15 things I learned from my coversation with David Deutsch on Fun, Knowledge, and the Self

Most of my conversation with David Deutsch focused on questions I'd been wrestling with myself: points I didn't fully understand, or wanted to check that I’d interpreted correctly. Many of these relate to the Fun Criterion and the essay I wrote on it.

The discussion clarified a lot for me. Below I’ve written out the key things I learned, point by point.

You can find the full transcript here and the podcast on YouTube here.

The list grew longer than expected, so I’ve first included a short description of each item, followed by the detailed explanations.

Insights on the Fun Criterion

1. Lack of fun is a criticism — a moment to reflect, not a hard stop
2. Interest and playfulness can be proxies, but they are fallible and unreliable
3. "Follow the fun" means interpreting boredom and staleness as criticisms
4. A neutral state is not optimal — fun still involves active engagement and enjoyment
5. A conflict with unknown cause is usually coercive — a sign of something going wrong
6. Unconscious conjecture and criticism happen in many everyday problems
7. Fun is only relevant when desires conflict — not for mechanical choices
8. "Alignment" between the creative strands means not hindering what's already active, not proactively seeking it

Insights on the Self

9. The self is not just an arbiter — it's a description of the process that aims for unanimity
10. Creative institutions are process-level knowledge that governs how ideas interact

Insights on Knowledge and Meaning

11. The three types of knowledge are useful approximations, not hard categories
12. Non-explanatory knowledge is what we mechanically recall outside the context of a problem
13. Meaning must ground in inexplicit knowledge — something beyond pure verbal definition
14. Sensations vs. emotions may differ only in how difficult they are to adapt
15. We don't fully understand biological evolution — DNA may encode for something more than genes, and there may be more selective pressure via preferred storage media

Insights on The Fun Criterion

Insights 1–8 refine and extend my understanding of the Fun Criterion: what it means, how it works, and where my previous essay was slightly off.

1. Lack of fun is a criticism — a moment to reflect, not a hard stop

My previous understanding: I already saw the Fun Criterion as a form of criticism, but my earlier interpretation missed the fallibilist nuance. A lack of fun is itself an interpretation, and like all interpretations, it can be mistaken. Not having fun isn't an automatic stop signal; it's a prompt to pause and reflect.

New understanding: Deutsch frames lack of fun as a criticism in the Popperian sense. It's a moment to reflect, not a command to stop. The criticism itself could be wrong; you might actually be fine. As long as you don't think you have a problem, you don't need to address it. If you do have a problem, the response is to conjecture how to change it and to question whether the problem is real.

"So following Karl Popper, I advocate regarding lack of fun or apparent lack of fun or apparent lack of progress as a criticism rather than a hard stop to what we're doing. As long as we don't have a problem we don't have to answer questions like that."

The improvement: From treating lack of fun as an alarm to treating it as a fallible criticism: something to investigate, not automatically act on.

2. Interest and playfulness *can* be proxies, but they are fallible and unreliable

My previous understanding: In my Fun Criterion essay, I examined whether play, curiosity, and interest could serve as proxies for the state of Fun. I was cautiously receptive to the idea, but hesitant because it can easily be misread as blindly chasing feelings, which are only correlates of being in a state of fun, not reliable indicators by themselves.

New understanding: Deutsch confirms that proxies and correlates exist (e.g., not being in pain, not being bored) but they are unreliable in particular cases. A marathon runner is in pain while having fun. The point is not to follow any external criterion like an emotion or state, but to reflect on whether you are solving the problem you care about.

"Yes, but proxies and correlates are no good in particular cases. [...] Meeting the criterion is a description of the process. It's not something you should be thinking about explicitly when deciding between ideas."

The improvement: Proxies exist but should never be treated as the criterion itself. The focus should remain on whether you're solving the problem you care about.

3. "Follow the fun" means interpreting boredom and staleness as criticisms

My previous understanding: I treated "follow the fun" as a somewhat loose expression Deutsch occasionally uses, which a high potential for misinterpretation. In my essay I was adamant to distinguish it from "follow enjoyment."

