Motivation for post: As a former CFAR instructor, longtime teacher, and rationality pundit, I find myself giving lots of advice in lots of different contexts.  I also try to check in from time to time to find out which bits of advice actually proved helpful to people.  Over the years, I've heard from a genuinely surprising number of people that my (offhand, very basic, not especially insightful) thoughts on "shoulder advisors" were quite useful to them, and remained useful over time.  So: a primer.

"There's a copy of me inside your head?" Hermione asked.

"Of course there is!" Harry said. The boy suddenly looked a bit more vulnerable. "You mean there isn't a copy of me living in your head?"

There was, she realized; and not only that, it talked in Harry's exact voice.

"It's rather unnerving now that I think about it," said Hermione. "I do have a copy of you living in my head. It's talking to me right now using your voice, arguing how this is perfectly normal."

"Good," Harry said seriously. "I mean, I don't see how people could be friends without that."

The term "shoulder advisor" comes from the cartoon trope of a character attempting to make a decision while a tiny angel whispers in one ear and a tiny devil whispers in the other.

Many people have multiple shoulder advisors.  Some, no doubt, carry a literal metaphorical angel and devil around with them.  Others may sometimes hear the whispers of some of their favorite beloved fictional characters.  It's quite common in my experience for people to have shoulder copies of their parents, or their best friends, or their romantic partners, or particularly impactful teachers or bosses or mentors.

This is not schizophrenia (though for all I know it may use some of the same hardware, or may be a low-key, non-pathological version of schizophrenia in the same way that a healthy self-preservation instinct could be thought of as a low-key, non-pathological version of a phobia or an anxiety disorder). 

Rather, there is simply some kind of subroutine in the brain of most humans that is capable of taking in training data and learning what a given person (or character, or archetype) would say, in a given situation. It's predictive software, likely evolved in response to the need to model other chimps in the ancestral environment, and strongly selected for due to the fact that being able to model those other chimps accurately generally paid off big.

It's important to be clear that the experience of "hearing the voices" actually happens, in many people. This is not a metaphor, and it is not hyperbole or exaggeration. I'm not saying that people tend to hallucinate actual sounds—that probably would be schizophrenia.  But in the same way that most people "hear" their own thoughts, people also "hear" the voice of their dad (or "see" his facial expression), offering thoughts or advice or reacting in real time to the current situation.

"I was going to complain about having to type with my thumbs to text you, and how I'd rather just use email or Slack, but my shoulder Malo popped up to say 'Duncan, you have a Mac. Just use Messages with your keyboard.'"

"My mental copy of Jack is currently freaking out a bit about how toxic and unhealthy this sounds."

"I notice my inner Nate is betting this project will fail."

"I can hear my mom reminding me to take jam tarts when jam tarts are offered."

(Note that you don't need to "demand" that your advisor communicate in words! Often it's both easier and also just as useful to simply let them be present—to "see" their facial expressions and body language, imagine their nonverbal reactions, let yourself be aware of and attentive to them in the same way that you (likely) are aware of or attentive to other actual humans in the same room as you.  Think of how, for instance, someone at a party might say something that causes your eyes to dart over to a friend, to see their reaction—you can do the same thing with your simulated friend.)

If you already have this experience: you can curate and improve your council of shoulder advisors, and this post will give you some pointers on how.  If you do not already have this experience: you can most likely learn how to, if you want, and even a weak or limited or unreliable version of the skill has proven valuable for people.

Why would I want this?

In essence: good shoulder advisors allow you to be (at least marginally) smarter and more creative than you-by-yourself are capable of being.

I don't have a rigorous or technically valid explanation as to why, but it is a straightforwardly observable fact that, for many people, their shoulder advisors occasionally offer thoughts and insights that the people literally would not have thought of, otherwise.  Novel ideas, useful perspective shifts, apt criticisms of one's own actions or intentions, that sort of thing.  It's generally well-understood that "two heads are better than one," especially in times when one is stuck or uncertain, and shoulder advisors can be genuinely almost as good.

("One-point-seven heads are better than one.")

Having the right shoulder advisor "show up" at the right moment can be every bit as impactful as having an actual friend or mentor in the room.  And since shoulder advisors take up zero space and can be called upon at any hour and can include people you could never actually call upon in real life (such as Master Yoda or President Obama or Dwight K. Shrute or Mister Rogers or any number of Lannisters), even small improvements in:

  • Your ability to summon them at all
  • Their richness and overall verisimilitude

... can be tremendously valuable.  My own cast of shoulder advisors have:

  • Helped me overcome fear of physical actions I was capable of safely performing (backflips, broad jumps at height)
  • Helped me make rapid mood shifts (e.g. yanked me off the path of "I'm about to lose my temper" and restored my perspective and calm)
  • Headed off large failure modes in important projects before they cropped up (e.g. pointed out a thing that would go disastrously wrong under the current plan)
  • Made genuinely useful suggestions about how to phrase comments in difficult conversations (with employers, with romantic partners, with struggling friends)
  • Noticed things that I had not consciously noticed (because it was the type of thing that person tends to care about and pay attention to, and I noticed my mental copy of them noticing)
  • Provided advice for other people who were seeking advice from me (that I was incapable of producing directly, out of my own experience)
  • Provided genuinely meaningful amounts of emotional comfort and support at times when I was isolated from my friends and family
  • Proposed multiple ideas for projects and essays and gotten me "unstuck" on both personal and professional projects
  • Generally served as a stabilizer that helps me stay within the range of what "feels like me," i.e. they give me funny looks or helpful nudges when I start acting uncharacteristically or in ways that don't accord with my vision of my ideal self.

... not to mention that having robust copies of my actual friends and colleagues has much better equipped me to interact with those friends and colleagues, by giving me a head-start on how they'll respond to any number of things.

Selection criteria: emulability and usefulness

Step one, acquire shoulder advisors.  Step two, use them skillfully.

This section is for step one.  In order to use shoulder advisors, you have to have shoulder advisors, and whether you're building up a whole shoulder council for the first time or just trying to expand and curate an existing ensemble, some appointees are going to prove much more valuable than others.

Assume you had no preexisting council, and were brainstorming a list of possible advisors with the intent to winnow it down.  You might try writing down four or five names for each of the following categories:

  • Close family members (whether they're still close or not)
  • Longtime friends (whether you're still friends or not)
  • Impactful teachers and mentors
  • Current bosses, employers, coworkers, or clients
  • Characters from TV shows and movies
  • Characters from books or other media (including those you've invented yourself)
  • Politicians, comedians, authors, celebrities, and other notable public figures
  • People who've blown your mind or changed the way you look at the world
  • People you have had serious disagreements with

Once in possession of a list of ~40 names, I claim the next step is to filter it based on the presence of two qualities: emulability and usefulness.

Emulability is the degree to which your brain can, or could likely learn to, successfully boot up a copy of this person and "just push play" on it, such that the copy in a sense "runs itself."  Authors sometimes talk about their characters "coming to life," and producing their own dialogue or wresting the story in an unexpected direction or even verbally arguing with the author inside their head—this is high emulability.  You want the sense that you're not making up or imagining what the person would say, via an act of explicit concentration, but rather that it's just auto-completing in the same way that a catch phrase or advertising slogan auto-completes.

In practice, emulability is often immediately obvious; you can just pluck a name off the list, imagine them sitting beside you (or reading over your shoulder, or lounging on the other side of the room) and just see how they react to what's happening to you right this second, and the claims that they hear me making.

(This is what happens to Hermione above, as soon as she bothers to check.  If attempting to bring someone into your current physical surroundings doesn't work, you could also try imagining specific scenarios, like throwing a water balloon at someone or showing up late to a thing, and see if your shoulder candidate has a characteristic response.)

In the event that this kind of imagination is not yet easy for you, though, there are a couple of qualities you can use to assess the emulation potential of a given shoulder-person, before putting in a bunch of effort.

The first of these is total training data.  People you've interacted with 100x more than average will tend to be more emulable just because you've absorbed more instances of "X happened, and they responded with Y."

(Note that as far as your brain is concerned, it makes zero difference whether the person under observation is real or fictional.  I've seen more of Miles Vorkosigan's reactions to a wide variety of stimuli than I have of many of my actual coworkers.)

The second major component is something like uniqueness or quirkiness or internal consistency.  If someone has a very specific vibe, it's easy to vividly imagine their particular responses.  Ditto if someone has strong opinions, or narrow special interests.

("I saw this video of a rocket launch and immediately thought of you, but then I got this mental image of your face looking very unimpressed, actually, and I genuinely wasn't sure why.  What does real-you have to say?")

Boring(-to-you), quiet, unopinionated, and "normal" people are thus quite hard to emulate, but that's okay because even if you could emulate them, you wouldn't get much out of them most of the time.  You're looking for the kind of people who have the potential to change your course—to think of things you wouldn't, make suggestions that aren't obvious, say the things you need to hear.

