My blog about philosophy, coding, and anything else that interests me.
Edwin de Wit’s Comments
✓
Identity verified
Thank you for the thoughtful and detailed reply
Re: inconsistency and confusion
I appreciate the examples you give regarding Lulie’s accounts of fun and the inconsistency between advice framed as fun-based productivity vs. productivity-based fun. I agree that this indicates reversal and confusion and can therefore also be potentially harmful (as I also acknowledged in my original reply). However, my main remark about the lack of context was directed more at Dennis’ line of criticism, which seemed to call out proponents of the Fun Criterion for not being productive—whereas I don’t think they themselves ever claimed to be productive because of applying the FC (certainly not David).
I agree that’s plausible. Personally, I don’t care to speculate further, but I do hope to ask her directly about her views on intuition and feeling on my podcast at some point.
It isn’t always clear in your comment when something is a literal quote vs a scare quote. I never wrote or implied that the FC itself isn’t productive. That wouldn’t make sense. Productivity is a property of people.
I see where my phrasing may not have been ideal, but the point I’m trying to make is this: criticizing the Fun Criterion on the basis of its proponents’ productivity only really holds if they themselves claim to be highly productive through the use of it. In that case, calling out the inconsistency makes sense. But as far as I can tell, that isn’t what David is doing. So I think the better mode of criticism is to engage the content of the Fun Criterion theory itself, rather than the productivity of those who advocate it. I mean this epistemologically; a theory is best criticized on its content, not its source or the personal results of its proponents.
Re: emotions
David has indeed said that not-fun can be felt through feelings. No disagreement there. My understanding, though, is that this doesn’t establish emotions as a central part of fun itself—it just shows that we can notice when we are not having fun, when problem-solving feels thwarted.
The follow-up question then becomes: can we can identify fun itself based on feelings? On that topic David has said that fun is not an emotion. That implies to me that fun itself is not something we can identify through feelings. Furthermore, he's also pointed out that many different emotions—positive or negative—can accompany the state of fun. To me, that reinforces the idea that fun can't be identified by any particular feeling and that fun is emotion-agnostic.
That’s why I also disagree with this statement:
“And we have no direct, explicit visibility into whether we have attained the state of fun; it is emotions that tell us whether we have.”
Emotions may sometimes signal when we’re not having fun, but they don’t in tell us when we are. For what it’s worth, I see this not only as David’s position but also as the best current explanation of fun, and the most practical one for applying it as a heuristic.
Re: unconflicted
The article once read like an unconditional endorsement of David’s and Lulie’s ideas. That has since changed. This paragraph, in particular, illustrates the very reversal I now caution against:
On the whole, in the long run, I expect greatness almost always to come from enjoyable activities, and rarely from suffering. In other words, of all the greatness at any given time in history, most of it will have been the product of fun – and I expect the ratio to improve in the future. Do not take that to mean that fun will always lead to greatness – any search for a guarantee of greatness will almost certainly produce suffering. Greatness is not the goal. Fun is the goal, even when you have fun pursuing greatness. And if you pursue fun, you may sometimes achieve greatness as a pleasant side effect.
Got it. Thank you. That’s a strong example, and your articulation of the risks of the reversal is a valuable takeaway.
I’m really glad you made this thorough analysis and brought fresh takes on the Fun Criterion (FC). My main takeaway, and one I agree with, is that while the FC is incredibly valuable in many cases, it’s insufficient as a moral heuristic for deciding “what to do.” I also share your view that the boundaries of what “Fun” means haven’t been as clearly defined as they could be, leaving room for misinterpretation and misapplication.
Detailed remarks
I appreciate your fair and accurate presentation of the primary interpretation of the FC, namely unconflictedness, and its merits. Regarding the risks and shortcomings you point out around “emotion” and “productivity,” I have a few thoughts.
