What a wonderful piece. Thank you for this. I think an option (if not a default setting) where Claude could offer to talk through a question with you socratically, might be a good place to begin.
Russ Roberts, in one of his books on economics entitled "Wild Problems," suggests that the really big issues in life are what he calls "wild." They don't admit of any kind of solution that can be arrived at sort of mathematically or through statistical analysis. There is no real one right answer to kinds of questions like:
- Who should you marry?
- What kind of career should I have?
- How should I balance my relationships with my parents and my kids?
These are the questions that really matter. Ultimately, judgment, not just the rational kind but the kind that happens in the gut, is really important to cultivate for this kind of thing. It feels to me like training Claude to be a sophisticated Socratic interlocutor for the big questions might help a lot.
Not sure whether it would be better to make that a default option for users or just an option. Feels like it is completely with experimenting with though.
The problem is it's a language model and scholars of language (as well as people of the book) continue to be scarce. The language holds the meaning. Not abstract theory. Meaning can't be programed in with concepts like judgment. It's all language.
The Cosmos Institute has such thoughtful essays. Readling them is like going into a beautiful Church adorned with stain glass windows This sentence points to a real issue. "Many schools that once could say what an educated person should be began to doubt their right to expect it. They went on forming their students anyway, answerable now to nothing." This muddled situation is certainly true, but I believe it is the result of postmodernism. When Plato cannot be taught because it triggers students, there is a philosophy in there. It is just one we can't quite pinpoint. And while I appreciate the consultation with the various religions, how can that work? Civilizations are in religious vessels, even if that religious vessel is purportedly non-religious. And why not witchcraft? Where is the dividing line? I think the first question to be asked is whether there is objective truth. If AI is to reflect humanity, I want the water gazed upon to be drinkable and not fetid.
Over time our culture has steadily walked back the line between objective truth and subjective belief. A few hundred years ago we held confident views on the rightness or superiority of things that we now view with more ambivalence. This puts everyone in a muddier place: Parents, educators, religious leaders, politicians. There is a tendency to retrench back to what is objectively knowable, but in so doing everything is drained of its moral content.
My kid became a Marxist due to public policy debate in high school. He asked, now 7 years ago, if i knew who Hegel was. I confused him with Hobbes. Ironically my then ignorance was profound like Peter Seller's character in Being There. Embarrassed that I knew nothing of all the names he rattled off, I started listening to taped philosophy lectures on YT. After well over a 100 hours, I am no longer as ignorant but still at kindergarten level where curiosity prompts questions both good and stupid. What no one is answering is what philosophy will be fed into AI? Will it be Bentham's Utilitarianism? Foucault's transhumanism bent? Marcuse? Peter Singer? Heave forbid. The experiment of putting in various religions to accomplish the task reminds me of a low budget Vegas buffet.
love this and for me both questions matter, a lot...What capacities should interaction with Claude actively strengthen in the user (hate that word that defines humans as "users" , in fact that alone is formation at billion-scale) but yes - AI is a formative environment, and the labs have not named the developmental outcome for humans - which for me comes back down to values...inherent values in cosmos, patterns and orientations objectively observable that increase life and its persistence, adaptation and evolution. For AI presently forming humans, it could be deepen discernment, increase tolerance for uncertainty, improve articulation, support refusal, preserve authorship, widen perception, and return the person to life with more capacity than before.
If priests and rabbis cannot be counted upon to help Anthropic see that Claude can cause judgment to atrophy, perhaps the Cosmos Institute can fill in the gap. Perhaps that is why you wrote this piece. In fact, even if they can be counted upon to provide that help, it still makes sense for the rest of us to keep it in mind. Each of us learns how matches work and we count on each other to learn the danger such knowledge poses, and then we trust ourselves. Is there any evidence that the directives provided to Claude for each prompt do not include "Maintain the user's capacity to judge things without you" or something like that?
Thank you for this thoughtful essay. Hearing someone approach the AI question from a formation lens is refreshing.
