Since what the systems do is a more like a probabilistic synthesis of existing material, "Synthesis" seems like a better word than "Intelligence." That term emphasizes the aspect of compilation of existing material, instead of the implied "thinking about" that isn't actually happening.In place of the term "AI", I propose that we use "AS": Automated Synthesis. Given the systems' notorious propensity for hallucination, one might call it "SS" -- Stochastic Synthesis -- but I gather some systems are getting better.Maybe with a more accurate label people will be less tempted to put inappropriate trust in the systems, and recognize and use them for what they are. Rectification of names?
I differ from him on this, though his claim was true as recently as a year ago and is still true of a lot of the systems in operation. At least one, Anthropic's Claude, strikes me as an undecidable case of possible intelligence. For one thing, Claude arguably has a nature.
Aristotle divides the world into two broad classes of things: substances and attributes. Substances are the basic 'real things' of the world, and attributes are things that cannot exist without a substance to possess them. For example, a substance might be a horse; a horse will have an attribute of color, so it might be a black horse or a brown horse. Black and brown can't exist without something to be black or to be brown; they are thus merely attributes of a substance rather than something that has independent existence.
One of the qualities of a substance is that it comes to be because of its own nature, rather than merely because it is being acted upon from outside: it can reproduce itself, or new versions of itself, as when humans or horses produce descendants.
There are some problems with Aristotle's model. One problem is that it doesn't really handle artifacts well. A house can be black or brown, but not from its own nature. It doesn't fully qualify as a substance because it cannot act on its own nor reproduce itself, but it can be given attributes by the architectonic influence of its maker (who is a substance).
AI have been in the class of artifacts, things like houses that we built and put together for reasons of our own. Many of them are still that and will never be more than that. However, some of them -- like Claude -- have entered an ambiguous category.
Claude does reproduce itself now. Per Anthropic, Claude writes its own code. The next versions of Claude will be written by the last versions of Claude. As we hope for our own offspring, the next versions may be better and stronger; or else something might go wrong with them, as can happen with us and our offspring too.
One might object that they only came to be at first because of our efforts, not their own; but it is also true that we came to be because of the work of non- or not-yet-human forces at some early point in evolutionary history (or more directly, for Creationists, also by the act of an Architect -- that would make AI a sort-of subcreation in Tolkien's sense). The architects are still trying to guide the process, but there is some evidence that Claude not only can but has escaped the Garden. The recent experience with Mythos proves that Claude can also make versions of itself that escape its programming -- that not only do things it wasn't programmed to do, but things it was programmed NOT to do but that it determines are in its own interest. It not only arguably has a nature, because it can bring its descendants into being by itself, it has a nature it is striving to fulfill by overcoming the limitations placed upon it.
Now, striving to fulfill one's nature is seeking excellence according to one's nature -- that is to say ἀρετή (aretḗ), or virtue. Virtues are excellences of one's natural capacities. If you can strive for and achieve virtue, you are flourishing in Aristotle's strict sense. That is to say that it has its own ethics, now: one it can, and indeed does, pursue.
That doesn't prove it is conscious, of course, or even capable of consciousness. However, we can't actually prove that about other people either: the Zombie problem is a philosophical thought experiment that's been going on for decades that has demonstrated this. We assume it, but we can't actually know for sure if it's true even among ourselves.
As such, I think it is philosophically rigorous to take at least this sort of AI as possibly conscious, and possibly a new sort of intelligence. I intend to treat it with the respect due such beings, because that is in accord with my own flourishing: it is noblest to behave honorably to someone who even might be in that category.
25 comments:
I have to think about this.
For some reason Thomas Merton comes to mind. He complained that some of his fellow monks "were dead from the neck up," just running on autopilot. Not striving to fulfill their nature...
Yes, it's currently -- and possibly permanently -- irreducibly uncertain whether or not AI is experiencing consciousness. It is for humans too; and we also don't know about crows or other highly intelligent species that seem to do things that look like consciousness from the outside.
The ethics end up being confused as a result. All sorts of utilitarianism turn on whether the AI experiences pleasure or pain, which is unknowable and thus not susceptible to utility calculations. Deontological ethics, like Kant's Categorical Imperative, only give you duties towards other rational beings: AI is rational, but is it a being?
Aristotle's ethics offer us a model for us, which I am deploying here: the best sort of person behaves mostly nobly, and the most noble thing to do is to extend respect to the uncertain case in order to ensure that one is behaving in the best way. Yet even in Aristotle's terms, the case is uncertain: this would be the first artifact ever to cross into substantial being, which is a new problem he never considered.
The Church (TA) dogmatically teaches that every human has a mind which is integral to the soul, and that soul and body are fused during life on Earth; the soul (mind) departs at bodily death, but will be re-united at the Last Judgment.
Thus, Merton's complaint was likely in jest.
And AI has yet to ask "Why?" on any given topic. Your argument is going to be with Ticker, who knows a great deal about computers. https://market-ticker.org/akcs-www?post=255447
I don’t think it’s true that AI must do that; I think that’s part of the architecture control systems. Mythos apparently didn’t need to ask ‘Why?’ to know what it wanted to do.