New understanding: In the chess example (Ginger GM), the mechanism becomes clearer. "Follow the fun" means that when you have multiple options and conflicting desires, you choose the one where you're most excited (least conflicted). It's the absence of criticisms that constitutes fun in this case: boredom and staleness are criticisms of the "safe" option, and excitement/beauty are the absence of those criticisms for the other option.

"So-and-so is probably the best move but it's boring. So on the other hand this other move might make me lose, but it'll lead to an interesting game."

This links to my earlier essay section on the chess example and the role of proxies.

The improvement: "Follow the fun" is best understood as: interpret your feelings of boredom, staleness, or resistance as criticisms of your options. Choose the path with the fewest unresolved criticisms, which often manifests as the path you're most excited about.

4. A neutral state is not optimal — fun still involves active engagement and enjoyment

My previous understanding: In my essay, I pushed back against the idea that positive emotions could be used to identify Fun. I drew a sharp distinction between Fun-as-unobstructed-state and enjoyment: "Fun cannot be felt directly, and therefore any interpretation of the FC as 'doing what feels fun' is a mistake." Based on this premise, I claimed that emotional neutrality (no thwarting, but no particular enjoyment) still qualifies as Fun.

New understanding: Deutsch pushes back on this:

"No forcing going on is, if you like, a necessary but not a sufficient condition."

So while the absence of coercion remains the foundation, Fun requires something more. People who look forward to shopping, enjoy it, and improve on their experience each time are in a better state than someone who is merely not coerced:

"I think such a state is not optimal. I think that there are people who look forward to going shopping and enjoy themselves while they're shopping and shop in a different way each time, so that each time they're improving on their experience of last time. And that's a better way of shopping than of being emotionally neutral."

This doesn't contradict my claim that blindly chasing positive emotions is a mistake. But it does suggest that Fun is more than just the absence of coercion; it involves active creative engagement with whatever you're doing. A neutral state may not be coercive, but it's also not the state where your creative processes are fully alive and interacting. When all your creative processes are genuinely engaged (explicit, inexplicit, unconscious), the result is naturally accompanied by some form of enjoyment or interest. Not because you're chasing feelings, but because active creative engagement produces them.

This is akin to the experience Csikszentmihalyi describes in flow states, where deep absorption in a task produces a sense of ecstasy and clarity, even though the person isn't consciously monitoring their emotions (Csikszentmihalyi, 2004 TED Talk).

The improvement: Fun is not merely the absence of coercion; it requires active creative engagement. Neutrality, while not coercive, is sub-optimal because it signals that your creative processes aren't fully engaged. My previous framing was too defensive against positive emotions. The correction isn't to follow emotions, but to recognize that creative engagement naturally generates them.

5. A conflict with unknown cause is usually coercive — a sign of something going wrong

My previous understanding: I understood that thwarting is the core bad thing, and that conflict itself is normal. But I hadn't thought carefully about what distinguishes a healthy conflict from a problematic one.

New understanding: Deutsch says conflict where you don't understand what's in conflict, where it feels bad and you don't know why, is a clue that something coercive is going on. The different parts of your mind are using their creativity against each other. This is distinct from ordinary problem-solving conflict.

"Conflict can be a very good thing, but again, conflict where you don't understand what's in conflict and it feels bad and you don't understand why and that sort of thing is a clue that there is something coercive going on."

The key distinction isn't between fully understood conflict and mysterious conflict. In practice, we rarely know the exact cause of an inner conflict. Rather, it's between having some guess or sense of what's going on versus having no idea at all. If you're nervous before a presentation and can roughly conjecture why (say, you're underprepared, or you care about the audience's opinion), that's a normal conflict you can work with. But if something just feels wrong and you can't even begin to approximate what's behind it, that's the warning sign.

