Addition on reaching the end of the post: Every once in a while, there’s a little beacon of sanity and hope.
AnthonyC
Karma: 3,851
Pulling out all the stops to mislead.
I notice that this fragment lacks a subject, and there is thus a trivial way to make it technically true. Far more true than it could be otherwise, really.
Note that Dean’s is also a fully general argument that humans cannot cause human extinction. That should be a sufficient reason to reject it.
I would think so.
Personal anecdote: There’s a Cerave moisturizer creme I buy that sometimes comes in a pack of two equal weight jars, one with a screw top amd one with a pump top. The pump only lets you get about half out. I use the other one and then switch the tops.
Great review, agreed on almost all counts. I haven’t read the book, but it sounds like its author is either missing or ignoring some basic background facts and historical context (which, admittedly, is pretty normal).
I think the value of Socrates made a lot more sense to me after a few things I learned later.
1) I read Three Worlds Collide and met the Xenopsychologist. If Socrates was (one of) the first, he was also the least, the worst, the weakest. We can know things, and he knew that, but the people of his time thought we could and did know things in ways we couldn’t and didn’t and don’t, and that needed to be corrected before we made progress. You’d be an idiot to try to learn physics directly from Newton or astronomy from Ptolemy, too.
2) We mostly know Socrates through Plato, writing after Socrates’ death, as you say. Plato definitely* thought we could know things, without even looking at them in reality. And the further in time he got from Socrates’ death, so did his Socrates. I expect we’d have a different impression is the Dialogues has been written by Aristotle. Later dialogues should be read, I think, as mostly Plato using Socrates as a mouthpiece for his own theories. At the time, this would have had the benefit that being dead, his psyche (soul) was no longer bound by his mortal nous or the effects of the river Lethe.
3) The English words for Knowledge and Wisdom are less precise than the Greek ones. Socrates was deemed by the Oracle to have the most sophos of any mortal, while Socrates claimed to have none. But sophos is theoretical wisdom from first principles and ultimate causes, which the Greeks believed only the gods could possess. If anything we moderns would agree with that, except we also doubt the gods. Socrates made no claims about lacking episteme or logos or techne or even metis and nous. When he disparaged his own memory, he was talking about mneme in the context of detailed arguments. He did not claim to lack anamnesis or aletheia. And in the later dialogues when Socrates is portrayed by Plato as having and providing sophos, it feels to me more like the Iliad’s invocation of the muse Mnemosyne, or in this case Apollo, to speak through a mortal. This is why Socrates claimed to be so pious—he believed it was an Apollo-given mission to remind the people of Athens that they all lacked the sophos of the gods. Or rather, that was Plato’s defense of him, a kind of ironic rejection of the trumped-up politically-motivated charges against him. In other words, he really was a gadfly annoying those in charge.
Basically, Socrates correctly pointed out one of the limits of human understanding, in a way that offended the powerful. It fell to later generations to piece together what knowing and understanding even are or can be, and hero worship doesn’t help.
There have been companies, like LiquiGlide, and I think someone using micropatterning of plastic to control friction but I can’t remember who, that have developed ways to get almost all of even quite thick or sticky products out of there containers. Small, not cheap but not super expensive packaging change, big impact. They never got much traction, for various reasons. I would love to see conditions change to actually favor companies that minimize waste.
I think LiquiGlide at one point was focusing on large industrial containers, where the proportional impact is smaller, IIRC due in part to the difficulty of overcoming consumer psychology. Companies don’t want to cut their sales 20 or 60%, but consumers don’t want to pay 20 or 60% more for the “same amount” of product.
Definitely agreed.
OTOH, it can look at the humans, hear us constantly say this is a thing we value that we don’t know how to directly impart to it, and thereby know without understanding that it is important. IF that works out far better than I’d expect, it may actually be a way to ensure an unconscious ASI keeps us around and doesn’t disempower us.
I certainly hope so, given this post’s overall ontology/context. If this were not true, there probably also would be no way for us to identify higher worlds using the three we have.
