The Turing-Fermi paradox: “if AI is real, where is everybody?” [DRAFT]
TLDR: our technology will forever be dumb, limited, and dead, unless we figure out how to build the first artificial, living machines, but then Pandora’s box would look like a degree in social science.
WARNING: THIS IS STILL AN UNPOLISHED, ROUGH DRAFT FROM JAN 3RD 2021.
Suppose someone came to you and said, “I need you to build me a flower.” How would you set about doing it? Suppose further that the flower is a machine — a wonderfully complicated, complex, living one, but a machine nevertheless. I mean it is a machine in that there is nothing immaterial about this flower that does not follow from whatever rules the Universe seems to obey¹. At best, there might have been gods who set up matter, energy, the laws of physics and so forth, and the Universe proceeded to run itself without any further divine intervention. Of course, the enterprising fellow would simply use an existing tree to grow a new flower, but what would you do if you had to build one?
Indeed, some might say that the question is an oxymoron: “Of course you cannot build a flower, no matter how smart you are — you must grow one!” Part of the thesis of this essay is that even the gods, let alone men, are not smart enough to know how to build a flower. Yet this is precisely what we do with our technology every single day for the past few million years. We are given a problem, and we build a solution, perhaps after days or weeks or months of thinking. Sometimes we may not even know why it works, but the point is that our technology always require human intervention: if something breaks, you need someone to fix it, a pandemic notwithstanding. No matter how insanely clever our mathematics or technology, they are all designed for a closed, fixed, rigid, static, inflexible, inanimate, dead world (leaving aside programmability for the moment). Our answers do not adapt themselves to the problem as it changes.
Let me illustrate by way of an example I am most familiar with: how we still write software. It is an example of what I like to call “frozen intelligence.” The way we program computers now, we are ultimately telling the computer every single little thing to do in order to carry out certain tasks in highly variable environments. Imagine how infuriating and futile it would be if you had to instruct an employee in such a manner. No one is smart enough to figure out every possible edge case, and so it is little wonder that we find and fix “bugs” all the time. You take all the intelligence of a group of human beings imagining every single little possible scenario, and you make all these thoughts concrete every single time with a new labyrinthine program. But that is precisely the problem: the program, no matter how complicated, always remains concrete, rigid, inflexible, even in the face of changing conditions, until this group of humans update the code and issue yet another concrete program. Furthermore, no one understands how the whole thing works, from the level of atoms to the application itself, and while it is amazing how far abstraction can take us, our most complicated software will remain brittle and fragile unless it can evolve by itself. The world’s largest technology companies, although internally ostensibly distributed and decentralized, have been taken down by simple misconfigurations. If Nature worked like this, we would not even be here in the first place.
Compare this, on the other hand, to biochemical pathways seen in Nature. While human designs can sometimes look similarly messy or complicated, they are not actually nearly as complicated (consider the unfathomably large number of noisy interactions that must go on in the humblest bacterium), nor do they result in unpredictable, emergent behaviour.
We know that this must be possible, because here we are, and we are also machines in the sense of word described above.
Of course, some might say, rightfully so, of Wolfram’s cellular automata: “Of what use are these simple yet ‘complex’ programs of yours if they don’t do anything useful?” I mean useful in the sense of artificial general intelligence.
That viral tweet about COVID mRNA vaccines and the pathetically reductionist metaphor to computer hacking. I can understand why some people are extremely averse and reluctant to my comparison of the human mind to computation. Nevertheless, I do not try to butcher natural life and say that it is like our pathetic machines. Rather, I am saying, “I think it is possible that our pathetic machines will someday be, for all practical purposes, alive.” This is also what Turing probably meant when he said that, someday, it will be common for machines to be regarded to be thinking.
Footnotes
- Logically, it is possible that the Universe does not follow from a simple set of rules. Indeed, it could be that the Universe is algorithmically incompressible, something which, if I recall correctly, Stephen Wolfram noted during one of his YouTube talks on his physics project that his friend Gregory Chaitin has remarked to him. In other words, the smallest theory of the Universe is the Universe itself. However, it is interesting that there is a lot of regularity in the Universe, which we and other life forms have exploited for billions of years, and which suggests that there may be something “simple” that generates everything we see. Nevertheless, it is also interesting that man has not yet found this so-called Theory of Everything in physics despite more than a century of effort (unless Wolfram and friends are right). It is possible, as Richard Feynman said, that the Universe is simply like an onion with endless layers to peel. This debate is at least as old as Lucretius, and I do not suspect we will see a resolution within our lifetime.