On emotions, I think your concerns are justified, and I’m glad you’ve laid them out. That said, I don’t see this as a shortcoming of the FC in my interpretation of “Fun” in the Deutschian sense. As you point out, David has said several times (such as in the Dwarkesh Patel interview and in this tweet) that fun is not about emotions. Therefore, although I appreciate and support your account of the risks of following your feelings instead of reason, I don't view it as a critique of the FC. Admittedly, there are quotes and descriptions that could lead people to misinterpret it as “Follow the fun.” That's a shame, and I agree David and Lulie could be clearer and more consistent in defining the boundaries of fun when they mention it. Several of the utterances of Deutsch and Tanett indeed go directly against the idea behind the FC and undermine it, as you rightfully point out.
On productivity, I think your criticism is especially valuable because many people, myself included, see the FC as a productivity method. Fun can be a useful heuristic for how to do something and what not to do (avoiding coercion). But it doesn’t give direction or structure. It doesn’t inherently push us to act, take responsibility, work rigorously, or confront the unknown. If we want to go beyond simply making our local problems more fun, we need additional moral heuristics. That said, I haven’t seen Deutsch or Tanett claim outright that the FC is the ultimate moral heuristic for maximizing productivity or efficiency. The quotes you cite don’t convince me otherwise since they’re narrow and lack context. So the sharp, personal criticisms that “the FC isn’t productive” because its proponents aren’t productive don’t seem like strong or necessary arguments to me. The more substantive critique is that the FC falls short in answering moral questions about what to do. That’s a valuable line of criticism, and I hope you continue supplementing it with the heuristics and insights you uncover in your research, such as those from Objectivism.
Other notes
On the merits of the FC, I agree with fun as an error-correction signal. But I wouldn’t describe the experience of this signal as “fading fun” since fun isn’t an emotional experience. I’d instead call it the emergence of a thwarted conflict. A hang-up.
My reading of David’s view is that emotions are downstream from the three types of knowledge, much like Ayn Rand describes.
I don’t share your interpretation of Lulie Tanett’s comment in the FC video that you can “deliberate by intuition.” I don't think she meant letting your thought processes be guided by emotions you can’t account for. Instead, I pictured it like a grammar tutor being asked whether sentence A or B is correct: you have a sense (produced by inexplicit knowledge) that A is correct. I view that as inexplicit deliberation. I’m not claiming this is the correct interpretation, but rather offering it as another perspective to consider.
You mention rethinking some stances from your original “Unconflictedness” article (one of my favorites). I reread it, and couldn’t find anything I disagreed with. I’m curious which parts you feel you’ve moved away from.
Interesting read! Do I understand correctly that you say that "true or false" (in a binary way) is to be used to describe whether an explanation holds up to criticism or not (in isolation)? And, that "better or worse" (as opposed to truer or falser) is more fitting terminology when comparing explanations to each other?
Yes, you bring up a good point that in-the-moment awareness is a better time to detect what’s going on. I do think it’s tricky to know exactly what kind of “push-through” scenario you’re facing. There’s this idea of a Goldilocks zone of challenge that leads to flow (or fun) —where you are in flow when it’s right at the edge of your competence. Too far above or below your skill level, and the challenge turns into conflict. Staying in that flow state without some degree of nudging or self-coercion to guide you back does seem difficult (but not impossible).
Thank you again for sharing that definition—it's a great one, and I agree with it. When I first started writing this reply, I was trying to justify micro-coercions in service of genuine long-term goals. While that’s probably how many of us operate—and I know I’m guilty of it—it also carries a troubling implication: that certain inner conflicts are things we must just push through.
After some reflection, I don’t think I agree with that. I tend to self-coerce reflexively, forcing my attention toward what I should be doing instead of trying to understand the conflicting impulse in the moment. Justifying coercion for the sake of a greater goal feels like a crude, imperfect solution—and there must be better ways to handle these inner tensions.