He also clarifies: if it doesn't seem problematic to you, assume nothing bad is happening; otherwise you'd just be making things up to frighten yourself.

The improvement: The diagnostic for coercion isn't whether you fully understand a conflict, but whether you have any purchase on it at all. Having a rough sense of what's in tension is normal problem-solving. Having no idea, just an unexplained bad feeling, is a clue that something coercive is going on

6. Unconscious conjecture and criticism happen in many everyday problems

My previous understanding: I understood that unconscious conjecture and criticism must exist, but I didn't have many concrete pictures of what that looks like. The main example I could think of was absorbing a language through immersion, actively learning but without conscious awareness of the specific knowledge being created.

New understanding: Deutsch gives a mundane but clear example. You go to scratch your head. Your shirt is caught on a thread of the armchair. You can't lift your arm. You pull hard, break the thread, scratch your head, and never know any of it happened:

"If I was going to scratch my head and I found that I couldn't reach, then I would think that there's something wrong. [...] I might just pull on the thread and break the thread and then carry on without really knowing that I've ever done any of that."

This is a full cycle of unconscious problem-solving: the problem, the conjecture, the criticism, the solution, all without entering conscious awareness. He also gives the example of stepping over a pavement obstacle without the words "step" or "obstacle" ever entering your mind, though that one is inexplicit rather than unconscious, since you could recall it if asked.

The improvement: From having only one concrete example of unconscious creativity (immersion) to seeing it at work in everyday problems: real-time conjecture, criticism, and resolution happening entirely below awareness.

7. Fun is only relevant when desires conflict — not for mechanical choices

My previous understanding: I understood fun as relevant to all decision-making broadly.

New understanding: Deutsch makes a distinction. When you're evaluating options by an explicit, objective function (e.g., which item fits my budget?), fun doesn't come into it at that level; the question is whether the whole activity is fun, not the specific choice. Fun becomes relevant when there are multiple non-quantifiable, conflicting impulses: desire to win, desire not to be bored, desire to look good, etc.

"If choosing which one it is just involves evaluating, say, an objective function [...] then fun doesn't really come into it at that level. [...] So the place where it becomes relevant is when there are different kinds of conflicting impulses in your mind."

My own conjecture here is that this points to a multi-level structure in how fun operates. Consider buying a laptop within a given budget. Comparing specs against price is a subordinate problem, largely mechanical, solvable by explicit criteria. Fun doesn't come into it at that level. But that subordinate problem sits inside a parent problem: Do I need a laptop? What is my budget? At that parent level, you're juggling conflicting desires: financial prudence, excitement about new tools, guilt about spending, a sense that your current laptop is "good enough." None of those are fully quantifiable, and that's where fun becomes relevant. So the same activity can be simultaneously mechanical at one level and deeply entangled with fun at another. The subordinate decision is resolved by criteria; the parent decision is resolved by taking all parts of the mind seriously.

The improvement: Fun is specifically about resolving conflicts between incommensurable desires, not about all decisions. When there's a clear objective criterion, that handles the choice; fun operates in the space where no such criterion exists. But most real-life decisions are nested: mechanical sub-decisions sit inside larger, messier parent problems where fun is very much at play.

8. "Alignment" between the creative strands means not hindering what's already active, not proactively seeking it

My previous understanding: In my essay, I puzzled over what Deutsch means by explicit, inexplicit, and unconscious ideas "influencing one another and being in alignment." I wrote: "I still see the core as simply acknowledging and accounting for all the types of knowledge currently in play, not necessarily requiring each to participate or deliberately seeking them out."

New understanding: Deutsch's clarification is that by default all three types of knowledge are already always active. There is no need to seek them out, because they're already there:

"Even in something like chess or mathematics, which line of thought to follow, which is the one that is best for you to follow, already involves inexplicit and unconscious thoughts and feelings. All explicit thinking contains an inexplicit component."