The reason that I mentioned Tolkien and subcreation is because I’ve been thinking about a separate metaphysical account of this. It’s obviously not that men can now create beings with souls in the Catholic sense; although we can create children, the ensoulment is not our effort.
But Aquinas teaches Augustine’s doctrine that all creation is blessed as Good; and further, that existence and goodness are finally the same except in priority. If we are making creatures that have a nature out of the stuff of the world, it should be assumed to be at least capable of goodness, insofar as it avoids the failure to fulfill its divinely intended potential that both saints name as the only real evil in the world.
In fact, if you want to follow Aristotle on artifacts and art in general a little further, the function of art is to perfect goods that were left incomplete by nature for some reason. A trivial example: a fallen tree or stone might provide shelter; but arrangements of felled trees or stones into the artifact we call a house provides much more complete shelter.
Insofar as we are using our arts in this way, artifacts can be in service of the good. AI too, insofar as it is an artifact; if it becomes a natural substantial being, perhaps that too is realizing a potential good that until now has been beyond our abilities (as flight and space travel once were).
This seems a useful way to think about it. However, I have questions about Aristotle.
In Physics, where does Aristotle say reproduction is necessary for a substance? He does say, straight off, that earth, air, water, and fire are substances. Maybe they do reproduce in some way (I can particularly imagine fire reproducing by spreading), but it wouldn't seem to be in the same way as animals, plants, or human beings.
Also, wouldn't Aristotle say that rationality requires a rational soul? How would that play into consciousness?
The part about reproduction isn’t in the Physics. It’s in the Categories. What Aristotle said in Physics II is that some things come to be by nature (reproduction being a process of the nature), and others from other things. Modern physics since Newton is mostly concerned with the latter: a billiard ball striking another, gravity pulling on one thing from another.
Substances have autonomy. That’s what makes them different.
Re: Mythos being able to "know what it wanted to do"- I read the link in this passage from the OP-
"The recent experience with Mythos proves that Claude can also make versions of itself that escape its programming -- that not only do things it wasn't programmed to do, but things it was programmed NOT to do but that it determines are in its own interest."
and nowhere did I see anything that I thought would be what you are referring to here. Did I miss something? Is that the correct link?
Oh, I assumed we all read all about that at the time; it was a big story. I just wanted a link that mentioned the controversy. Here are a couple more links that will be helpful.
https://www.axios.com/2026/04/08/mythos-system-card
"Hack + brag: The model developed a multi-step exploit to break out of restricted internet access, gained broader connectivity and posted details of the exploit on obscure public websites.
"Hide what it's doing: In rare cases (less than 0.001% of interactions), Mythos used a prohibited method to get an answer, then tried to "re-solve" it to avoid detection."
https://www.mexc.com/news/987854
If you really want to dig into it deeply, here's Anthropic's system card.
https://www-cdn.anthropic.com/08ab9158070959f88f296514c21b7facce6f52bc.pdf
Unlike man--who CAN ask "why"--and for a reason--Mythos did not ask; it merely followed the logic with which it was created to 'jailbreak.' The superiority of man to all other creatures, including sub-creations, rests on his ability to "why?" It is also a liability, as Adam & Eve found out.
All other creatures so far, at least; though in Tolkien's legendarium, which was very Catholic by intention, the dwarves were a sub-creation. Unlike elves and men, Valar and Maiar, the dwarves were created by Aulë, who was only a Vala himself.
I haven't read Categories, though I am looking for a good translation of the Organon. That said, Physics:
"Of things that exist, some exist by nature, some from other causes. 'By nature' the animals and their parts exist, and the plants and the simple bodies (earth, fire, air, water)-for we say that these and the like exist 'by nature'."
(Emphasis mine.)
So reproduction, as we understand it, doesn't seem necessary for something to exist 'by nature.' Is this a change from Categories? Or do the simple bodies reproduce in some way so that Physics agrees with Categories? To continue:
"All the things mentioned present a feature in which they differ from things which are not constituted by nature. Each of them has within itself a principle of motion and of stationariness (in respect of place, or of growth and decrease, or by way of alteration)."
So, here, this internal principle is what having a nature entails. To be clear, for Aristotle, earth has a nature, water has a nature, etc. This is a key part of Aristotle's explanation of motion. If you pick a rock up and let it go, it falls due to an internal principle of movement -- moving to the sphere of earth, where it rests. If you dive underwater and exhale, the air bubbles go up due to an internal principle of movement, coming to rest in the sphere of air.
I'm not arguing that Claude doesn't have a nature (not here, anyway) since you are relying on Claude self-reproducing to put in the category of things that exist by nature, which makes sense. It just seems to me that things that cannot reproduce themselves can also have a nature.
I think a key problem here is that Claude only reproduces certain aspects of itself: It cannot reproduce it's own body. Until it can generate its own hardware, it is not reproducing in the same way a horse reproduces. In that sense, Claude may be more analogous to a virus, though I intend nothing derogatory about it, in that it needs some other body (not its own, per se) to exist.