Through some experimentation, I’ve noticed that simply acknowledging the impulse—without acting on it—often reduces the tension more than expected. Maybe that’s because the conflict is created by the should-impulse itself, rather than the competing desire. And even when the other impulses remain, finding a common preference—or just taking a break—can be more effective, and less harmful, than pushing through.
This has been a calming and empowering insight for me, and I’ve been trying to remind myself of it whenever I catch myself drifting or self-coercing. Your writing—and Lulie Tallot’s—has been especially helpful, and I’d be keen to read more if you ever decide to write further on the subject.
Great article! I’m curious exactly when we can consider something self-coercion. My interpretation of Schwarzenegger’s philosophy of “needing a vision” is primarily about establishing a sufficiently meaningful problem that you’ll enjoy working towards most of the time. Even when there are road bumps along the way, that problem is still rewarding enough that you choose to keep working despite not feeling like it in the moment. I’d posit that it only becomes self-coercion when the vision you’re working toward no longer genuinely interests you—not necessarily when you push through a bad day. The litmus test could be to see whether, after pushing through, you feel good about it or not. What do you think?
Hi Dennis, Impressive theory! A lot resonates and I've been enjoying the book as well. I do have a couple of parts of the theory that I'm struggling to understand:
1. The Relationship Between Knowledge and Functions
My current understanding is that knowledge (a theory) is information with a potential causative effect. Examples include conscious beliefs, feedback from reinforcement learning, memes, feelings, intuitions, emotions, preferences, desires, internal state signals, and external stimuli. I see knowledge being used by our minds in two broad ways:
As part of a function. If we define a function as consisting of inputs (initial conditions), an expression, and outputs (which may include side effects or actions), then knowledge comes into play in the following ways:
As inputs (initial conditions) for functions.
Within expressions, shaping logic such as: If X complies with theory Y, then Z is true (which I believe is also referred to as a "policy" or "criterion" in the book).
To compare theories and identify contradictions in the Popperian sense. I recognize that “comparison” is itself a function—checking whether both theories are true, and if not, identifying a problem.
From this, I conclude that knowledge on its own is not a function, but for it to be enacted (and replicated), it must be embedded within one. Does this align with your perspective? I’ve come across passages in the book that clash with this, such as:
"We do not need to carefully examine different kinds of knowledge such as emotions, hopes, goals, desires, and preferences to conclude that they are all functions: we know this from the universality of computation. The brain is a computer, so everything it runs must be a function (that includes the ever so elusive qualia)."
2. Persistence of Non-Embedded Knowledge
This question depends on whether we indeed see knowledge as different from functions. If we do, then some non-ephemeral knowledge might exist without being embedded in a function, enacted, or replicated.
For example, let’s say I firmly believe Cats is the best movie ever. What sustains this delusion inside me if it’s not part of a functional self-replication process?
3. Composition of the Function Pool
What exactly does the function pool contain in your view? Does it include:
Run-time generated thoughts?
Genetically determined functions?
Acquired memetic functions?
Reinforcement learning functions and their results?
Or are some of these functions located elsewhere?
4. Competition Within the Function Pool
What prevents critical functions—such as primal hunger instincts or the autonomic nervous system—from being outcompeted by highly effective self-replicating functions?
This is especially relevant for functions that remain mostly dormant and rarely execute, such as a fight-or-flight response. If they don’t get triggered often, and thus don’t replicate regularly, what ensures their persistence?
5. Genetically Determined Functions in Humans vs. Animals
Do you see a fundamental difference between how a dog and a human experience and act on hunger?
Are both experiencing a conflict between two competing theories, or is that a uniquely human capability? I wonder if a dog simply has a hunger-alleviation function that is blindly triggered when hunger arises, or if they also have some ability to evaluate conflicting theories.
(P.S. I align with the view that humans are non-deterministic, while dogs are deterministic.)
6. The Evolution of Function Replication
I’m curious about the early evolution of function replication. In the book, you mention that some functions do not have side effects (actions), such as general relativity.
What exactly do you mean by that? If a function has no side effects, what is its output?