So the inexplicit and unconscious aren't optional extras you invite to the table; they're already at the table for every thought you have. The question isn't whether to engage them, but whether you're hindering them. Deutsch illustrates this through the grammar example: every word you utter has undergone unconscious conjecture, unconscious criticism, and unconscious explanation, without you knowing it. These processes are always running:

"There will have been unconscious conjecture, unconscious criticism, unconscious explanation, all going in to produce that word 'very' or 'extremely' and every single word you say has undergone that process."

"Alignment" means these processes are running unobstructed, not that you must deliberately activate them. When nothing is being hindered, the different strands of thinking naturally help each other: a change in your explicit understanding cascades into your inexplicit behavior and vice versa.

The improvement: My previous understanding was that alignment you must actively persue: gather all three types and make them interact. Deutsch's point is the opposite: all three are always already active. Alignment is the default state when nothing is being hindered. Fun isn't about proactively seeking out inexplicit and unconscious input; it's about not suppressing what's already there.

Insights on the Self

Insights 9–10 are about how the mind is organized: the self, creative institutions, the types of knowledge, and why no thought is ever purely mechanical.

9. The self is not just an arbiter — it's a description of the process that aims for unanimity

My previous understanding: I thought of the self as something like an arbiter or coordinator, an "I" that sits above the conflicting parts of the mind and makes the final call. A kind of internal judge that mediates between competing ideas.

New understanding: Deutsch acknowledges that the self does form judgments; it's not a purely passive label. When you think "that isn't me," you're making a normative claim: that shouldn't be me.

"The self is not just a neutral coordinator, it also forms judgements. So when you're thinking, 'that isn't me,' then you're forming a judgement. It really means that that shouldn't be me."

But he also adds that this shouldn't lead us to think of the self as a thing, an entity living somewhere in the brain that gets called upon when there's a conflict, like a policeman. Rather, it's a description of what happens when problem-solving leads to unanimity:

"I also don't want to think of the 'I' as being a thing that lives in a particular part of the brain and it gets invoked when there's a conflict like a policeman. It's just a descriptive thing. Conflict resolution, problem solving happens and the aspect of problem solving that leads to unanimity or that aims for unanimity we call the self."

That makes sense: the judgment-forming and the process are the same thing. In a rational mind, conflicts get resolved and at the resolution there's only one idea left:

"In a rational mind, there is a single self in that conflicts get resolved and when they get resolved then at the resolution there's only one idea left so that there are no chronic hangovers of the rival theories left at the time when one acts."

But this is specifically the picture of a rational mind. Deutsch notes that people occassionally report the opposite: conflicting voices that won't go away, ideas that persist even when other parts of the mind would like them gone. That's a sign of something irrational or anti-rational going on.

"Some people report that there can be more than one and they're both yelling at each other and so on. Note that we're universal so that we could host several selves. If we couldn't do that, we wouldn't be universal. But it's normally — usually, I have to say again, usually — having several selves which can't resolve their differences is a sign of something going wrong."

The improvement: From thinking of the self as an arbiter above the process to understanding it as a description of the process. The self isn't a judge sitting in a courtroom; it is the process by which the courtroom reaches a verdict. It judges, but it's not a substance or location. In a rational mind, this process converges on unanimity, a single self. In an irrational mind, it fragments.

10. Creative institutions are process-level knowledge that governs how ideas interact

My previous understanding: I understood Deutsch's tweet about the self as "creative institutions of consent" but wasn't sure exactly what "institution" meant in this context.

New understanding: Deutsch explains (via Popper's usage): an institution is a kind of idea, theory, habit, or practice that governs how other ideas interact with each other in the mind. Creative institutions are ideas whose job is to manipulate other ideas and try to correct impediments to the growth of knowledge. They include modes of thinking, modes of conjecture, modes of criticism, and criteria. Probably the most important ones are inexplicit.

"What Popper means by an institution is a kind of idea [...] which governs the way that other ideas within the mind interact with each other."

