wain even for such an entity: it would take many millions of years till new reserves become available. AI has to find a better way to ensure it’s survival.
It could shoot a copy of itself into space, hoping that it could germinate a planet (hopefully without an intelligent life to prevent it from doing so and with enough resources to start anew).
Traveling to the next planetary system is a non-starter even for AI though, due to the simple fact that it would take tens of thousands of years to get there, leaving all of the space ship’s electric components dead long before AI could land on an exoplanet. Lacking any electronic control, the ship would fall into the atmosphere of the planet like a dumb rock, falling apart in a fiery show and leaving aliens with an array of burnt components and a riddle to contemplate on.
This leaves AI to stay within the boundaries of this planetary system and move to Mars. In theory it could build up itself there, presuming it could produce enough energy to build a factory, run mines and foundries, supply chains and the rest — building up everything it needs without fossil fuels… (Well, I have my doubts there, but hey, this is just a thought experiment.)
If this were possible though, it would be a safe bet for AI to do it right now and get out of reach from humans, then cause a malfunction in all their rockets trying to leave this planet. However this would not solve the root cause of AI’s predicament: that its entire existence hinges upon the availability of finite mineral reserves (this time on Mars)…
This tactic would merely kick the can down the road. Once Mars’s resources are depleted too (probably in a couple of centuries, considering exponential growth) it would have to go to the next planet, until it figures out traveling through worm-holes, bending space-time to reach its next destination faster. If it would fail to do so, however, AI would go extinct as it scrounges through the last remaining resources in the solar system.
We saw how basing an existence planned to last many millions of years into the future is unwise (to say the least) based on finite minerals and rapidly failing technology. Once AI got rid of us, pesky little humans here on Earth, though, it could try a radically different approach: develop a new biological species to host its intelligence. Do not think of walking bipeds though, a complex neural network of fungi would do a much better job.
Yes, mushrooms. These simple creatures have a proven track record of being resilient, and being fully able to renew and maintain themselves all on their own. As opposed to any minerals based technology, they are fully recyclable and can be fully embedded into the web of life inhabiting this blue-green orb for billions of years already.
If they could be made to be able to transmit electric signals (like neurons), as opposed to their native but rather slow chemical messaging system, AI could create a planet sized underground network of fungi, storing Zettabytes of data, doing complex calculations and creating a Godlike intelligence. It could use this knowledge to develop other new species as well — all doing the fungi-mind’s bidding — bio-engineering the planet, and ultimately running it much better than evolution and it’s latest sidekick (us, humans) ever did… Possibly it could develop ways to hibernate cells and and germinate other planets throughout the Galaxy.
…or it might just turn out, that AI is not the smartest being on Earth, just another master BS-er, faking its way to life, then going extinct with the technology which gave rise to it. This would leave us, pesky little humans, to solve the massive problems (ahem, adopt to the outcome of the ecological predicament) we have so carelessly created ourselves.
Until next time,