Wouldn’t it still invoke an action of some kind? If it does, then it would lead me to two possible replication mechanisms:
Replicating the output function itself.
Replicating the physical action that the function’s output produces.
I believe the book primarily focuses on (1), drawing a parallel with the RNA world hypothesis. However, I struggle with that theory because I can't show how an output of a function would make a copy of the original function. It could be an input for another function, trigger a loop, or cause a chain reaction. But for me, it's hard to imagine how a function would directly produce an imperfect copy of itself.
On the other hand, you also hint at the alternative approach (2) in the book:
"If your program can copy a function’s implementation without reading it, based solely on inputs and outputs it tries out and perhaps some clues about its implementation, you may be on your way to building AGI."
This raises an interesting possibility: Could early function replication have been driven by parsing and mimicking actions rather than directly copying functions?
If so, this aligns with the hypothesis of behavior parsing and memetic evolution. It’s conceivable that our ancestors evolved self-replication mechanisms in the same way that animals developed meme-parsing abilities—to improve survival at runtime relative to competing gene-machines.
Perhaps the earliest function replication was not self-replication but rather an attempt by another part of the body or mind to parse and reverse-engineer an action into a new function. If so, this could imply that function replication initially emerged as a function-copying function rather than a self-replicating function. Perhaps this is part of the meta-algorithm itself?
Thank you for the thoughtful and detailed reply
Re: inconsistency and confusion
I appreciate the examples you give regarding Lulie’s accounts of fun and the inconsistency between advice framed as fun-based productivity vs. productivity-based fun. I agree that this indicates reversal and confusion and can therefore also be potentially harmful (as I also acknowledged in my original reply). However, my main remark about the lack of context was directed more at Dennis’ line of criticism, which seemed to call out proponents of the Fun Criterion for not being productive—whereas I don’t think they themselves ever claimed to be productive because of applying the FC (certainly not David).
Re: emotions
See my reply to Dennis: #3733.
Re: determining the correct perspective
I agree that’s plausible. Personally, I don’t care to speculate further, but I do hope to ask her directly about her views on intuition and feeling on my podcast at some point.
Re: productivity
I see where my phrasing may not have been ideal, but the point I’m trying to make is this: criticizing the Fun Criterion on the basis of its proponents’ productivity only really holds if they themselves claim to be highly productive through the use of it. In that case, calling out the inconsistency makes sense. But as far as I can tell, that isn’t what David is doing. So I think the better mode of criticism is to engage the content of the Fun Criterion theory itself, rather than the productivity of those who advocate it. I mean this epistemologically; a theory is best criticized on its content, not its source or the personal results of its proponents.
Re: emotions
David has indeed said that not-fun can be felt through feelings. No disagreement there. My understanding, though, is that this doesn’t establish emotions as a central part of fun itself—it just shows that we can notice when we are not having fun, when problem-solving feels thwarted.
The follow-up question then becomes: can we can identify fun itself based on feelings? On that topic David has said that fun is not an emotion. That implies to me that fun itself is not something we can identify through feelings. Furthermore, he's also pointed out that many different emotions—positive or negative—can accompany the state of fun. To me, that reinforces the idea that fun can't be identified by any particular feeling and that fun is emotion-agnostic.
That’s why I also disagree with this statement:
Emotions may sometimes signal when we’re not having fun, but they don’t in tell us when we are. For what it’s worth, I see this not only as David’s position but also as the best current explanation of fun, and the most practical one for applying it as a heuristic.
Re: unconflicted
Got it. Thank you. That’s a strong example, and your articulation of the risks of the reversal is a valuable takeaway.
TL;DR
I’m really glad you made this thorough analysis and brought fresh takes on the Fun Criterion (FC). My main takeaway, and one I agree with, is that while the FC is incredibly valuable in many cases, it’s insufficient as a moral heuristic for deciding “what to do.” I also share your view that the boundaries of what “Fun” means haven’t been as clearly defined as they could be, leaving room for misinterpretation and misapplication.