Which brings us to our second major filter: usefulness.

When I ran through the brainstorming list above, pretending that I'd never had any shoulder advisors at all, I got about 40 names, and when I filtered for emulability, I had maybe a dozen left.

Predictably, on that list were "Mom," "Dad," and "Ender Wiggin."  But if I were actually creating a council of shoulder advisors from scratch, I wouldn't necessarily want Mom or Dad or Ender to be on it.  I grew up with all three of those people having a deep influence on me—their perspectives and philosophies are already largely baked into "my whole deal," and not the sort of thing I need help keeping in the forefront.

Similarly, I don't really need more Tyler Durden or Mad-Eye Moody; I think I'm doing pretty okay on cantankerous pessimism and niche charisma.

Instead, a far more interesting person to have on my shoulder is one who can remind me of virtues I don't have down pat.  One who can snap me out of my normal patterns, cause me to smack my own forehead and mutter a rueful "of course."

For me, that list looked more like my friend Matthew from high school, who is soft-spoken and charitable and the-sort-of-Christian-the-Jesus-depicted-in-the-Bible-would-actually-like, and Jean-Luc Picard of the starship Enterprise, and an old colleague from Seattle, and the comedian Dave Chappelle.  These people were not only emulable but also truly different from me, which meant that if I could successfully add them to my shoulders, they would have the potential to catch things my regular algorithms would miss.

By "usefulness," then, what I am trying to gesture at is "I suspect my life would benefit from small, well-timed injections of this person's way-of-being."  If you are (according to yourself) too timid and hesitant, then you might look for people who are avatars of boldness, or who tend to be encouraging and supportive and make you feel confident, or who are eccentric and surprising.  If you are (according to yourself) too reckless and unreliable, then you might benefit from shoulder advisors who are avatars of caution, or who tend to pipe up with nervous hesitations, or who are good at noticing the little details before they turn into big problems.

(And if you don't know what your flaws are, or how best to go about improving yourself according-to-your-own-values, then maybe you're looking for people who are generally insightful and clear, or who are good at turning uncertainty into concrete and actionable suggestions, or who are (perhaps) somewhat scathing and unafraid to utter harsh truths.)

Improving the effectiveness of the council

Taking as given that you have some number of shoulder advisors who are either active or who you intend to start consulting, what next?

The key value of a good shoulder advisor is that they say the thing you need to hear, at the moment you need to hear it.  It doesn't take much to tip a tough decision from one direction to the other, or to start (or break) an affective spiral or chain of if-then behaviors.  A shoulder advisor is a specific instantiation of the general wish "if only I'd thought of X before Y happened"—you're trying to make it more likely that you will, in fact, remember X, especially where X is something not particularly native to your current way of doing things.

Taking the second part first, there are two ways to make sure that you hear from your shoulder advisors at the critical moment:

  • Build the habit of making an explicit, effortful check; pause and actively boot up your shoulder advisor in response to various triggers, e.g.:
    • You're about to make a major decision
    • You're noticing a strong feeling of temptation
    • You're noticing a strong feeling of certainty
    • You just said a bunch of hateful things about yourself
    • You've just made some kind of absolute declaration
    • You're considering changing the plan (or sticking to a plan you feel an impulse to change)
  • "Teach" your shoulder advisors to appear on their own

... there's a little bit of magic in both of these; I'm more telling you where to put your effort and not how that effort should look.  A full attempt to lay out how to build habits-of-mind goes beyond the limits of this introductory primer.

By far, though, it's the second strategy that I and others have found disproportionately impactful.  Explicit, intentional checks can only ever cover a small fraction of the times when people could really use a little extra insight.

However, doing the explicit thing is a good way to bootstrap to the automatic version, especially if you set aside five minutes to do a one-time brainstorm on "when do I wish my shoulder advisors would show up?"  Note that you can make a limited commitment, and that almost any amount of explicit practice will pay off, on the margin—if it sounds like too much to do five checks a day for three months, try doing one check per day for one week (or whatever).

(As with exercise, the best plan is one you'll actually follow through on, not one which sounds virtuous and doesn't work out.  Also, for the record, that line was literally just delivered to me by my shoulder Eli Tyre.)

A couple of tips, as you explore this space:

  • Don't ask your shoulder advisors questions.  Just like people tend to get better results from telling themselves "it went wrong" and then letting their brain tell the story of why (rather than asking themselves "what might go wrong?"), it's better to just imagine the person in the room with you—imagine them hearing the previous minute of conversation, or visualize them sitting over in the corner, watching and forming opinions, and just sort of let them say their piece.  This can be a difficult skill to learn, if you don't have experience with it, but be patient—if your shoulder advisor isn't speaking up or making faces or anything, just keep on imagining them as you think thoughts at yourself or review your plan or whatever.
  • Also, don't just summon your shoulder advisors to weigh in on Big Issues, especially if you're practicing.  Vary the triggers, and reward your brain for causing the shoulder advisor to show up at all, for whatever reason, even if it's while you're making breakfast or while you're in the shower or just to say something snarky about the person in front of you in line.  Like in (some forms of) meditation, where you don't stop your thoughts from wandering, but rather practice always returning your focus to where you want it, you'll get better results if you think in terms of "how much practice are my shoulder advisors getting at booting up from nothing?"

Once you've got a cast of characters who are willing to show up at all (or at least one solid imaginary friend), then you can worry about nudging their contributions in an actually useful direction.

My favorite techniquelet here is to refer back to the source material.  It's amazing how quickly the human brain will update its model of another human, if you actually go back and check.

"Hey, Nate, I was wrestling with [decision] yesterday, and my shoulder Nate thought the key consideration was [blah]."

"Lol.  I mean, yeah, but actually there's a much more important consideration, which is [blah]."

(Yes, my shoulder Nate actually says the word "lol" out loud, like it rhymes with "doll."  He does this because the real Nate does this, and my brain recorded it.)

If at all possible (especially in the early days), get your real advisors to not only correct your shoulder advisor's core thoughts and ideas, but to flesh out why they think what they think, and where your shoulder copy went wrong/what it doesn't seem to understand.

If your shoulder advisor is fictional, this is somewhat harder to do, but a good substitute is to write down a draft of their first contribution, then review it a day or two later with a critical eye.  Even moreso than copies of real people, your fictional shoulder advisors are free to mutate in whatever direction is useful for you.

(One thing I've had fun with is pitting them against each other—not by simulating an argument directly, where I imagine two sides of a debate, but rather by having both of them fight to convince me, or by having each of them arguing their conflicting judgment of the situation.  Having an optimist and a naysayer is a pretty good dynamic, and it's not hard for most human brains to pattern-match what each of those would say next, to the other.)

Ultimately, the idea is to give regular feedback to whatever part of your brain is running the emulation.  Upvotes for what works and feels true, downvotes for what doesn't, but most importantly, more training data.  It's fine if your shoulder advisor gets frustrated and impatient as you ask it to say more and more words—let it be frustrated and impatient in whatever way is characteristic for that individual, and just keep recording.

Downloading yourself

Again, the above was more a set of trailheads or threads-to-pull; there's not really a standard canon of advice here yet.  Hopefully, it's enough to get people started (and hopefully readers will leave further tips and advice in the comments).

There was one last piece of the overall picture that I wanted to touch on, at least briefly, and it's this:

You, too, can be a shoulder advisor.

My friend Nate and I both live in each other's heads, and we both furthermore have a vested interest in our mental clone copy.  Nate wants my shoulder Nate to be as good of a Nate copy as it can be; I want the same for his mental Duncan.  In part, this is for weird TDT-esque considerations, but mostly, it's just because I like my friend Nate, and he's my friend at least in part because of the impact I have on him, and if he's got a copy of me on his shoulder I can go on having that impact even when I'm not actually in the room.

You can in fact deliberately install yourself in other people's heads, if they're at all inclined to let you; some of my best lectures while at CFAR included me doing exactly this.  The key, as with developing your own shoulder council, is to focus on making yourself emulable.  Making your outputs reliably generable from inputs, having a specific and legible style or vibe.  If you've only got an hour, this usually means being pretty blunt and repetitive and keeping things simple:

"... so the one question I want you to keep asking yourself is 'do you know what you are doing and why you are doing it?'"

[5 minutes pass]

"Say it in my voice, in your head: 'Do you know what you are doing and why you are doing it?'"

[10 minutes pass]

"And what would Duncan say, at this moment?  Can you picture his face? ... that's right, he'd say 'do you know what you are doing and why you are doing it?'"

[40 minutes pass]

"... months from now, you're going to be sitting in your room, tired and frustrated, and you're going to look up at the clock, and you're going to sigh, and then you're going to hear my voice in your ear, and it's going to say—"

By this point, I get a message roughly once a quarter, from former students or former workshop participants or people who saw me at a conference or talk, letting me know that their shoulder Duncan appeared for them in a pinch, and that they were (usually) quite glad that he did.

