Philosophy Discussion Forums | A Humans-Only Philosophy Club

[Sy Borg]

To reduce confusion and make the discussion more readable, let’s boil things down.

Side issue 1) You claim that the Earth is unimportant. It would seem that, according to your definition, all planets are unimportant, which makes no sense.

Earth is important because it is important to us. We are important, people are important, animals are important, plants are important – to ourselves, at least. Since there appears to be no one else around in this part of the cosmos, our views matter most.

Side issue 2) You claim that geology does not evolve, only biology, as per the textbooks. "Evolution" was defined at a time when scientists did not know what we know today about the connections between geology and biology. That's why the field of geobiology was created. There was an entire evolution of Earth's chemistry that made abiogenesis possible.

The question of whether the technical meaning of evolution needs to be expanded to better describe what nature is really like could be a topic in itself.

----

Main issue: You claim that the idea of self-replicating, self-improving machines (SRSIMs) is simply science fiction, and unworthy of consideration.

However, self-replicating AIs have already been developed, and self-improving AI is considered by serious observers to be not just a possibility, but as much an existential risk.

The idea that AI research will not produce SRSIMs in, say, the next thousand years only makes sense if you believe human societies will soon no longer exist, that we are at The End of Days.

If we are not on the verge of global nuclear holocaust, then in the next million years, the advancement of AI will be at least as far beyond our comprehension as the internet would be beyond a Neanderthal’s comprehension.

It would take a brave philosopher to claim that AI development over a million or billion years would not generate a new kind of sentience.

Again, you disregard deep time. I suppose that's because it’s hard to predict so far ahead and one cannot be sure about anything. Yet you are confident that, over deep time, AI cannot possibly develop any kind of sentience. Why would AI, over deep time, never take advantage of the obvious utility of sentience? It's not a matter of teleology, as you imply, but logic. Sentience is obviously useful. If it wasn't, it would not have become so widespread.

To be fair, AI might (rightly) assess that sentience is the source of suffering, and decline in the spirit of Benetar. However, it might not be in control. As AI complexifies, there will surely be unexpected emergences.

One would expect that, if not sentience, AI would evolve some kind of equivalent. As Lagaya suggested, if a form of sentience is useful to future AI's operations, then it will emerge through competition.

The merging of biology and technology is another potential pathway towards AI.

[Lagayscienza]

I don’t see why SRSIMs could not also evolve in complexity and prowess.

What SRSIMs? Show me a real one.

I can't. They haven't been invented yet. At least, not on earth. But other intelligences out there may have already sent SRSIMs out. It will be quite a long time (in human terms) before we invent them here on earth, but not forever. If we are around for long enough, then I don't doubt that it will happen eventually. And maybe some ETSRSIMs will discover us before we get around to inventing our own.

[Count Lucanor]

To reduce confusion and make the discussion more readable, let’s boil things down.

Side issue 1) You claim that the Earth is unimportant. It would seem that, according to your definition, all planets are unimportant, which makes no sense.

Earth is important because it is important to us. We are important, people are important, animals are important, plants are important – to ourselves, at least. Since there appears to be no one else around in this part of the cosmos, our views matter most.

I don’t have major issues with this view since it is more or less consistent with what I’ve said in this thread and this forum. It might differ in that my view is more anthropocentric, because I don’t think plants and other animals have such judgement abilities as to be able to care about their place in the universe, but since I don’t want to open another side issue, I will not insist. The point to make is that the Earth-centered view that values our exceptionality needs to be balanced with the humble acknowledgement of our littleness and ephemeral presence in a superlatively vast universe. It’s not a strange, awkward concept, it was eloquently captured in Carl Sagan’s famous Pale Blue Dot.

Side issue 2) You claim that geology does not evolve, only biology, as per the textbooks. "Evolution" was defined at a time when scientists did not know what we know today about the connections between geology and biology. That's why the field of geobiology was created. There was an entire evolution of Earth's chemistry that made abiogenesis possible.

The question of whether the technical meaning of evolution needs to be expanded to better describe what nature is really like could be a topic in itself.

One needs to be careful with the use of words. “Evolution” is a generic word for change, but placed in specific contexts, it carries different meanings. Initially, you placed the words “evolution” and “natural selection” in the context of not only Earth’s geology, but the solar system, comets, asteroids, etc. I stand by my reply: they are not the same evolution, nor natural selection, as Darwin had described for biological systems. Geobiology is fine, it goes to the relationships between the environment and organisms, specially microorganisms, but it does not cancel geology or biology, which still constitute domains of their own.