Detailed remarks
I appreciate your fair and accurate presentation of the primary interpretation of the FC, namely unconflictedness, and its merits. Regarding the risks and shortcomings you point out around “emotion” and “productivity,” I have a few thoughts.
On emotions, I think your concerns are justified, and I’m glad you’ve laid them out. That said, I don’t see this as a shortcoming of the FC in my interpretation of “Fun” in the Deutschian sense. As you point out, David has said several times (such as in the Dwarkesh Patel interview and in this tweet) that fun is not about emotions. Therefore, although I appreciate and support your account of the risks of following your feelings instead of reason, I don't view it as a critique of the FC. Admittedly, there are quotes and descriptions that could lead people to misinterpret it as “Follow the fun.” That's a shame, and I agree David and Lulie could be clearer and more consistent in defining the boundaries of fun when they mention it. Several of the utterances of Deutsch and Tanett indeed go directly against the idea behind the FC and undermine it, as you rightfully point out.
On productivity, I think your criticism is especially valuable because many people, myself included, see the FC as a productivity method. Fun can be a useful heuristic for how to do something and what not to do (avoiding coercion). But it doesn’t give direction or structure. It doesn’t inherently push us to act, take responsibility, work rigorously, or confront the unknown. If we want to go beyond simply making our local problems more fun, we need additional moral heuristics. That said, I haven’t seen Deutsch or Tanett claim outright that the FC is the ultimate moral heuristic for maximizing productivity or efficiency. The quotes you cite don’t convince me otherwise since they’re narrow and lack context. So the sharp, personal criticisms that “the FC isn’t productive” because its proponents aren’t productive don’t seem like strong or necessary arguments to me. The more substantive critique is that the FC falls short in answering moral questions about what to do. That’s a valuable line of criticism, and I hope you continue supplementing it with the heuristics and insights you uncover in your research, such as those from Objectivism.
Other notes
Interesting read! Do I understand correctly that you say that "true or false" (in a binary way) is to be used to describe whether an explanation holds up to criticism or not (in isolation)? And, that "better or worse" (as opposed to truer or falser) is more fitting terminology when comparing explanations to each other?
Yes, you bring up a good point that in-the-moment awareness is a better time to detect what’s going on. I do think it’s tricky to know exactly what kind of “push-through” scenario you’re facing. There’s this idea of a Goldilocks zone of challenge that leads to flow (or fun) —where you are in flow when it’s right at the edge of your competence. Too far above or below your skill level, and the challenge turns into conflict. Staying in that flow state without some degree of nudging or self-coercion to guide you back does seem difficult (but not impossible).
Thank you again for sharing that definition—it's a great one, and I agree with it. When I first started writing this reply, I was trying to justify micro-coercions in service of genuine long-term goals. While that’s probably how many of us operate—and I know I’m guilty of it—it also carries a troubling implication: that certain inner conflicts are things we must just push through.
After some reflection, I don’t think I agree with that. I tend to self-coerce reflexively, forcing my attention toward what I should be doing instead of trying to understand the conflicting impulse in the moment. Justifying coercion for the sake of a greater goal feels like a crude, imperfect solution—and there must be better ways to handle these inner tensions.
Through some experimentation, I’ve noticed that simply acknowledging the impulse—without acting on it—often reduces the tension more than expected. Maybe that’s because the conflict is created by the should-impulse itself, rather than the competing desire. And even when the other impulses remain, finding a common preference—or just taking a break—can be more effective, and less harmful, than pushing through.
This has been a calming and empowering insight for me, and I’ve been trying to remind myself of it whenever I catch myself drifting or self-coercing. Your writing—and Lulie Tallot’s—has been especially helpful, and I’d be keen to read more if you ever decide to write further on the subject.