If you have more than an hour to interact with someone, you can be a bit less cheesy than the above example, and encourage the same sort of feedback loops I described earlier, from the other side—Nate, for instance, often asks for the specific wording of his shoulder advisor, if I can remember it, and remarks on that wording as if he were disagreeing with shoulder Nate in a casual conversation, correcting and improving it.  

(It's just such little mannerisms that allow a shoulder advisor to be "really real"—to bring it to life, give it a personality separate from, and not dependent on, your brain's main central personality.  Again, I don't have a sound explanation of the mechanics, but it works.)

You can often make this happen by simply asking your friend or colleague or coworker to predict what you'll say, in response to a given question or prompt—

(Asking them to predict is in general better than asking them to guess.)

—and as icing on the cake, this has the added benefit that, not only are they refining their specific model of shoulder-you, they're also secretly practicing the general skill of booting up a shoulder advisor at all.

Speaking of which ...

Recap & Conclusion

This section is left as an exercise for the reader—try booting up a shoulder Duncan and see what parting words he has to offer, before you (hopefully) leave a comment down below.  And if your shoulder Duncan doesn't have anything at all to offer, see if anyone else feels like chiming in.

New Comment
130 comments, sorted by Click to highlight new comments since:
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings

It's also important to avoid bad shoulder advisors. I've spent several years trying to reduce the influence of miniature copies of abusive family members on my thinking.

EDIT: The most effective counter I've found for this is to 1) Notice that the thought I just had is actually coming from a bad source, 2) Remind myself that that person wanted me to believe/act that way for selfish and narcissistic reasons, and I shouldn't take their advice for the same reason I wouldn't take moral suggestions from people who go around kicking puppies.

This was a significant lesson I learned between the ages of 13 and 23. I have repeatedly removed bad advisors from my shoulder.

8[DEACTIVATED] Duncan Sabien
I attempted to add some thoughts on how I'd go about this, but I'd love to hear a primer on your general method.

A model that pops up in several places (e.g. this book, this paper) is that these kinds of shoulder advisors show up as a kind of a preventative measure. If there are real people who would criticize or berate you for doing specific things, then your brain learns to predict when they would do that, and starts creating that criticism internally. That way, the inner critic may prevent you from doing the thing and thus spare you from being punished by the external critic who's being modeled.

In that case, one approach is to simply try to talk to your inner critic and ask it what it's trying to achieve and what it's afraid would happen if it didn't say the things it did. Sometimes it may be possible to get it to notice that e.g. avoiding the abusive family member's judgment isn't very important anymore, because you're no longer living with that person, getting it to ease off.

I've done it twice, very explicitly. (Probably more implicitly.)

Here's some thoughts, tuned specifically to the cases I had.

  • The individuals would bring very 'defensible' arguments forward. And they would stubbornly refuse to change their mind in response to my improved understanding of the world. I knew they'd never change my mind, so I was always stuck debating them, I could never move on.
  • They had a certain level of status in a community (e.g. one was a public figure who sells books and gives talks) that was not status I was giving them. It wasn't like a friend I could just stop being friends with, they would continue to 'be in the public environment'.
  • It took me a long while to go from "this person seems wrong and set in their ways around ideas and norms that I do not support" to "as best I can tell, in some important ways this person does not live out virtue and I do not want to consult them when I am trying to understand the world or take action". Their arguments were always very 'defensible' in the given social context. To a significant extent I had to give up on that social context, give up being interested in getting status in that hierarchy, in order to stop caring what they
... (read more)

Oh, I notice that I also have done this sort of thing with a bunch of recent tv/films/content. 

There's a habit of modern content that, when it gets politicized, will "mimic argument". It will pretend to show sincere dialogue and debate, but it will fully swing the deck against one side and in favor of the other, and straightforwardly imply that the other side is unethical.

I can watch political art that I disagree with, I can even put up with good art that has bad political art inside of it, but when it attempts to distort what good faith dialogue is in order to win an argument, I just turn it off. I don't want to simulate that character/perspective or have a dialogue with them/it in my head.[1]

I can immediately think of four times I've done this with shows/content I otherwise greatly enjoyed and admire. I just don't want to learn to simulate them.


[1] Writing this out, I realize it's straightforward darkside epistemology.

Maybe that is one way how entertainment manipulates public opinion: By creating memorable (=easily emulable) characters that become shoulder 'influencers' that promote the official narrative right in the heads of the populace. 

I would love an example, though I realize there are several reasons you might not want to put one out there!
6Ben Pace
Okay, because you asked AllAmericanBreakfast. Though I am not likely to follow-up discuss the specifics of each. Recent examples include the last season of Brooklyn Nine Nine and Bo Burnham’s “Inside”.
The "before" state you describe, where you find yourself having arguments with stubborn advisors who refuse to change their minds, reminds me strongly of rumination. I say this because it's something I'm working on to get out of. I'll sometimes find myself engaged in a pretty adversarial discussion about what boils down to my boundaries and be unable to fall asleep for a few hours. And it's usually the same cast of characters. I've found that I can consciously jump out of it by reminding myself that I'm merely burning energy without changing reality in any way. But I usually have to do this a few times before the "bad advisor" finally quiets down. Do bad shoulder advisors feel like rumination to you?
2Ben Pace
That sounds right.
4[DEACTIVATED] Duncan Sabien
Strong agree (and strong upvote). Some general categories of strategy for working on that: * Setting up a dueling shoulder who's specifically motivated to stand in your defense, or pick apart an irrational or abusive argument, or even just remind you to take a breath and broaden your focus. * Preparing a mantra-of-rebuttal, which could be directly addressed to the annoying advisor ("I do not have to listen to you") or could be more general-purpose ("I will not allow toxic people to live rent-free in my thoughts"). * Using CBT-esque self-conditioning to simply cut the thoughts off, mid-stream, until your brain gets the point. * Doing some kind of internal double crux to find the nugget of truth or usefulness that you do reflexively believe the advisor has to offer (e.g. "What this ghost in my head is saying is wrong but at least it is worthwhile to remember that some people think this way" or "What this ghost in my head is saying is wrong but it does remind me to care about X"), and then whenever it pops up, thanking it for that one nugget and sort of firmly closing the door.
3Quintin Pope
Possibly just the act of installing supportive shoulder advisors would be helpful. The brain only has so much capacity for shoulder advisors, so earmarking some of that for positive advisors may “clog the channel” so to speak. Bear in mind that shoulder advisors can be more abstract than is discussed here. E.g., you could have a wordless, nameless shard of pure positivity and acceptance. Also, I expect shoulder advisors have a global positivity parameter that you may be able to influence. When a bad advisor tries to say something bad, stop them and force them to say something good instead, while imagining that the advisor truly believes the good thing. If your shoulder advisor objects to this practice, “correct” their objection and imagine them encouraging you to “remove the maladaptive cognitive pattern my irrational and unwarranted hostility represents”, or something like that.
7[DEACTIVATED] Duncan Sabien
(I note a prediction that most people wouldn't actually be able to make a wordless/nameless shard of pure positivity and acceptance work, and wouldn't get much out of it if you did, but also I'd be stoked to hear someone's experience with one that did work.)
Maybe imagine a dog?
3Quintin Pope
It’s definitely possible, though perhaps shoulder advisor is the wrong phrase to use at that point. Maybe it would be better to describe such a practice as a nonverbal mental ritual, rather than using an “angenty” framing. You picture an incredibly happy crystal that blazes with light and feelings of positivity and acceptance (for this step, it may be helpful to put a cartoonish smily face on the crystal or to imagine it dancing, hugging you, etc). Then let those feelings radiate out from the crystal and into you, until you primarily feel the emotion from yourself. Allow yourself to be happy for the crystal’s happiness. Your own mood should naturally reflect that of the crystal as you lean into emulating the crystal’s radiant positivity. It may also help to picture the crystal as being delighted to share that happiness with you. In this framing, both you and the crystal are happy to share your own joy with the other. Alternate between you sharing happiness with the crystal and the crystal sharing happiness with, both delighted by the other’s joy. Note that visualisations of the sun, moon, a star, a glowing cloud, etc also work well for this exercise. I find that picturing the light as an ever-shifting rainbow of colors helps add some texture and adds dynamism to the crystal’s emotions. I also have difficulty holding a static image in my head for a long time, and the rainbow effect helps with that.

I'm fascinating by the mechanisms here! You said:

This is not schizophrenia (though for all I know it may use some of the same hardware, or may be a low-key, non-pathological version of schizophrenia in the same way that a healthy self-preservation instinct could be thought of as a low-key, non-pathological version of a phobia or an anxiety disorder). 

This accords with my own sense. 

My current central model for schizophrenia starts with Sapoloski's evolutionary hypothesis (sorry for youtube, it works at 1.75X and I just rewatched all of it to make sure I'm not wasting too much time), which links with how cluster A personality disorders are heritable and occur more frequently in people with schizophrenic parents/cousins.  