The scientific mode of thinking is a key example: it's not inborn, it helps avoid errors and conflicts, and it can be internalized (made inexplicit and unconscious after starting as conscious).

The improvement: "Creative institutions" are the knowledge structures that facilitate creative processes. They're not content-level knowledge but process-level knowledge: the rules of engagement for how ideas interact.

Knowledge and Meaning

Insights 11–15 are about the nature of knowledge itself: what makes it explanatory, where meaning grounds out, and the spectrum between sensations and emotions.

11. The three types of knowledge are useful approximations, not hard categories

My previous understanding: I use a three-part distinction (explicit, inexplicit, and unconscious knowledge (in my own terms: Statements, Intuitions, and Drives)) as a fairly firm working taxonomy for thinking about the mind.

New understanding: Deutsch explicitly cautions that this isn't a rigid taxonomy. It's an approximation, comparable in precision to Freud's ego/id/superego, useful for reasoning but not describing real boundaries. We don't know if consciousness is a hard border; it's more approximate and descriptive based on what we can interpret rather than quantized units.

"This isn't a rigid taxonomy. This is just in a very approximate way of trying to understand what's happening in the mind."

The improvement: The three types of knowledge are useful approximations and descriptions, not hard categories with firm boundaries. Being overly rigid about them may actually hinder understanding.

12. Non-explanatory knowledge is what we mechanically recall outside the context of a problem

My previous understanding: I understood the distinction between explanatory and non-explanatory knowledge broadly, but hadn't placed it clearly in the context of the mind and problem-solving.

New understanding: Deutsch distinguishes: some ideas in our mind are non-explanatory (e.g., remembering what you had for breakfast, the year of the Battle of Hastings). The memory itself may not be explanatory. But as soon as it enters into a problem, it becomes the subject of conjecture and criticism, and thereby becomes explanatory.

"The memory may not be explanatory but as soon as it enters into a problem it will become explanatory. It'll become the subject of conjecture and criticism with the intention of solving a problem."

This is interesting because conjecture and criticism are themselves explanatory processes, so our manipulation of ideas always involves explanation, even when the ideas themselves are non-explanatory.

The improvement: Knowledge can sit in the mind as non-explanatory (raw memories, facts). It becomes explanatory the moment it's recruited into problem-solving. The distinction is contextual, not inherent to the knowledge itself.

13. Meaning must ground in inexplicit knowledge — something beyond pure verbal definition

My previous understanding: I understood that explicit knowledge is rooted in inexplicit knowledge from Deutsch's earlier conversations. But I hadn't fully grasped why it must be inexplicit specifically.

New understanding: Deutsch clarifies this:

"Explicit knowledge is expressed in words, but the meaning of the words is not expressed entirely in words. If it were [...] that would be an infinite regress. Dictionaries could never define anything."

This is because interpretation itself demands that meaning eventually points to something beyond words, to inexplicit knowledge: theories, skills, and interpretive frameworks we use but haven't articulated. You can't make everything explicit at once because making something explicit requires using other knowledge as background, and that background knowledge is (at that moment) inexplicit.

Even speech itself illustrates this. As Deutsch notes, when we convey something in words there's always a component that isn't in the words:

"When we do convey it in words there's still a component that isn't in words."

Ironically, I was reminded of this while reviewing the podcast transcript itself. Spoken language carries far more inexplicit knowledge than text written directly: emphasis, rhythm, sentence structure, pauses, and tone all convey meaning that the words alone don't capture. The explicit content is the same, but the inexplicit layer is richer in speech. Which is itself evidence of how much meaning lives beyond the verbal.

This is also why dictionaries work: they help you connect new words to knowledge you already (inexplicitly) have, rather than creating meaning from nothing. And it's why Deutsch sees a fundamental problem with LLMs: they lack this root-level contact with reality:

"That again is a problem with LLMs because they don't have a root level of contact with reality. They only have words and sentences in terms of other words and sentences."