Great article! I’m curious exactly when we can consider something self-coercion. My interpretation of Schwarzenegger’s philosophy of “needing a vision” is primarily about establishing a sufficiently meaningful problem that you’ll enjoy working towards most of the time. Even when there are road bumps along the way, that problem is still rewarding enough that you choose to keep working despite not feeling like it in the moment. I’d posit that it only becomes self-coercion when the vision you’re working toward no longer genuinely interests you—not necessarily when you push through a bad day. The litmus test could be to see whether, after pushing through, you feel good about it or not. What do you think?
Hi Dennis, Impressive theory! A lot resonates and I've been enjoying the book as well. I do have a couple of parts of the theory that I'm struggling to understand:
1. The Relationship Between Knowledge and Functions
My current understanding is that knowledge (a theory) is information with a potential causative effect. Examples include conscious beliefs, feedback from reinforcement learning, memes, feelings, intuitions, emotions, preferences, desires, internal state signals, and external stimuli. I see knowledge being used by our minds in two broad ways:
As part of a function. If we define a function as consisting of inputs (initial conditions), an expression, and outputs (which may include side effects or actions), then knowledge comes into play in the following ways:
To compare theories and identify contradictions in the Popperian sense. I recognize that “comparison” is itself a function—checking whether both theories are true, and if not, identifying a problem.
From this, I conclude that knowledge on its own is not a function, but for it to be enacted (and replicated), it must be embedded within one. Does this align with your perspective? I’ve come across passages in the book that clash with this, such as:
2. Persistence of Non-Embedded Knowledge
This question depends on whether we indeed see knowledge as different from functions. If we do, then some non-ephemeral knowledge might exist without being embedded in a function, enacted, or replicated.
For example, let’s say I firmly believe Cats is the best movie ever. What sustains this delusion inside me if it’s not part of a functional self-replication process?
3. Composition of the Function Pool
What exactly does the function pool contain in your view? Does it include:
Or are some of these functions located elsewhere?
4. Competition Within the Function Pool
What prevents critical functions—such as primal hunger instincts or the autonomic nervous system—from being outcompeted by highly effective self-replicating functions?
This is especially relevant for functions that remain mostly dormant and rarely execute, such as a fight-or-flight response. If they don’t get triggered often, and thus don’t replicate regularly, what ensures their persistence?
5. Genetically Determined Functions in Humans vs. Animals
Do you see a fundamental difference between how a dog and a human experience and act on hunger?
Are both experiencing a conflict between two competing theories, or is that a uniquely human capability? I wonder if a dog simply has a hunger-alleviation function that is blindly triggered when hunger arises, or if they also have some ability to evaluate conflicting theories.
(P.S. I align with the view that humans are non-deterministic, while dogs are deterministic.)
6. The Evolution of Function Replication
I’m curious about the early evolution of function replication. In the book, you mention that some functions do not have side effects (actions), such as general relativity.
What exactly do you mean by that? If a function has no side effects, what is its output?
Wouldn’t it still invoke an action of some kind? If it does, then it would lead me to two possible replication mechanisms:
I believe the book primarily focuses on (1), drawing a parallel with the RNA world hypothesis. However, I struggle with that theory because I can't show how an output of a function would make a copy of the original function. It could be an input for another function, trigger a loop, or cause a chain reaction. But for me, it's hard to imagine how a function would directly produce an imperfect copy of itself.
On the other hand, you also hint at the alternative approach (2) in the book:
This raises an interesting possibility: Could early function replication have been driven by parsing and mimicking actions rather than directly copying functions?
If so, this aligns with the hypothesis of behavior parsing and memetic evolution. It’s conceivable that our ancestors evolved self-replication mechanisms in the same way that animals developed meme-parsing abilities—to improve survival at runtime relative to competing gene-machines.
Perhaps the earliest function replication was not self-replication but rather an attempt by another part of the body or mind to parse and reverse-engineer an action into a new function. If so, this could imply that function replication initially emerged as a function-copying function rather than a self-replicating function. Perhaps this is part of the meta-algorithm itself?