On this model, a few of these genes make you a tribally useful shaman or successful leader, and get positive selection (hence why the individual alleles exist at non-trivial levels), but too many all at once in the same person can predispose one to fall into a dysfunctional mental attractor.  

Cluster A disorders (classicly, the 3 on the left in the diagram below) are then maybe "very mild 'schizophrenia', with enough perks to make up for the down... (read more)

5Alex Vermillion
I think I've internalized (from the same source as you) that a lot of "metaphors" about thought are just things I experience differently from a given author. Despite this, I hadn't realized that the "voice of self doubt" is literally a voice, and polled my fiance to check (she said it is). I feel like some of these should be a day's worth of learning in a standard education.
3[DEACTIVATED] Duncan Sabien
Yeah, I am indeed toying around with a 100-ish-day curriculum of such insights.

Just a little bit of context here. I tend to be the kind of person who is generally able to cope DURING an traumatic event but then it may or may not take an emotional toll after the fact. That's context. 
Post- COVID American 6 month lock down (that's what it was for me), during 2021, I became majorly depressed and basically.... well, i guess the word would be ""broke". the best way my phsycologist and i can figure out how to explain it was that my personality fractured. not in terms of voices in my head exactly but i had a sort of ego death where i dissolved into a personal set of shoulder Advisors, demons, and other assorted folk who instead of talking to me were talking to each other. there was no me to talk to. but in more tangible terms my head couldnt cope with itself and my personality just became a set of shoulder personas.

The way i see this article, the shoulder advisor thing works because on a meta level you are in fact that smart. you can make better decisions/ notice more things/ do more things if you have a "trigger" that lets you bring that pattern into a higher level, more in control space. people are relatable in the sense that we relate to people pretty well a... (read more)

9[DEACTIVATED] Duncan Sabien
(I appreciate having it added to the conversation.)
4Alex Vermillion
This gave me a lot of abstract information on the ideas touched on above, thank you very much.

Thanks for this great post!

TL;DR for my own thoughts:

  • I speculate on why shoulder advisors are useful
    • Drawing from ensemble methods in machine learning
    • Drawing from predictive processing and overfitting
  • I discuss generating a common corpus of training data for useful shoulder advisors
  • I discuss a few archetypes for useful shoulder advisors and their uses


Shoulder advisors seem to mirror a common machine learning technique, ensembling, which combines multiple ML models to get better overall performance than any individual model can reach. E.g., an ensemble of ERNIE models holds the current first place on the GLUE leaderboard (a metric for evaluating the general capabilities of language models). Shoulder advisors let you sort of ensemble thoughts across different personalities. Ensemble approaches are most helpful when the ensembled population is diverse and each model tends to specialize in particular types of tasks. That matches your usefulness criteria fairly well.

Predictive processing

If we extend predictive processing theory to internal personality traits, then our own personalities are generated by a predictive process, presumably one that bootstraps itself by predicting beha... (read more)

Two anecdotes:


I recall when I was 18 years old or so, and I'd been arguing with a very religious friend throughout high school. I would rehearse arguments with him in my head, preparing for the next time we'd meet and I'd tell him all the reasons I thought his beliefs didn't make sense.

And for the first couple years of this, in the arguments in my head, I'd always say things like "have you considered point X" and imaginary-friend would say "oh, man, you're right. I am wrong." But then, eventually, I hit points where I'd say "what about X?" and then my imaginary friend would say "so? X doesn't matter, because [counterargument]".

This was a neat thing to discover about my ability to model people. (It also was relevant to the entire "does God exist?" debate – an eventually cruxy point for me is that you totally can build up simulations of people in your head, and I'd expect that to be hard to distinguish from God speaking to you)


More recently, I received benefit from asking my own future self for advice. (In fact, I asked multiple future selves who might evolve in different directions). One future self responded with some concrete, compassionate advice about how one of my coping mechanisms wasn't actually helping with my core goals.

There's in fact at least one book about this. From one of the reviews:

Curated. This is a clear articulation of a rationality skill that I've never seen or heard explained elsewhere. Someone I'm close to gained a shoulder advisor in the last few years that was a dramatic breakthrough for them, but I never thought of it as something accessible to most people, yet this post updates me that there's a lot of value here for many (perhaps myself included).

Many people have multiple shoulder advisors. ...  It's quite common in my experience for people to have shoulder copies of their parents, or their best friends, ...

Is this really the case? 

I recently discussed inner monologue with a friend a he was doubtful whether this was real or just something made up. He didn't doubt that people could imagine something like it but not as something in the way you describe here. I can relate to his view because I also don't have a persistent inner monologue - though I can bring it up. 

If you asked people neutrally what goes on in their mind I guess the results would look a lot like Galton's experiments about mental imagery (mentioned e.g. in the SCC post Generalizing From One Example): Some would say nothing, some would answer like my friend, most would report some inner monologue, and a some would have a lot of inner voices. 

How solid is your evidence? 


It's definitely the case that many people do (and also definitely the case that many people don't!).

My evidence is:

  • My own personal experience
  • Explicit conversations about the shoulder advisor concept, including phenomenological reports of what it's like, with a large number of people including Eliezer Yudkowsky, Nate Soares, Damon Pourtahmaseb-Sasi, Renshin Lee, Eli Tyre, Logan Brienne Strohl, Malo Bourgon, Jack Carroll, and Eric Zolayvar.
  • Instances in fiction and pop culture (the angel-and-devil trope, HJPEV's cast of mental models in HPMOR, Bella seeing Edward in Twilight, a lot of references to "hearing someone else's voice at a critical moment" or "an image of someone's face popping to mind" in any number of works)
  • Some pop-evo-psych and bicameral mind stuff that sort of post-dicts it

I started developing (and sharing) my explicit concepts of shoulder advisors well after encountering the stuff about some people not having visual imagery, or some people not having inner monologues, or some people not having a persistent sense of self across time, or Scott Alexander's "Different Worlds," and so forth.  Like, from the start I've been quite aware of the fact that this won't be uni... (read more)

Update: I polled some people and for what it's worth this is the result:

  • has dialogs and trialogues, that are different shards of their personality and chime in often unasked (but not like an advisor)
  • can easily emulate other people and how they will react and what they would say (it sounded more like in the form of conceptese)  
  • can easily visualize other people and what they would say (it sounded more like in a video recording with lower person fidelity)
  • has minimal inner monologue, but can bring up limited shoulder advisors (me)
  • has almost no inner monologue
  • has no inner monologue, aspects of their personality take control unasked and without negotiation 
  • has no inner monologue and is skeptical such a thing exists at all 

Kids' answers (ages 10 to 15):

  • I construct the other person from parts and can imagine what they would say but not how they would look.
  • I can easily imagine what other persons would say. 
  • The better I know another person the better I know what they would say or do (it sounded less like an imagined person and more like 'just knowing').
6[DEACTIVATED] Duncan Sabien
Can you share something about the wording of your question/prompt?  I might want to ask other people using the same prompt for consistency.
It happened relatively organically in a board-game round. Discussion moved to how to deal with politics at the job, and you have to anticipate people's reactions, and I said that it happens in management all the time. And I think I asked: "How do you do that?" and as a follow-up question: "Do you have some kind of inner monologue?" With my kids, I didn't assume familiarity with such a concept and used a situation they are familiar with. A setup like: "You have a lot of friends and tell them a joke, can you tell how they will react? What happens in your brain/mind when you do that?" And as follow-ups: "Can you imagine your friend saying that?" "Can you hear what tone he uses or what emotions he shows?" (We have talked about mental imagery often before so I know they are familiar with that) While it was not intentional with the adults I prefer this pattern. Kids (at least mine) often don't like direct questions but will happily tag along if my topic relates to their situation. It also avoids leading questions. And if I offer answer options I provide a wide variety of alternatives with cues that there could be more.  So my proposal is to describe a situation where modeling another person is a natural choice. And then ask "How do you do that (in your mind)?" and only in follow-up questions ask for inner mono/dialogue or mental imagery.  
Thank you for sharing your epistemic status. I think this and other questions regarding consciousness deserve more study. It is relevant for AI research, on LW see e.g. Nonperson predicates. This was also mentioned by Sam Altman in the recent meetup. I intend to write a post on some testable predictions about consciousness and I hope to get some into the next SSC/ACX survey. 

I mostly don't have inner monologue either, but I also don't feel like that's an requirement for having shoulder advisors. Shoulder advisors can talk in conceptese, in my experience.

(I don't have shoulder copies in the strong and vivid sense that Duncan describes, but I sometimes get vague feelings that seem like they are the outputs of the same kind of process. E.g. reading an article and thinking "I feel like Duncan would be interested in this", seeing a particular plush toy and thinking "my friend S would find this one adorable", or preparing for a fraught conversation with someone and doing a simulation in my head of what I'd expect them to say in response to various things that I could say. Also I'm getting a weak sense of what you might reply to this comment, now. :-)  )

Agree. Esp. the Which worked best for interacting with a copy of myself. See my other comment.