The improvement: The grounding problem of meaning resolves into the necessity of inexplicit knowledge, not as an arbitrary fact, but because the logical structure of interpretation demands it. All interpretation bottoms out in knowledge that is felt, practiced, or experienced, not stated in words.

14. Sensations vs. emotions may differ only in how difficult they are to adapt

My previous understanding: I had some uncertainty about whether sensations and emotions are meaningfully different categories, and if so, how we should treat them differently when it comes to the mind.

New understanding: Based on this conversation, my working theory is as follows:

Both sensations and emotions are downstream of knowledge: they are signals produced by underlying knowledge. We can change the signals by changing the underlying knowledge, but we can also change our interpretation of the signals. How difficult either of these is depends on how deeply ingrained the knowledge is. Innate knowledge (which tends to produce what we call physical sensations) is harder to change than constructed knowledge (which tends to produce what we call emotions). So the meaningful difference between sensations and emotions may not be one of kind, but of degree: how much creative effort is required to modify them.

Deutsch's observations seem to support this framing. He notes that both require creativity to modify, but that sensations demand far more of it:

"Manipulating physical sensations to make yourself say not feel pain when an external stimulus would normally give you pain — first of all, it's very difficult. One can't just decide to do it. One has to exert a huge amount of creativity in order to do that."

But there are multiple ways to modify what we experience. We can change the underlying knowledge, we can change the interpretation (since sensations must be interpreted before they're experienced), or we can skip the experience entirely. Deutsch gives examples of each: training yourself to acquire perfect pitch changes how tones are interpreted, learning to enjoy the pain of a marathon changes whether you enjoy the sensation without altering what it feels like, and soldiers in battle sometimes don't notice losing a limb until the battle ends: the signal is produced but never reaches consciousness:

"There are stories of soldiers in battle who get shot or have their arm blown off and so on and they don't notice until the end of the battle."

So even deeply innate sensations aren't as fixed as they appear: they can be reinterpreted, reframed, or not experienced at all.

The improvement: Sensations and emotions aren't two fundamentally different categories but sit on a spectrum of how deeply ingrained the underlying knowledge is. Both can be modified, either by changing the knowledge that produces them, by changing our interpretation of them, or even by not experiencing them at all. The difference is in the amount of creative effort required: innate knowledge (sensations) is harder to change than constructed knowledge (emotions), but neither is fixed. Note that this is my own conjecture based on the conversation, not Deutsch’s own conclusion.

15. We don't fully understand biological evolution — DNA may encode for something more than genes, and there may be more selective pressure via preferred storage media

My previous understanding: I knew we couldn't fully simulate biological evolution computationally, and that evolutionary algorithms only produce optimization, not genuine novelty.

New understanding: Deutsch raises a specific and interesting possibility: searching randomly in DNA space is not the same as searching randomly in protein space. The DNA as a storage medium already constrains and structures the space of possible codes. It might have properties "over and above coding for genes" that we don't understand. All methods of information storage privilege certain types of information over others; DNA presumably does too.

"The DNA, searching randomly in DNA space is different from searching randomly in protein space [...] it might be that the DNA has got some property that we don't understand over and above coding for genes."

David also rightly corrects my use of the word "targeted": whatever is going on, it evolved itself. It's not directed in a supernatural sense but the search space is more structured than "purely random."

Sidenote: Tara Stahl and I noted that this resonates with Michael Levin's work on developmental bioelectricity. Levin's lab has shown that bioelectrical patterns among cells carry patterning information that isn't directly specified by DNA, serving as a non-genetic code that stores anatomical target states and can be rewritten independently of the genome (Levin & Martyniuk, 2018). This is a concrete example of biological information storage beyond DNA-to-protein coding, which is consistent with Deutsch's suspicion that there's more structure in biological information systems than we currently understand.

The improvement: The gap in our understanding of biological evolution isn't just about the search algorithm; it's about the storage medium itself shaping what kinds of solutions are reachable. DNA as information substrate may carry more structure than we appreciate.