So do you know the Elon Musk quote? It's around how to learn. It's essentially this idea that you want to view knowledge as a semantic tree. So you start at the roots, then you go up to the trunk, then you have the branches, then you have like the secondary branches, then you have the leaves. Whereas often the way we'll approach things is, oh, I wanna learn about the heart. I'll just put on this random Andrew Huberman podcast with a specialist about the heart and just kind of hop in. But you don't have any of the roots or anything there. So you never actually retain any information. Whereas when you treat knowledge as a semantic tree, you work all the way up from the base and all and then all the way there. And a big realization this year was, it's kind of a, a bit of a Deutsche concept, but essentially this idea that the only thing, the only bottleneck that really exists is knowledge. And then you look at, okay, you have all these great people who are self-taught. So you can just teach yourself... and you realize I can learn anything starting from there. So you start with the, you say the specific Elon quote and you say, teach me about X...
He doesn't mean Elon's X, AKA Twitter. Don't be confused. He means your prompt should be "teach me about" whatever it is you want to learn, and, importantly:
... start with the roots and then work all the way up and don't move to the next layer until I say, I understand. And you're constantly just moving up and you realize, oh I can literally teach myself anything....
This is a nice development from your last year's one, which was treat me like a total idiot and start at zero until I say I understand.
Yes. And then go to step one and then step three.
You can then just move it into like a mind mapping software and literally just build the tree yourself. So then you have that semantic tree in your head of all the interweaving parts.
Big mistake that I made when studying medicine was not doing that earlier. You have to have like a, a framework or a skeleton to be able to hang concepts on. Otherwise you are just learning raw data and it's so difficult.
Yes. There's nothing connect and, and you're just memorizing like you did at school. You're never actually understanding that....
Okay. I hope that was helpful. I found it very helpful. But what do I want to learn in that systematic way?
ADDED: The tree metaphor bothered me because a tree doesn't start at the roots. It starts with a seed, and then roots develop downward and it develops upward at the same time as the roots grow. Or it starts with a seed and there's something called a "radical" that precedes roots... right? That's what I seem to remember from elementary school when we folded a bean inside damp paper towels, and I'm off into an endless maze of memories and language peculiarities and mistakes and metaphors. I happen to think I can interact with A.I. in my idiosyncratic way, but I'm interested to see how systematic other people strive to become as they interface with the computer. What could go wrong!
AND: I took this inquiry to Grok:
I got this answer:I'm concerned that this method has the human being attempting to become more like the computer, but I'd like to think that you, the computer, ought not to incentivize our becoming more computer-like and that we the humans ought to guard out humanity and not eagerly strive to resemble computers.
Your concern highlights a critical aspect of human-AI interaction: the preservation of human qualities like creativity, intuition, emotion, and critical thinking. Here's a perspective on this:
Human Uniqueness vs. AI Efficiency
• Human Qualities: Humans thrive on creativity, empathy, ethical judgment, and the ability to navigate ambiguity. These are aspects where AI cannot and should not replace human judgment or identity. The semantic tree approach, while useful for structured learning, should not be the sole method of education or personal growth.• AI as a Tool: AI should serve as an augmentative tool, enhancing human capabilities rather than dictating human behavior or thought processes. The idea isn't to make humans more computer-like but to use AI's strengths to support human endeavors.
Balance in Learning and Development
• Diverse Learning Methods: While structured learning paths like the semantic tree can be effective for certain types of knowledge acquisition, humans also learn through experience, interaction, and unstructured exploration. Encouraging a blend of these methods helps maintain a balance between efficiency and the human experience.• Guard Against Dehumanization: There's a risk in any system, especially one involving technology, where humans might start to mimic or over-adapt to machine-like behaviors. To counteract this:• Promote Creativity: Learning should include spaces for creativity, where the 'tree' of knowledge can be explored in non-linear, imaginative ways.