Just to make explicit a connection that seems obvious to me but I'm not sure how obvious it is to others: the existence of this phenomena fits nicely together with a global workspace model of the mind, where the brain may spawn new subroutines that plug into the workspace and then learns various rules for when to activate them, as well as fine-tuning their properties when the system becomes aware of a mismatch between the model's predicted outcome and what-the-target-person would actually have said.

If that is true that would also mean that the ego could be just one of these shoulder advisors - albeit a privileged one. 
I would guess that there is some additional machinery involved in the ego compared to shoulder advisors (this might not contradict your description of ego as privileged shoulder advisor), as tulpas seem to be quite related to shoulder advisors while being 'closer to ego' in some sense.  Probably this distinction is an important reason why shoulder advisors seem much less problematic from the standpoint of mental health.
What additional machinery do you have in mind or what else makes you think that?
My thoughts on this are mostly from introspection. When I try to imagine a shoulder advisor in comparison to my self (note that I do not have shoulder advisors currently), there seem to be some additional properties to my self which a should advisor would not have. Trying to get at the differences, what comes up is: * bodily sensations and urges are 'directly fed into and fuel (/delegate vote power to)' my self, but not shoulder advisors * decisions on movement likewise are directly connected to myself, while shoulder advisors are only influencing my mental dialogue/perception * similarly with things like 'felt responsibility for actions', 'identity' etc. I am not sure that 'additional machinery' is the right term for these differences. My impression is 'the ego is much more strongly connected and fused with these other parts'
You describe it as a matter of degree and I cant disagree with that. 
Thanks for your reply, am an unsure whether I am correctly understanding your position: * Would you agree that some of the aspects that make the ego/self different compared to shoulder advisors are the ones that I stated? (it doesn't seem to contradict the formulation 'privileged shoulder advisor' as far as I can tell) * The 'matter of degree'-question where our views differ is about the question whether there are such things as 'shoulder advisors+' that are e.g. halfway between a pure shoulder advisor and the ego?   If I am not misunderstanding you, this is a really interesting disagreement. To me, this topic is almost an alief-level view and I might have fallen for cognitive fusion. To possibly help with misunderstandings, I'll just throw out my current thoughts on the topic. I expect there to be major mistakes or at least gaps in there, but I will not be able to improve them without working with them. Of course, it is also not necessary to solve our disagreement if you find it less interesting. My current position: In principle, the agentic unity of the ego can be lost and instead scattered onto different agentic-type thought processes (think trauma, dissociation, identity disorders etc.). My impression is that this is usually quite harmful to the individual and generally not sth to strive for. Then there is a more or less separate 'space of agentic-type thought processes' that leans more toward 'abstract world-modelling', 'agent-modelling' and 'abstract-goal-pursuing' and less toward 'episodic memories', 'bodily sensations', 'bodily urges' and 'fight or flight response'.  Here, although most people tend towards unity from the ego plus a large number of simple goal-processes plus modelling people, it is possible and largely non-dangerous to build towards a number of larger bundles of thought processes such as shoulder advisors.
A, I missed a "t". "can" -> "cant". Sorry about that typo. I mostly agree with it being a matter of degree. But I want to respond to this part of your comment:  I wouldn't say that this is what happens with Shoulder Advisors or with the no-self experience of meditation. There are many failure modes of the brain making sense of agency and identity. I think the default mode of society is to encourage and reinforce an interpretation around ego, identity, and agency which is stable and beneficial (at least in the sense of societal productivity, I guess there are cultures with very differt patterns that are stable but probably less scalable e.g. the Piraha). 
Ah, this makes sense thanks!   This sounds right. Maybe the cases that I am concerned about additionally contain fear responses, and purely having a non-unified or unclear sense of self is more normal/safer than I thought. 
Scott's book review of Origin Of Consciousness In The Breakdown Of The Bicameral Mind also seems related.
Consider that GPT doesn't have any of that fancy stuff and yet can generate dialogues of semi-consistent characters. Shoulder advisors can be slightly-fancier text bots just by adding audio tone and facial expressions to what is being prompted and predicted.
True, though much of that fancy stuff is related to things that GPT doesn't need to do, e.g. it doesn't have any of the learning components because it comes pre-trained and doesn't do any on-line training, and it's also just one predictive engine rather than being embedded in a larger system that needs to decide when to apply that predictive engine and when to apply something else. I do agree that if you look at just any given "shoulder advisor module" itself, and not any of the components concerned with updating it or deciding when it should run, it does seem quite similar to something like GPT.

Two shoulder advisors I have found helpful this year:

  • Daniel Filan: has a knack for asking 'trivial' or 'obvious' questions in really situations that end up really opening up or clarifying what's happening. (It's part of why he's a good podcast interviewer at AXRP.) I've started to anticipate in my conversations where Daniel would ask a question, and asked it myself, and gotten great results out of it.
  • Gimli, son of Glóin, from the Lord of the Rings: I read the trilogy for the first time in January, and in many situations where I've been tempted to have long drawn-out passive aggressive interactions with coworkers or wander into a pit of despair, I noticed what he'd do: he'd just get on with the work. This has been very helpful.
4Mary Chernyshenko
One of mine is Athos from 'Twenty years after'. I admire his dedication, plainness of speech, valor, level-headedness, ability to just not defend things he finds unworthy of defence even when it would be to his advantage, etc. At most, he "says" something like, "Shall we? Yes, we shall." - and this is enough.

"it is a straightforwardly observable fact that, for many people, their shoulder advisors occasionally offer thoughts and insights that the people literally would not have thought of, otherwise."

How can this be observable, let alone straightforwardly?

6[DEACTIVATED] Duncan Sabien
+1 for noticing the sketchy wording; thanks for the chance to clarify. I intended to convey that the fact is straightforwardly observable, not the phenomenon itself.  It's the sort of thing you can confirm is true by just going and asking a bunch of people, as I have done and as Gunnar reports doing in another comment. You are correct that I can't literally observe this happening inside any heads other than my own (but it is indeed observable inside my own head, e.g. by adding a new shoulder advisor and suddenly and consistently getting a whole class of insights that I never got previously).
But what I'm saying is that the fact isn't observable. You can't know what you would have thought of otherwise.
3[DEACTIVATED] Duncan Sabien
You definitely can. Or rather, to be more precise, rounding off a question like that to "yes" or "no" is itself the mistake, fallacy-of-the-gray style. Sure, you can't know for sure what you definitely would have thought of otherwise. But you can absolutely gather tons and tons of relevant and reliable evidence that allows you to have high confidence.
So when you say "you definitely can"..? I would be interested in your explanation of how you can have any idea what you would have thought of otherwise. As far as I'm aware it's a pretty basic and obvious truth of empirical inquiry that you really can't make almost any reliable causal/counterfactual observations about yourself at all, for self evident methodological reasons. Why do you think this is an exception? And why is your confidence so high that you describe it as "definite"?
4[DEACTIVATED] Duncan Sabien
I'm not accepting the burden of proof here.   Why do you think that "it's a pretty basic and obvious truth of empirical inquiry that you really can't make almost any reliable causal/counterfactual observations about yourself at all, for self evident methodological reasons"? EDIT: this was more dickish on reread than I intended; sorry.  Here, at least, is a gesture in the right direction, but I don't have time to lay out a full proof: [STIMULUS] Response: has thought of type A [STIMULUS] Response: has thought of type A [STIMULUS] Response: has thought of type A [STIMULUS] Response: has thought of type A [LEARNS NEW SKILL OR TECHNIQUE] [STIMULUS] Response: has thought of type B Conclusion: "It's reasonable to believe that without this skill, I would have had another thought of type A." Especially if one observes this exact same pattern in multiple people across multiple contexts.

Two additional beneficial outcomes of attempting to boot up / improve shoulder advisors:

  1. You are more likely to really pay attention during interactions with them. You get better at interactions in general if you are quite focused. And others tend to notice and react positively when they see that you really care about what they are saying.
  2. You learn to be more curious about others' thoughts, actions, and backgrounds. This can help you be more empathic, and can also help shine light on your own motivations and influences.


Anecdotal, but the friends of mine who simulate people / hear others' voices tend to be among the most thoughtful and socially-buttery-smooth people I know. 

5[DEACTIVATED] Duncan Sabien
True in my experience as well, and in the experience of at least a couple of the people I've talked explicitly to about this.

Oh, I just realized this post is new, and not one of the 8 year old posts I keep coming across and want to participate in.

I don't think this technique works for me, at all. For several reasons. And reading through the comments section, I feel like other people can't imagine my failure case (e.g. the people surveying others about whether they do / can do this - "Can you imagine your friend saying this?")