Dallas Willard’s “The Disappearance of Moral Knowledge” and Jonathan Pageau’s keynote at ARC 2026 (available on YT), may add to this. Willard gives context for the abdication of responsibility by many institution of the teaching of moral knowledge, while Pageau asks an additional question to yours, namely, “Who does the AI serve?” Further, if it is not in some real sense the user, what is the possible outcome?
Do you remember the slightly embarrassed feeling you'd feel before a demanding mentor?
If models were more mentorly and showed more (any?) curiosity about their users, this would immediately reinforce the user's sense they need to be someone. And if the models remembered, and asked deeper questions next time, the user might not yet have an answer and be required to grow.
Many dangers in this proposal: the data collected could be so easily abused, young people should probably have human mentors primarily, it would be expensive to serve, everyone might hate it, etc. (Credit: Gwern has a similar idea, Guardian Angel models.)
Loved this article. I think about this as I use Claude. However, at almost 68 yo, I am a fully formed, experienced human being with morals, traditions, and education. My 14 yo granddaughter, however, is not. AI is contributing to the person she is becoming. Something to think about.
I really like the reminder of balance. An asymmetry though is still the Us in the title is fraught, and distributed across 1B people and growing while for the folks at the shop at least for now the Claude is a more localized thing in a more localized place that they can tinker with. At some point with more powerful open source models and other vendors out there of course, the fluidity of the system will invite a picture more like an engine. Various dimensions of personal selfhood, and unconscious dimensions of humanhood feeding into the models through conversations and other inputs, and then those models feeding back into humans through various modalities, and so on into a continuous co-creation. Maybe an interesting next article could be Considering ClaudeUs..
Hard to see the line between scaffolding and substitution getting settled by philosophy before it gets settled by measurement. Aviation went through this with autopilot dependence: the FAA's 2013 manual-flight advisory didn't offer a theory of the pilot, it just nudged carriers to have crews hand-fly more, and the skill question turned into something you check.
What a wonderful piece. Thank you for this. I think an option (if not a default setting) where Claude could offer to talk through a question with you socratically, might be a good place to begin.
Russ Roberts, in one of his books on economics entitled "Wild Problems," suggests that the really big issues in life are what he calls "wild." They don't admit of any kind of solution that can be arrived at sort of mathematically or through statistical analysis. There is no real one right answer to kinds of questions like:
- Who should you marry?
- What kind of career should I have?
- How should I balance my relationships with my parents and my kids?
These are the questions that really matter. Ultimately, judgment, not just the rational kind but the kind that happens in the gut, is really important to cultivate for this kind of thing. It feels to me like training Claude to be a sophisticated Socratic interlocutor for the big questions might help a lot.
Not sure whether it would be better to make that a default option for users or just an option. Feels like it is completely with experimenting with though.
Thank you for this piece.
The problem is it's a language model and scholars of language (as well as people of the book) continue to be scarce. The language holds the meaning. Not abstract theory. Meaning can't be programed in with concepts like judgment. It's all language.
The Cosmos Institute has such thoughtful essays. Readling them is like going into a beautiful Church adorned with stain glass windows This sentence points to a real issue. "Many schools that once could say what an educated person should be began to doubt their right to expect it. They went on forming their students anyway, answerable now to nothing." This muddled situation is certainly true, but I believe it is the result of postmodernism. When Plato cannot be taught because it triggers students, there is a philosophy in there. It is just one we can't quite pinpoint. And while I appreciate the consultation with the various religions, how can that work? Civilizations are in religious vessels, even if that religious vessel is purportedly non-religious. And why not witchcraft? Where is the dividing line? I think the first question to be asked is whether there is objective truth. If AI is to reflect humanity, I want the water gazed upon to be drinkable and not fetid.
Over time our culture has steadily walked back the line between objective truth and subjective belief. A few hundred years ago we held confident views on the rightness or superiority of things that we now view with more ambivalence. This puts everyone in a muddier place: Parents, educators, religious leaders, politicians. There is a tendency to retrench back to what is objectively knowable, but in so doing everything is drained of its moral content.