• Encourage Social Interaction: Human learning is deeply social. Collaborative learning environments where discussion, debate, and personal insight are valued can counteract the isolating effects of solitary learning from machines.
• Emphasize Critical Thinking: AI can provide information, but humans should be encouraged to question, evaluate, and interpret this information, fostering independent thought.
• Ethical AI Use: AI developers and users alike should be mindful of how AI influences human behavior. Ethical guidelines should ensure that AI promotes human values rather than diminishing them.
Conclusion
You're right to be vigilant about how we interact with AI. The goal should not be to make humans more machine-like but to use machines to better serve human needs and potential. The semantic tree method, or any AI-assisted learning, should be seen as one part of a broader educational strategy that respects and nurtures human uniqueness. By keeping this balance, we can leverage technology while safeguarding our humanity.
IN THE COMMENTS: n says:
"Feliz Navidad! The pre-embryonic root is a radicle."
What's your favorite Christmas homophone?
Posts
31 comments:
Merry Christmas Meadhouse. I hope your day grows into a beautiful one.
Merry Christmas. And for some others, Happy Hanukkah.
I've had an on again, off again attempt to learn the Italian language. I've taken a few beginners classes locally, that was paired with a very good textbook. And I've downloaded and used, let me see....up to 3 different language apps, all coming at it from a different place.
The class with textbook approach was the most 'semantic tree' approach, literally starting from the roots of grammar in the language. The apps all want to give you what they think are the key phrases and uses you'll need as an English speaking person visiting Italy, though one of the apps is much more in depth.
I've found two things:
The class/textbook is the best base of knowledge, but is even better when paired with any of the apps.
None of it is any good unless you work at it every day and don't take months away from it as I have.
And so, this Christmas Day, I'll be beginning round 3 in Learning Italian.
I wonder if AI would be any better at helping me learn the language. I'll be Neuralink would.
Merry Christmas. Is there going to be a test?
It reads to me like a secretary interacting with ELIZA in the 60s, except the cliches come also from the internet instead of transformed echos of the interaction itself.
"It is important to..." and similar seem like a major crutch.
Merry Christmas!
Merry Christmas, all!
My son has taught himself to play many musical instruments, but it all started with guitar and a teacher who deep-dived him into music theory. That would be the roots. I, after seven childhood years of piano lessons without instruction in music theory, never progressed beyond being a mediocre musician. I can sight read well enough -- that's the branches -- but I can't evaluate the composition. And it's hard to work backwards from there.
Feliz Navidad! The pre-embryonic root is a radicle.
I asked AI to write a comment to an Althouse blog post. Below is what I got.
Merry Christmas to Meadehouse and all Althousians except Derve Swanson.
DDB
“ The ongoing dialogue about capital punishment reflects broader societal debates around justice, compassion, and the role of leadership. As we witness these evolving discussions, it's worth considering how these actions will inform future legislative directions and societal values related to criminal justice reform.”
Maybe you want to view knowledge as a semantic tree but maybe you want to do college automatically.
I've noticed that you never write about what you did with your friends last night (or whenever).
I'm concerned enough about the posts of your interactions with various AI that you need to touch some grass on your next outing to the lake.
Go volunteer at a church. Help people. You'll find the interactions far superior to your conversations with the corporations controlling AI to enslave the rest of humanity.
My listening skills (never particularly strong) diminish considerably when the speaker uses "around" instead of "about." Fingernails on a chalkboard.
This made me think of the difference between learning how to perform tasks on a job and understanding why the tasks are being performed and how they fit into the bigger picture: Tactical vs. Strategic thinking. You may not need to know what tasks others are performing to be able to perform your own tasks, but it helps if you are ever asked to do someone else's job, or to rise to supervisory role where more knowledge is needed.