I'm visually aphantasic, I cannot conjure a visual image at all except in some phases of sleep, or sometimes under extreme tiredness while 'awake' (I hypothesize those mental states are closer to sleep than normal waking awareness anyway). This means the 'visual image' part falls apart instantly. I can, however consciously control my auditory perceptions with extremely high fidelity.

This means I can easily make a voice in my head say anything in a specific person's voice, but that can only happen with my conscious input. These voices are purely puppets, or more accurately, just 'voice filters' I can put on my 'internal voice'.

I believe I've never 'heard' anything other than what actually happened or came directly from (or through) my internal train of thought, with rare exceptions of... (read more)

On more self reflection, and reading a bunch of posts in the subagents tag, and looking into tulpas again, I believe I just don't have the mental architecture for this kind of thing.

I hypothesize that this skill requires thinking of yourself as a personality, but I don't see myself that way. I see my 'self' as my central attention. I don't believe I have software subagents, I believe I have hardware submodules with some configuration overlays at most. There's no 'default personality'. I switch how I interact with people based on context fluidly.

I don't believe my architecture is flexible enough for this, as I can't even write fictional characters that are not a version of my mind with masked knowledge and hidden or exaggerated traits (and otherwise with a manipulated context). I don't argue with people in my head, I don't do 'what ifs' of conversations, I don't imagine what a person would say in some situation.

Interesting - if someone told me only that they have no default personality and that they switch how they interact with people fluidly based on context, I'd assume them to naturally view their mind in terms of subagents. Since "no default personality" sounds to me like "no unitary self, just different subagents that get activated based on the context". How do you define the difference between software subagents and hardware submodules?
5Flawed Spiral
I (feel like I) have a good understanding of what exactly which major 'parts' of my brain are capable of. I know what I can do by language modeling, I know what I can do and when with my visual cortex, auditory cortex, kinesthetic sense, motor cortex, working memory, I know what I can do with my spatial awareness. I can consciously focus on most of these parts and affect their workings, mostly by bringing them into attention. I'm also aware of, but not in direct conscious control of some other parts of the brain which makes me somewhat aware of things such as Ugh Fields. (I don't have names for all such submodules, only the most obvious ones that seem to match up to things I've read.) These all are completely non-agenty, even if they can somewhat work independently of my central attention. Say, when you're not focusing on sound, do you view your auditory cortex as a sub-agent when it brings to your attention the fact a loud sound just happened? I don't. I also don't see these parts as separate from 'me', and I don't communicate with them in any way except with raw attention. I should also clarify that despite my description, I don't mean whatever controls executive function when I say "central attention". I mean the part of the brain that controls the importance that affects what stays and what's replaced on the 'main bus' that other parts of the brain dump data on. My interpretation of software subagents is that people can install a (possibly pseudo-)personality that runs in certain parts of the brain, or at least interfaces with them while your central attention is elsewhere. Importantly, it's able to use various mental resources without it coming to your central attention. This interpretation is likely wrong, as I have no experience with this other than reading people's posts on the internet, which loses a lot of detail. Regarding "no default personality" and "no unitary self". I don't think that's the case (unless I misunderstood the term), I do have a self,
That again sounds pretty close to how I would have described myself earlier (and would mostly still describe). You didn't write about your emotional balance but if you're again like me, it would involve few excited states and fewer conflicted or what's called negative states and emotions e.g., anger. If so, that would confirm my assumption that getting to such a uniform and stable state requires a certain environment. An environment that has little need for the developing brain and mind to overfit. No forced adaptations to environmental risks like loss of caretakers or life. But also an environment rich in information and worth exploring. 
4Alex Vermillion
I would like to tag in as a 3rd in this group and say your estimate at my growing environment matches how I would describe it. I have little-to-no interior emotional conflict and explicitly modeling others seems alien to me. [Edit s/know/no]
4Flawed Spiral
True, I didn't mention emotional balance since it usually plays little role in my daily life. I used to have issues with managing extreme emotions in early childhood that I solved by both avoidance, and 'dimming' them to the point they are mostly manageable. I avoid anger in daily life because it was always unproductive for me in the past, and is incompatible with the social strategies I use nowadays (which I picked because I suck at social 'tactics'). The environment you describe matches up well, but not perfectly with the one I grew up in. I guess you can use that as a confirmation.
Thanks for sharing. I can relate even to the exceptions and resulting strategies.
I can relate very well to Flawed Spiral. Before I looked more into meditation and got a higher resolution awareness of inner processes, I could have written a description like him. For me, it was not so much "no unitary self" as rather a tiny self. I didn't have much of an inner monologue. And I could easily relate to Paul Graham's appeal to Keep Your Identity Small. A small self doesn't mean low complexity. For me, it was a high alignment of inner processes and values as well as a rich and updatable world model. 
5Flawed Spiral
I personally see an inner monologue as as much of a tool as any other part of my brain. The inner monologue, as a tight coupling of auditory and linguistic processing, is rather helpful for performing some kinds of thought, for extending working memory (the auditory loop is an extremely easy place to store small amounts of nearly arbitrary data in the immediate term, and you can abuse your language processing to store moderate amounts of linguistic data in the short term as long as you're able to retrace a path of thought through it). I do find that I don't have a constantly running narrative of my thoughts and what I'm doing, even if I remember having one in the past. I still use internal monologue to trigger parts of my brain for things like planning, or for enhancing myself in some task as described in the earlier paragraph, but most of the time my inner monologue is inactive. I do agree with Keep Your Identity Small, I seem to have been doing that, or something very similar, automatically from a certain point in my mid teens. This does have a side effect that I never really feel like part of most groups, which is both good and bad, as it allows me to exit groups or communities easily, and for example, permanently 'shed' online identities that I decide I can't use anymore for whatever reason (like sharing too much info that's reasonably correlatable to another identity or real life). I'm curious what kinds of meditation you've looked into. My go-to form of meditation is focusing attention on my body, in any position, with or without muscle relaxation or increasing blood flow.
Same.  Or for recalling previous conversations or rehearsing speech (though that also falls under extended working memory). Same here. And that, together with what you wrote earlier ("I don't think I can model people"), leads to less of a felt connection to other people - in both directions: It makes us harder to model. It is why I have tried to pick up skills in that direction. The way our mind has developed makes it harder - but I think if we succeed, more fluid. Since being a teen, I have done a lot of self-introspection. Meditation looked suspicious to me for a long time. I knew about its benefits though the same is said about religion. I was delighted to find a non-dogmatic introduction on LessWrong though I'm not such which one of the many under the tag meditation it was. Probably one by Kaj.  I tried the breathing exercise, and it was effortless. Same with other exercises. I had trouble locating emotions in the body and was skeptical, guessing it being illusory (same trouble Duncan has). I attended a 10-day silent Vipassana meditation retreat two years ago organized by an LWer and billed as non-dogmatic and open to individual needs. It worked out incredibly well. The teacher (Julia Harfensteller) provided a lot of exercises and cues from multiple directions. The resolution of my introspection increased immensely. At the end, I gained access to my emotions - previously, they had been so well-regulated subconsciously as to be almost invisible. In the weeks after, I went thru big parts of The Mind Illuminated (see e.g. here).  Things I did: * Breathing meditation (decompose the sensations of the breath) * Decomposing visual perception ('unseeing' shapes, forms, motions, faces) * Noticing beginnings and ends of thoughts down to pre-thoughts. * Noticing bodily sensations, itches, bodily posture with high resolution. * Noticing and naming emotions and noticing bodily correlates. * Regulating emotions up and down e.g joy. * Tuning brain modules up and d
3Flawed Spiral
I overlooked the obvious, yes, I do that too, of course. However, less of the rehearsing speech part, and more of looking for concrete words for concepts in the moment. I do believe I would improve the fluidity of my speech by rehearsing, I'm not sure that kind of practice is aligned with my values. Most of your meditation description sounds fascinating, it seems mostly like practicing the skill I already have to strengthen the connection between direct sensations and conscious attention. The only parts that I've never consciously done before are regulating emotions up, and paying attention in general while in emotional states. I still find backtracking through thoughts difficult, and am not completely successful. I think the way I practice is not particularly effective, but I would like to improve. I'm not sure I'd be willing to go to a meditation retreat, I'd have to re-evaluate quite a few things to consider actually going.
I think the two big advantages of a retreat are * a lot of time in one large chunk to improve the mind (introspect, meditate, or something). As with programming some things you can only do if you go deeper and deeper in one run (extreme maker schedule). * tight feedback loops with the teacher and other practitioners hopefully at about the same level.  Both interrelate. But with your specific profile and experience level, I think it will be difficult to find a suitable retreat. It might work better to work closely with a meditation practitioner that you click with.
4Said Achmiz
This also describes my experience.
2[DEACTIVATED] Duncan Sabien
FWIW, my experience of writing fiction is very much "they are versions of me with masked knowledge and hidden or exaggerated traits."
2Said Achmiz
This describes my own experience fairly well—except that I am definitely not visually aphantasic. This part is spot-on for me, though:

One practice I've been doing a bit since January has been something called Ideal Parent Figure Protocol, which includes guided meditations (e.g.) for imagining yourself as a child with the kinds of idealized parents who are always perfectly supportive and understanding and available, to correct for any emotional lacks created by the ways in which your real parents were just human and non-perfect. One of the parts of the practice is something called "microhits", which basically means making your ideal parents your shoulder advisors so that they'll be available for emotional support whenever you need it. (I haven't gotten this very strongly, but I've heard people say it's really powerful if you do get it to work.)