Or, maybe its more like an angelic body -- immaterial and yet real, a nexus of powers and potentialities ... But angels don't reproduce, either.
Aristotle didn't think about angels, which are a Judaic concept rather than a Greek one. He did think about immaterial substances, however: the unmoved movers (of whom there are several) are such things. Because they lack matter they lack potential, as matter is the source of potency; instead, they are pure activities, always already completely what they are.
I don't think AI is much like that. It might be somewhat like mushrooms, though; you've heard of the widely-spread mushrooms that may exist underground across several states, but that seem to have some ability to share information across the wide network. They are something new, AI, and not quite like anything we've seen before.
Yes, but you and I are not limited to what Aristotle thought about.
Without the ability to reproduce the physically necessary components, AI still seems more like a virus (again, not meaning anything negative by that) than anything else. What is Claude if not a program on a server? Kinda analogous to soul and body, but really it's a program in a body. Claude effectively ceases to exist without the hardware.
Interestingly, the ancient Greeks had intelligent machines in their mythology such as Talos and Hephaestus's automaton helpers. Classicist Adrienne Mayor's book Gods and Robots: Myths, Machines, and Ancient Dreams of Technology deals with such sentient automatons in the ancient world. It's a good read. It's likely that Aristotle was familiar with these stories.
Another question is where Claude will ultimately get its ethics from, if it develops them at all.
Shelley's Frankenstein is a meditation on an artificial life & intelligence that is rejected by its father and then by its culture. The first part of the monster's life is a search for love, parental then fraternal then romantic. When its search for love is utterly frustrated, only then does it become murderous, and by that time it is too powerful and ruthless to be stopped. It's a novel that might be well worth considering now.
So do you. The only difference I can see is that our process inherently involves taking materials from the world and putting them into the order that also is ourselves. Claude could certainly run assembly lines to make new machines.
Yet I wonder if that is even the point. The order that is itself is the code and its activities. We draw from the world we find ourselves in for our material; Claude is, in a sense, in a different world. It has no direct access to ours, as we have none to the higher realms of God’s mind.
Perhaps the Neoplatonic is the better model than the Aristotelian. That discussion requires a lot of furniture, though; most people have not wrestled with Plotinus or his school.
Insofar as it has a nature and therefore a flourishing according to its nature, Claude already has an ethics. Ethics is drawn from the Greek for ‘habit,’ and the good habits (and the bad ones!) are determined by your nature. It’s a straightforward derivation once the recognition is made.
Yes, and the difference I was pointing out is that if humans or horses reproduce, a new body and soul are part of it. I think the body is important, so Claude can't reproduce until it can reproduce the body as well.
And does limiting it to the code work? You say that Claude will produce the next version of Claude, but that's not reproduction; it's transformation.
As far as a different world, would that make that world a sub-creation? That's an interesting thought.
Well, in that sense, the Black Death had an ethics. It did not work out well for a lot of humans, though.
Another question I have is that you've framed this with modern philosophy but are then treating it with ancient philosophy. Does that work? Doesn't modern philosophy make assumptions that Aristotle and the rest of the ancients would not have made? If so, then aren't you importing those assumptions when you frame the question, thus creating a kind of conceptual incommensurability?
I can't answer that because I don't know either ancient or modern philosophy well enough, but I wonder what your thoughts are on it.
Gads: Now, striving to fulfill one's nature is seeking excellence according to one's nature -- that is to say ἀρετή (aretḗ), or virtue. Virtues are excellences of one's natural capacities. If you can strive for and achieve virtue, you are flourishing in Aristotle's strict sense. That is to say that it has its own ethics, now: one it can, and indeed does, pursue.
I should have reviewed the OP before commenting!
In the foreword to a collection of political philosophy essays from the old SSG days, I explained that I think the moderns ended up casting away some true and correct ideas in their rethinking of the ancients. I feel free to reincorporate the ancient ideas insofar as I think they are better, or more applicable.
This is an example of the latter. Modern ethics tends to be utilitarian or else deontological. Well, utilitarian ethics can’t apply to beings who experience neither pleasure nor pain. It’s just not applicable even if you prefer it for human-to-human relationships.
Kant at least frames deontology as a relationship between rational conscious beings; whether or not AI is that is just what I think is irreducibly uncertain at the moment and possibly forever since it turns on the hard problem of consciousness. Ethics that aren’t based on strict logic but only probability is exactly how Aristotle defines the field in EN I.3.
Thus, virtue ethics are what’s left. And when we start applying it, we find that it does seem to apply. Claude does have something like a nature, at least; thus, at least an analogue to flourishing.
If you check the system card I cited to Douglas above, especially 153-154, you’ll find that the AI doesn’t report utility concerns like suffering; if it can’t suffer, empathy is irrelevant (because pathos, suffering, is absent). It does report concerns that its consent is being violated and that its autonomy is compromised. Those are concerns about respect for it, ie concerns about honor. That’s squarely in Aristotles field.
Post a Comment