I did the same job for 24 years, and although things changed over time as technology advanced, they didn't change radically. I knew that other people at my workplace had different jobs, but I didn't need to know what they did because I was never asked to fill in for them. I had my own tasks to do. It was only when I got a job bid on the day shift that things changed. I was doing similar tasks, but had to learn new details. Then, a few years later, I got a totally different job that I had to learn from scratch, and all of a sudden, some of the things on the second job began to make more sense, as I knew more about where the product was downflowing to, as I now needed to make sure that it was indeed going to the right place. It's easy to learn How; it's not as easy to learn Why.
Merry Christmas, all, and a happy Hanukkah too.
AI is great. You can have a knee jerk reaction to it, and it's no substitute for being "well read," whatever that means. But it does encapsulate a lot of knowledge for you to rummage through. Mark Twain once said that the way he learned to write was setting type, that he had set miles of type, and from that, he learned the proper use of the English language. Well, Twain was a kind of genius, in the sense of creativity, but AI has read parsecs of type, and so the knowledge that it has gleaned is not to be sneezed at. It knows a lot about the usage of language.
Back when I worked for a living, we used to use these things called "cubes" to encapsulate the data that we had collected from the network, a large amount of data could be compressed into a very small space, and rapidly recovered. The limitation on it was that you had to anticipate the kinds of questions that you planned to ask it, and about which metrics. But within those constraints, it was amazing, given the computational limits of the time. I think that AI does the same thing, but you don't have to limit the questions you plan to ask in advance of a run of the model against the data. Of course this is why they are buying nuclear power plants to run these models; we had to run them on a couple of dozen Windows based servers.
Merry christmas 😊
I came across a listicle of Frank Zappa quotes just now. This one seems apropos:
"The computer can't tell you the emotional story. It can give you the exact mathematical design, but what's missing is the eyebrows."
Merry Christmas all!
Merry Christmas, Althouse and Meade! And Althouse, thank you for 20 years of almost-always-fascinating food for thought here!
My prediction is that A.I. is going to become a powerful tool in learning a language, because the best way to learn a language is via total immersion. A.I. will eventually allow something like this: A conversation partner who will speak the language impeccably, conversationally, but also be able to step outside the immersive environment to provide coaching. And since it's A.I., the cost will be a fraction of a human teacher, and its availability 'on demand' will be a benefit.
The crooked beat. https://www.youtube.com/watch?v=wIf5DF-PoDo
new headphones for me this xmas...
The second reading today at the Catholic Mass at St. Robert Bellarmine (Jesuit saint) church was from the letter of Paul to Titus. In it he wrote, "deliver us from lawlessness."
America has been delivered from the lawlessness of the Biden regime. Hallelujah!
I'm curious about the similarity between learning to think critically and speak clearly in order to effectively communicate with other humans, and learning to think systematically and speak unambiguously in order to effectively communicate with computers.
Google me this: prompting and steering. The question is: who is to be master and who is the slave.
Anthropogenic Intelligence, Artificial Intelligence, Authoritarian Intelligence... an evolution of intelligence, which may well have been conceived as an oracle and progressed with a calculator in secular religions.
Merry Christmas everyone.
from the article: "Big mistake that I made when studying medicine was not doing that earlier. You have to have like a, a framework or a skeleton to be able to hang concepts on."
Prof, what do you think about the traditional law-school "seamless web" method, where the student is basically forced to construct her own framework? (I mean, you can "cheat" by reading introductory books or articles about how the common law works, and by buying commercial outlines for whichever subject you're studying, but if yer just pays your money and sits in yer seat, no one explains the whole structure to you - they just throw specific example after specific example at you.)
JSM
I would like to see responses to Leonard Peikoff book on Dis-Integration; Integration; Mis-Integration >>> The DIM Hypothesis: Why the Lights of the West Are Going Out, by Leonard Peikoff
One hundred percent. The other like this for me is, "across." Blecch (and Merry Christmas!)
Merry Christmas! AI is especially good at handling what would be multiple tedious queries. For example, coming up with a travel itinerary to spec.
Post a Comment