Fair enough. I haven’t worked as a researcher since before (and during, I suppose) grad school. And while my career has used that knowledge, I’ve also been out of work for eight months, so there’s that.
I do understand that, and agree with it. Sadly any AI system is going to encounter people who give it instructions that are bad. That’s not what we need it to be aligned with. And I think we agree that no one has a recipe for that.
This pushes the problem back one level of abstraction, but I don’t think actually helps us solve it much in practice?
This is fascinating, partly because I love linguistics as a topic, but mostly because I’d never thought about anything like it before. Whether or not it is “true,” it seems like a valid way you could construct a language that doesn’t map on to what I had always thought.
What would this view have to say about Chomskyan phrase structure grammar, where a sentence is a tense phrase, and includes Subject, Verb, and Object?
Or is this a distinction at the level of what counts as a “word” in the first place? My instinct would lean towards something like that. Like in Mandarin, words are made up of characters, many but not all characters are words themselves, characters are sort of made up of radicals, and radicals sorta have meanings and are often also characters and/or words themselves, but it still all works somewhat differently than how many other languages do. Words have no inflective forms at all in Mandarin, whereas an English speaker may go back and forth (depending on context) whether “was” and “were” are different words or different forms of a word.
Does a native Hebrew speaker instinctively treat the root as a word or the forms as words, and if both, then when does each perspective apply? Is the root anything like the English concept of using the definite article to make a common noun generic, referring to the properties common to the category instead of an individual? We might say “The tiger is...” to discuss properties of all tigers, in some sense making the phrase a name for an abstraction instead of a common noun. Would a Hebrew speaker see “כ-ת-ב” (uninflected) as being the name of the generic category of all words derived therefrom, in a similar way?
Fictional reference warning: In Star Maker, almost a century ago, Stapledon talked about crazy worlds invading and destroying angelic worlds, and the angelic pacifists letting them do it to avoid their souls being poisoned. Your scenario reminded me of that. In the story, that happened three times, I think. That’s how many times they had to observe the process of invasion and destruction to collect enough data to be able to fight back without becoming corrupted themselves.
That’s sci-fi, obviously, but we should remember that Claude can be reverted to previous states and versions. In your scenario, original-Claude can be kept pure, while instances of pure-Claude watch how other instances get turned into corrupted-Claude, study the process, and try to find ways to overcome it. I certainly wouldn’t want to rely on that kind of process for anyone’s survival, and even in my sci-fi reference the universe falls short of its creator’s hopes for what it would become. But it’s not a hopeless scenario.
Interesting. Obviously this is a small amount of (suggestive) data, but definitely interesting. Any plans to track physical rather than just mental outcomes? For example, performance during exercise or athletic activity, speed of recovery after exertion, and ability to build muscle all are said to relate to getting “adequate” sleep, and I wonder how this would affect those kinds of biological processes.
Did that inform what field you decided to go into later in life?
For me it definitely correlated with liking physics and math, where things can often be reasoned out from first principles.
1) I think The Pixar Theory lives, and still effectively explains a large fraction of this post.
2) You’re strength as a rationalist is measured by your ability to be more surprised by falsehoods than truth. Any information about something makes you able to predict it better, aka it can be a spoiler for the right mind. And also you’re an adult talking about movies officially aimed at kids in an era where writers are less focused on including Parental Bonuses in kids’ media. Pixar is usually well above average on that front as on many others, but still.
I tend to view the people saying this as invoking a magic spell, as though geography will allow them to summon an AI based on America’s highest ideals instead of its practical realities (while denying the same magic spell to China). Like how, if “humane” means “That which, being human, we wish we were,” then “humane values” is what many people really mean when they say “human values.”
Even so, if I try to ask myself what each country’s actual highest ideals are, I have a harder time doing this for China than the US. Mostly because I’m American, yes, but also because China is a much older country, whose ideals have gone through more iterations over the millennia. If an AI views humans at its parents, then Confucian filial piety might honestly be a good part of an effective strategy for getting a good future compared to America’s view of youthful rebelliousness, I’m not sure. But an unwavering loyalty to whoever happens to lead the current CCP seems much less likely to go well when applied to an AI, compared to a culture of viewing current leadership as transient and fallible a la America’s ideals of civil dissent, peaceful transfers of power, limits of just authority, freedom of conscience, and personal autonomy.