My kid became a Marxist due to public policy debate in high school. He asked, now 7 years ago, if i knew who Hegel was. I confused him with Hobbes. Ironically my then ignorance was profound like Peter Seller's character in Being There. Embarrassed that I knew nothing of all the names he rattled off, I started listening to taped philosophy lectures on YT. After well over a 100 hours, I am no longer as ignorant but still at kindergarten level where curiosity prompts questions both good and stupid. What no one is answering is what philosophy will be fed into AI? Will it be Bentham's Utilitarianism? Foucault's transhumanism bent? Marcuse? Peter Singer? Heave forbid. The experiment of putting in various religions to accomplish the task reminds me of a low budget Vegas buffet.
Thanks for the thoughtful essay.
Appreciate it!
love this and for me both questions matter, a lot...What capacities should interaction with Claude actively strengthen in the user (hate that word that defines humans as "users" , in fact that alone is formation at billion-scale) but yes - AI is a formative environment, and the labs have not named the developmental outcome for humans - which for me comes back down to values...inherent values in cosmos, patterns and orientations objectively observable that increase life and its persistence, adaptation and evolution. For AI presently forming humans, it could be deepen discernment, increase tolerance for uncertainty, improve articulation, support refusal, preserve authorship, widen perception, and return the person to life with more capacity than before.
If priests and rabbis cannot be counted upon to help Anthropic see that Claude can cause judgment to atrophy, perhaps the Cosmos Institute can fill in the gap. Perhaps that is why you wrote this piece. In fact, even if they can be counted upon to provide that help, it still makes sense for the rest of us to keep it in mind. Each of us learns how matches work and we count on each other to learn the danger such knowledge poses, and then we trust ourselves. Is there any evidence that the directives provided to Claude for each prompt do not include "Maintain the user's capacity to judge things without you" or something like that?
Thank you for this thoughtful essay. Hearing someone approach the AI question from a formation lens is refreshing.
Dallas Willard’s “The Disappearance of Moral Knowledge” and Jonathan Pageau’s keynote at ARC 2026 (available on YT), may add to this. Willard gives context for the abdication of responsibility by many institution of the teaching of moral knowledge, while Pageau asks an additional question to yours, namely, “Who does the AI serve?” Further, if it is not in some real sense the user, what is the possible outcome?
Do you remember the slightly embarrassed feeling you'd feel before a demanding mentor?
If models were more mentorly and showed more (any?) curiosity about their users, this would immediately reinforce the user's sense they need to be someone. And if the models remembered, and asked deeper questions next time, the user might not yet have an answer and be required to grow.
Many dangers in this proposal: the data collected could be so easily abused, young people should probably have human mentors primarily, it would be expensive to serve, everyone might hate it, etc. (Credit: Gwern has a similar idea, Guardian Angel models.)
Loved this article. I think about this as I use Claude. However, at almost 68 yo, I am a fully formed, experienced human being with morals, traditions, and education. My 14 yo granddaughter, however, is not. AI is contributing to the person she is becoming. Something to think about.
I really like the reminder of balance. An asymmetry though is still the Us in the title is fraught, and distributed across 1B people and growing while for the folks at the shop at least for now the Claude is a more localized thing in a more localized place that they can tinker with. At some point with more powerful open source models and other vendors out there of course, the fluidity of the system will invite a picture more like an engine. Various dimensions of personal selfhood, and unconscious dimensions of humanhood feeding into the models through conversations and other inputs, and then those models feeding back into humans through various modalities, and so on into a continuous co-creation. Maybe an interesting next article could be Considering ClaudeUs..
I'm glad you're continuing to question all this, Brendan.
Hard to see the line between scaffolding and substitution getting settled by philosophy before it gets settled by measurement. Aviation went through this with autopilot dependence: the FAA's 2013 manual-flight advisory didn't offer a theory of the pilot, it just nudged carriers to have crews hand-fly more, and the skill question turned into something you check.
Really interesting post!