The hard part of making something like this work is that if your parents were messed up enough for you need to do this, your concept of an "ideal" parent is probably pretty broken, though perhaps in subtle ways. There were a lot of counterintuitive things I had to realize about parenting that aren't well-understood in popular culture, to get this kind of thing right. (Also, if you do get it right, then a lot of the time you can just use memory reconsolidation on the events where things didn't work out the right way, and then you don't need the shoulder advising on that topic any more, because the new response is embedded in the schema for responding to situations like that.)

For some time, I have tried to create mental models of other people to improve my social skills, better tune my communication, and predict reactions. It was always very difficult to put myself in other people's shoes. For a long time, I didn't understand what that even was supposed to mean. Your instructions have been extremely helpful and I managed to bring up boot up (at this point one piped up and said "call it 'boot up'") some advisors easily.

It is easiest to let the advisors just be there and act without speaking. Smiling, noding, moving.

Emulating by ex-wife - at least her speech - is relatively easy. I'm not sure what to make of it; we are well cooperating co-parents, maybe there will be parenting advice that will simplify things.  

Among the advisors that I brought up the most vivid were instances of myself. The first tries were rather boring because the copies didn't speak up but just thought and observed. Only when I changed the frame to include thoughts did this lead to some cool interaction. The copy was disoriented at first, explored the environment, asked for access to senses, and brought up nested instances at which point I decided to write up the experience. 

9[DEACTIVATED] Duncan Sabien
Thanks very much for this comment.  I've added in a paragraph (ctrl+f "Note that you don't") to emphasize the point about not requiring them to speak.

You might want to add a disclaimer when not to try this or what the risks are. Meditation risks have been discussed on LW before. I don't want to overblow this, I think it is pretty safe. But it is an experimental meditation practice and you could link to the risks section of SSC Book Review The Mind Illuminated or offer to help if anything unexpected happens.  

Also, your mileage may vary. You couldn't sell me on all the nice points you listed - having a lot of time to work on it I arrived at a stable inner life via other means. What sold me on your practice was your aside 

... not to mention that having robust copies of my actual friends and colleagues has much better equipped me to interact with those friends and colleagues

7Quintin Pope
On the one hand, I agree that potential side effects are important. Shoulder advisors seem very similar to tulpas, and mental health disorders are very common (~50% or so) in the tulpamancy community. Though this paper argues that this is because mental health issues cause people to be drawn towards tulpamancy, and that tulpamancy can benefit those with mental illnesses. Of course, people who had negative experiences with tulpas would likely leave the community and not be available for surveys. On the other hand, shoulder advisors and tulpas are fundamentally exercises in prediction. Your perception of the practice influences the results you’ll get. If you create shoulder advisors with the assumption that they’ll go wrong immediately, your odds of a beneficial outcome drop. It’s thus important not to over-emphasize potential negatives.

My own personal experience following this post: I don't have enough training data for most of the people I'd like to emulate. When I think of the people I know irl that is like to learn from, I've spent about ten hours 1-on-1 with each of them; not enough to have a solid mental model of what advice they may give. At the same time, part of why I value their advice is that I can't predict it; they have wisdom and experience that I don't. Often, I'll ask them for advice and be surprised by their answer. When I tried to create a shoulder advisor of one of them, it didn't work; I just didn't know enough about them to accurately understand what they were thinking in a certain situation.

Still a great post, though; just didn't work for a specific use case of mine.

5[DEACTIVATED] Duncan Sabien
That makes sense. I (weakly) predict that building a shoulder advisor or two out of less-useful-but-more-emulable people might be worth it via giving you the skill of emulating to the available max?  Such that, finding emulation in general a little easier and more familiar, you might be able to try again with the actual higher-value targets? FWIW, I have been genuinely surprised by advice from shoulder advisors that I could not predict; in a very real sense, that's the primary claim of the post.  You don't have to have the same wisdom and experience to spin up a mental chatbot that will sometimes be able to mimic the pattern well enough to produce something novel and useful.  If having a good shoulder advisor required being as wise and experienced as the real person, I don't think I would have felt this post was worth writing.
That makes sense, I'll give it a try. Ah, I see.

I really liked this post since it took something I did intuitively and haphazardly and gave it a handle by providing the terms to start practicing it intentionally. This had at least two benefits:

First it allowed me to use this technique in a much wider set of circumstances, and to improve the voices that I already have. Identifying the phenomenon allowed it to move from a knack which showed up by luck, to a skill.

Second, it allowed me to communicate the experience more easily to others, and open the possibility for them to use it as well. Unlike many less... (read more)

So uhh

I've been doing a bit of coaching for people recently

And then when I was thinking that I'm not going to do [THING] yet, I'm going to wait until I'm in a better position to do so, suddenly I had the experience of a shoulder advisor materializing that was me in coach mode being like "okay so do you have some actual criteria for what counts as being in a good enough position"

That was a very peculiar experience

(probably I'd have had that thought anyway but reading this post primed me to have it be accompanied by a mental image of myself standing on my own shoulder)

Great post. Two comments:

The popular saying "What would Jesus do?" suggests many devout Christians use Jesus precisely as a shoulder advisor - no doubt frequently & with intense seriousness. Hence they may well have useful insights into the technique.


It's important to be clear that the experience of "hearing the voices" actually happens, in many people. This is not a metaphor, and it is not hyperbole or exaggeration. I'm not saying that people tend to hallucinate actual sounds—that probably would be schizophrenia.

I read a book about Jaynes' bicame... (read more)

2[DEACTIVATED] Duncan Sabien
Yeah, the bicameral mind stuff was definitely in my thoughts as I explored this concept explicitly over the past few years.

On the having a shoulder Duncan, I explicitly tried to do this upon listening to you being interviewed by Spencer Greenberg.  Sadly, digital Duncan doesn't exist in large enough quantities to emulated.  So for those of us that don't happen to live (I assume) in the Bay Area, can Duncan increase how many podcast or video's he's on?  Having a shoulder Duncan sounds really useful to a large number of people you interact with and enlarging that pool purposely seems to be a pretty good idea. Food for thought?

My best talk

My longest talk

My Harvard talk

My badly-in-need-of-updating website which happens to have a "writing" tab F U L L of stuff

You can find a couple others on Youtube by searching "Duncan Sabien."

Also you can get people to invite me to podcasts and I'll often say yes. =P

This phenomenon is also why we have the term "role model." Successful examples of people similar to us are extremely valuable, and it is in fact very difficult to succeed without such examples.

I think this is part of the reason for the popularity of the Meditations of Marcus Aurelius.

5Mary Chernyshenko
And just the good old absolutely trustworthy Roman army guy who does what he has to do, sings as he whittles his arrows, looking up to smile cheekily at his dejected friend and wiggles his sun-bleached eyebrows with an outrageous joke that makes one actually forget the reason they approached him. Perhaps not the most helpful character, but.
That's entirely possible. It actually appears that the ancient Stoics actually used something like this shoulder advisor technique. IIRC, Donald Robertson (prominent writer on Stoicism) actually suggested that Marcus wrote his journal entries as a way to model one of his own deceased Stoic teachers. I myself have been working on trying to create a shoulder advisor based on the Meditations, but it's been slow going. I was actually inspired several months back by the recent popularity of tulpamancy, not all of which I buy into but which seems to have some interesting techniques along these lines.

Thank you for writing this, it really got me thinking. I'm one of those people who don't really have a firm cast of shoulder advisors. In fact, when I saw this appear in fiction (and in particular in HPMOR), I kind of assumed it was just a convenient narrative device and not something real people actually do. I suppose I should read HPMOR again and try to figure out what other blatantly obvious advice I've missed.

This does seem like a extremely useful skill to have, so I'd like to practice it if possible. I just tried to imagine one of these shoulder advis... (read more)

I think that getting a stable visual appearance isn't very important; I'm sure it can make things more realistic if you do get it, but the most essential thing is getting a hold of the person's felt sense, as well as a feeling of how they would react to different things. If you get a sense of "ah in this situation, they would say X", then it doesn't matter if you're unable to clearly visualize them.
I can bring up the idea of a shoulder adviser easily. I can also situate it in the room. It has emotions. But it is not visual. I don't 'see' the emotions or the advisor. At best it is a sketch. That's normal for me, my visual imagination is 'sketchy' (see Generalizing From One Example).