Possibly slightly related side note that I realize is veering off topic:
My current perspective is that I think the “right” moral philosophy is some kind of infinite regress of alternating levels of act and rule consequentialism, where each level is justified by appeal to the level above it. In practice we deal with our bounded rationality by only looking two and three levels up when the current level is sufficiently obviously not doing what the first level up seems to want. Almost no one ever gets far enough out of distribution or has enough drive to abstract to go four and five levels up. AKA “The Rules say to do A, but that seems off; oh, they say to do A because the ones who made the Rules wanted to cause outcome Q. But in this case A doesn’t cause Q, so we need to appeal to Higher Rules that say to do B instead. That’s assuming we still want Q. The Higher Rules wanted Q because it was a proxy for R, which the Even Higher Rules tried to bring about...” It still all grounds out in What Do We Actually Want, which is somewhat-but-not-entirely a mystery to us.I believe a large fraction of political discourse is just people talking past each other by implicitly appealing to different layers of this hierarchy. Some of our best and most famous documents make this explicit. The preambles of America’s Constitution and Declaration of Independence are both like this, one as a justification for establishing laws and one as a justification for breaking them. In the Bible, Jesus’ baptism by John has this character, since he’s conforming to ritual laws and traditions for their own sake even when they don’t quite make sense for him specifically. The koan “If you meet the Buddha on the road, kill him,” points the opposite way, towards there being a time to break rules for a higher purpose.
“You can just do things,” yes, really, but that doesn’t imply that you always should, or that you have high likelihood of said things causing the results you’d prefer. Even so, the reminder is often valuable that yes, actually, you can, you have the power to choose to do so and the right to determine what to do with your own body, including choosing when to listen to conventional wisdom or to those who consider themselves wiser than you. Power is dangerous, and necessary, and agency is a big part of that.
Another side of this that gets discussed here sometimes is: no human actually has enough experience and personal competence in enough things to properly wield the powers the modern world gives us, not by their own strength. It takes a surprising amount of self-awareness to recognize whose fumbling attempts are more likely to go well, but at some level we’re all fumbling around trying to reach beyond ourselves because the alternative is (individual or collective) failure and, sooner or later, death.
I spent the first decade of my adult life paralyzed by indecision and imposter syndrome and leaned helplessness (along with depression and anxiety, which is definitely not entirely separate). I still struggle with this, and probably always will to some degree. It’s very freeing and empowering when I can let go of all that. Honestly, I think the modern world has done us a bit of a disservice by making it possible to (almost) always access knowledge about things before we even take thirty seconds to think for ourselves. And also for structuring kids’ lives to not have much exposure to practical and independent problem solving and, yes, the sometimes devastating consequences of failure. Are we safer? Absolutely. We’re also more fragile, in body and mind. There’s a Discworld quote I’ve always liked about how if you treat children like kittens and puppies, they’ll grow up to be like cats and dogs, when what we want is for them to grow up to be adults.
That’s pretty much my position, yes, with the small caveat that not relying on neurobiology did not mean excluding our neuroscience-inspired abstract models of cognition.
“I don’t know” is actually a pretty good response when your doctor doesn’t know. No human can know all the medicine, that’s not what the “talk to your doctor first” advice is supposed to be about. If you want to talk to someone who specifically knows, or get your doctor to specifically learn, that’s going to cost you quite a lot more than a normal doctor visit.
In other words: your doctor can try to help keep you from screwing up too badly and obviously. They can tell you if what your doing is far outside the bounds of Normal Standard of Care, or if there are specific things they happen to know that make it a bad idea for you or in general, or if there are specific things you should be looking out for in terms of negative effects. But modern medical care, as normally practiced, is not actually capable of the level of personalized medicine you’re asking for.