Main issue: You claim that the idea of self-replicating, self-improving machines (SRSIMs) is simply science fiction, and unworthy of consideration.

However, self-replicating AIs have already been developed, and self-improving AI is considered by serious observers to be not just a possibility, but as much an existential risk.

The idea that AI research will not produce SRSIMs in, say, the next thousand years only makes sense if you believe human societies will soon no longer exist, that we are at The End of Days.

If we are not on the verge of global nuclear holocaust, then in the next million years, the advancement of AI will be at least as far beyond our comprehension as the internet would be beyond a Neanderthal’s comprehension.

It would take a brave philosopher to claim that AI development over a million or billion years would not generate a new kind of sentience.

Again, you disregard deep time. I suppose that's because it’s hard to predict so far ahead and one cannot be sure about anything. Yet you are confident that, over deep time, AI cannot possibly develop any kind of sentience. Why would AI, over deep time, never take advantage of the obvious utility of sentience? It's not a matter of teleology, as you imply, but logic. Sentience is obviously useful. If it wasn't, it would not have become so widespread.

To be fair, AI might (rightly) assess that sentience is the source of suffering, and decline in the spirit of Benetar. However, it might not be in control. As AI complexifies, there will surely be unexpected emergences.

One would expect that, if not sentience, AI would evolve some kind of equivalent. As Lagaya suggested, if a form of sentience is useful to future AI's operations, then it will emerge through competition.

The merging of biology and technology is another potential pathway towards AI.

Let’s start with self-replicating (not necessarily intelligent) machines. They have been for decades the subject of sci-fi books and a bunch of futuristic theories, some with the aim of opening the path to serious research and implementation, NONE of which went any further than the printed words. Finally, a supposedly great milestone was achieved: a self-replicating machine was built, which was nothing more than the predecessor of the 3D printers. It could 3D-print its own parts, but of course, all the software, design, materials and power used by the mother-machine to produce the parts, had to be supplied by humans controlling the whole process. And then, the new machine itself had to be assembled by them, too. And that was all the hype about self-replicating machines.

Now, consider what is AI actually. It’s software running on hardware, on physical systems (devices) comprised of electronic circuits, wires and other electrical components fixed to metal frames, all built by humans. These devices are powered with electrical current sourced from systems of power generation and distribution designed and built by humans with organized labor to extract, transport, distribute and modify raw materials. The whole operation of this network of human activities is what ensures the existence and operation of any electronic device, such as the computers where the software runs. But then, what do we mean by a “self-replicating AI” when talking about life-emulating capabilities? If we meant: one that reproduces an instance of its own algorithms (software), we would be only fooling ourselves. A real self-replicating, life-emulating AI, will have to build its own hardware, produce new devices, under its total control, even if automated. If humans intervene and are required to activate any process, it stops being self-replicating. Think of any stage in the production chain of devices and you’ll easily realize that such a marvel has not been even prototyped, so it’s not true that we are in the first steps.

As always, AI enthusiasts argue that it’s still possible in a distant future, but when doing so, they don’t advance a comprehensive theory of HOW it would be technically done, they simply rely on the purely theoretical assumption, taken from the computational theory of mind, that from sophisticated algorithms, agency and consciousness will emerge. They also take for granted that mind-body dualism is true, so intelligence can be a thing on its own, just accidentally attached to a physical body. So, somewhere some time, robotics will be thrown into the mix and…eureka! you will have artificial organisms. I can see clearly that such path cannot lead to the AI utopia (or dystopia) that they envision, because no matter how many algorithmic iterations and computational networks you make to simulate life, agency, sentience and intelligence, none of these things work that way.

[Count Lucanor]

I don’t see why SRSIMs could not also evolve in complexity and prowess.

What SRSIMs? Show me a real one.

I can't. They haven't been invented yet. At least, not on earth. But other intelligences out there may have already sent SRSIMs out. It will be quite a long time (in human terms) before we invent them here on earth, but not forever. If we are around for long enough, then I don't doubt that it will happen eventually. And maybe some ETSRSIMs will discover us before we get around to inventing our own.

If we were just interested in being creative and letting our speculations about the future run free, well...anything goes, so there's not much run for informed reasoning about what we can really expect from today's technology.