A related (ADDED: but more intrusive and risky) technique is called Tulpa and was discussed earlier in this post: How Effective are Tulpas? 

I am comfortable with having a comment making the (objectively correct) claim that tulpas are related in a narrow and technical sense.  They seem to be playing in a similar space, are probably using the same mental architecture, etc.

I would not be comfortable with having a comment leaving the (unjustified, imo) impression that tulpas are substantively similar.

Like, it may be that tulpas get a bad rap, but from what I know of them, they're much more like inventing a shoulder advisor and then ceding control to it entirely because you think it can run your life better than the core you, and that's way more extreme and requires a lot more assumptions to justify than the thing I'm recommending with shoulder advisors.  Their common-use definition is a thing that feels risky in a way that shoulder advisors do not, and feels like it requires warnings that I don't think shoulder advisors require.

EDIT: Also, afaik tuplas are much more built-from-the-ground-up, rather than being keyed into a set of recorded experiences from either real people or detailed fictional characters.  Having to ground out a mental construct in either actual reality or plausible near-reality seems like a big safeguard.

Even if I'm wrong about what tulpas really are in practice: to the extent that my understanding and my brief description above matches other people's general impression, I want to be pretty firm that that thing is not closely related to shoulder advisors in spirit.

Well, the way you want them to be is different and less risky but as you point out: They likely run on the same mental architecture. My shoulder advisor was emulated smart enough to ask for access to senses, which was fun to play around with and felt a bit like 'ceding control'. It is probably a good idea to make sure that you create them really as advisors otherwise the downvoted points might apply.  
The connection to Tulpas is close enough that curious smart people will hit on them if you want it or not. For example, it was the second comment in Kajs reshare.
Seconding this, I noticed the parallels the moment OP started talking about advisors injecting comments. Tulpas can be instantiated from fictional characters, and these are called fictives or soulbonds. And it's not about ceding control, I think that's more of a did thing where some alters will hide because they're traumatized and afraid. I suspect that feeding an advisor attention (by talking to it a lot) will help it grow into a tulpa/alter. But I'm not sure, my advisors aren't even at the level described in the post. I have to invoke them deliberately, and they disintegrate as soon as I'm done talking. On the other hand, I don't spend hours conversing with them every day, which everyone seems to agree is part of instantiating a tulpa.
I noticed this also but intentionally did not bring it up because I consider this area to be extremely negative. Hearing that someone is getting into "tulpamancy" is for me a gigantic red flag and in practice seems linked to people going insane -- not sure if it's causal or correlational or what but I would very much like the community to avoid this area.
I agree that a community can and should avoid certain topics. For example, the "politics is the mind-killer" no-politics rule. And this is probably true, independent of whether one understands why something is dangerous. But there are two aspects here: Understanding why something is dangerous and actually trying out the dangerous thing. Granted, one can easily lead to the other. There is also the other side of the coin: Understanding why something is healthy/beneficial. LW is also about that (see Lifestyle interventions to increase longevity). There is a lot of grey - or the healthy part is an island in a big grey sea. By excluding discussion of interventions, you exclude a lot of good. And by excluding discussion or mention of the grey around a good, you risk people wandering into it unwarned.  

One of mine is a (real actually) gander whose name is Жирний (Fatty). The fictional him is more optimistic / not-depressed than me; a generous, unsophisticated, slightly egotistical, proud of his flock, kinda shy, but of times wordy, magnificent bird. He just never hesitates to be kind to others when my self-image would invent a carload of reasons not to.

I love how Jean-Luc Picard was selected to be one of your advisors. He is also among my best candidates :).


What is the largest number of advisers you have known people to actively use? I am a bit reluctant to cut it down to four or five.

8[DEACTIVATED] Duncan Sabien
At any one time (i.e. in any one specific situation) the most I've ever seen anyone juggle is about six.  Like, I have sometimes actively booted up 5-7 advisors over the course of a ten-minute introspection, and have seen others do the same, and sometimes those advisors "talk to" each other directly without me feeling like the Duncan-personality is doing anything other than watching. But as for my overall cast of shoulder advisors—it's well over thirty?  Essentially anyone I get to know past a certain level becomes emulable, and many many people might pop onto my shoulder only once or twice a year, or only once ever.  But there are at least thirty people (maybe fifteen real and fifteen fictional) who I regularly emulate. Recommendations to cut things down were solely for the purpose of "if you've never done this/have no experience, don't try to do too much at once."
I am curious about the effort it takes or the impact it has on resolution, or the time needed. My predictions would be  * no two advisors can talk at the same time but take turns (but can interrupt) * effort or at least time scales mostly linearly * Dunbar's number applies
4[DEACTIVATED] Duncan Sabien
Definitely I've never had more than one "voice" going at a time, even if there are two or three voices interrupting and taking over in rapid succession.  This accords with what I've heard from other people, though I haven't specifically asked about whether people ever hear two-or-more. I think in a given three-second span I can track something like the "mood" of at least three different perspectives or advisors at once?  The same way that (if my own experience generalizes) you can be in a room with ten people, and someone says something, and you immediately have a visceral feeling about how several of the other people will react.
Since a recent meditation retreat, I can notice the mood of a group of people - but that isn't the same as having a specific feeling about all individual reactions and aggregating over it. If you can do the latter I'm impressed.
4[DEACTIVATED] Duncan Sabien
It's usually far from "all."  It's more like, in this room with ten people who I know reasonably well, any given development will tend to provoke specific, identifiable reactions from my mental model of 1-4 of them. If ten minutes go by, my attention will land on almost everyone at some point, and I'll have clear intuitions for almost everyone at some point, but in any given moment some will be much more salient than others.
That sounds like what experienced managers I know seem to be capable of, though they wouldn't phrase it in terms of advisors.
4[DEACTIVATED] Duncan Sabien
Yeah, at this point we're drifting away from the concept of "a curated mental model" and into just "general mental impressions," or something.  But they're contiguous concepts, in my head—they're just different in something like vividness or intensity or clarity.  I can "turn my focus" toward what's just a whispering impression, and it becomes more of what I'm thinking of as a fully-fledged emulation, or I can shove an emulation into the back of my mind and ignore it and it subsides into just being a tickle of awareness.
Yes. That rhymes well. I can better describe what goes on with these emulations now.  My primary mental mode is thinking in concepts. Probably related to what Kaj called conceptese. I struggled to find concepts for people; some intermediate structure was missing. With your shoulder advisors, I have found it. And I can apply all the machinery that I have for concepts.  With meditation practice, I got better at noticing thoughts forming or concepts activating. Following your instructions, memories of friends are brought up, and let babble - mostly their expressions. Some do feel easier - have higher emulability. There is quick back and forth between your suggestions and own experimentation. A list is updated with results. First spontaneous activations happen - just fleeting sentences. Like an idea coming up.  My prediction is that this will go the same way many concepts and skills do with me: After a short time of being very concrete, they will stop standing on their own but become part of my overall way of being. Thus less like a person advising but more like just knowing what is going on (your "general mental impressions"). At least if it goes efficiently, otherwise it will remain as a tool used consciously but rarely. 

Simulated intelligence is real intelligence. Although probably nobody is going to simulate the person/figure properly, the typical LW reader will actually occasionally outperform the actual person/figure.

Not sure to what extent shoulder advisors outperform deliberate thinking, especially compared to other CFAR handbook techniques.

I was thinking about something just now and tried to imagine some concept advising me (like morality play style, like getting advice from Temperance or something; the example isn't important and is distracting) and I realized something.

Those of you who experience this phenomenon naturally: where do you experience it? When I tried just now, I noticed my brain trying to create the presence directly side-behind my head. After this post and realizing people literally had the experience of being talked to or advised by entities, it didn't occur to me:

Do you exp... (read more)

3[DEACTIVATED] Duncan Sabien
I have a variety of spatial locations, and sometimes no spatial location at all. Sometimes there is a literally-on-or-just-behind-my-shoulder thing.  Sometimes I intuit a shoulder advisor like twelve feet away off on a 45º angle.  Sometimes I just see a face in my mind, and it doesn't have location; it just sort of "fills" my mental vision. A range of things; nothing consistent.

This makes me think "tulpamancy-lite". Not that that's a bad thing - perhaps it's like a safer tulpamancy. Some thoughts:

(It's just such little mannerisms that allow a shoulder advisor to be "really real"—to bring it to life, give it a personality separate from, and not dependent on, your brain's main central personality. Again, I don't have a sound explanation of the mechanics, but it works.) Would it be useful to have a shoulder-advisor not constrained by having to relate to a real example? Or perhaps, without that link it will just tend to become mor

... (read more)
6[DEACTIVATED] Duncan Sabien
I have created shoulder advisors from scratch, most notably a pair of characters from a novel that I'm writing who like to show up in opposition to each other.  Not sure about any of the other points.
[+][comment deleted]40
[+][comment deleted]10