Sunday, February 9, 2014

Lob's Theorem Cured My Social Anxiety

This post explains how I cured my social anxiety in three minutes (sort of), which is a surprisingly long story. If you're just interested in the practical advice that I expect to help other people, you can skip to the section that begins, "And so it began." If you're only interested in what actually worked for me once and for all, skip to "This is where it gets seriously strange." If you're only in it for the mathematical logic jokes, skip to the very last section.

I've had something like social anxiety for as long as I can remember. I haven't always recognized it as that. For a long time I thought I just hated humans. Despite encountering some humans I actually liked over time, it got worse with age. By the time I was 20, I was having panic attacks and running off to hide in closets during social events.

I knew my goals required I be able to deal with people, so when I started college I decided to learn to socialize. I didn't have to like it, but I had to be good at it. My understanding of how to learn things wasn't very sophisticated back then, so I just threw myself into the middle of socialization. (Diving in headfirst had long been my custom.) I joined clubs, ran clubs, went dancing on the weekends, and even took a job as an RA. Although I spent much of my free time during college huddled in my room exhausted and crying, I gained many skills very quickly in order to survive the ruthless training.

That whole time, though, I didn't think of myself as having social anxiety, as being constrained by a psychological illness that could be cured. I just thought of myself as extremely introverted. It was part of my identity, more like being obsessed with books than like having a paralyzed limb. As a result, all the techniques I learned for navigating social situations assumed the constraint. I framed questions as, "Given that my brain works this way..." rather than as, "In order to make my brain work differently...".

It wasn't until I returned from my first visit to the San Francisco Bay Area that the reality of my situation hit me. I took a workshop with the Center for Applied Rationality. One of the workshop activities was called "Comfort Zone Expansion", or COZe for short, and it was basically exposure therapy. They took everyone to a crowded mall and told them to get a little outside their comfort zones. Some of the men had their makeup done, for example, and others were pushing their boundaries just by shaking hands with a few strangers.

The night beforehand, I couldn't sleep. I was already way outside my comfort zone, spending nearly every moment of every day surrounded by strangers I had to interact with in relatively unstructured ways. During dinner and other break times, I would hide in my room instead of getting to know the extremely intelligent and fascinating participants and instructors. I felt like I was on the edge of a panic attack the entire day leading up to the COZe exercise. When the time came, I simply couldn't do it. I couldn't even go and sit silently in a crowded area reading a book. The thought of being trapped with other people in a car on the way there made it hard to breathe. I stayed behind.

During the following week, I thought about all the networking opportunities I'd missed. CFAR selects their participants carefully in order to create a certain culture, and to have the largest impact they can on the rest of the world. Thus, the people at their workshops are invariably extraordinary. And I'd more or less failed to make friends with a single one of them. Without the familiar structure of academic settings, my hard-earned coping mechanisms hadn't been enough.

It was not because of my failure that this was a tipping point. I'd failed before to accomplish social goals I'd set for myself. But I'd only wanted to want to do those things, on the meta level. They seemed like a good idea, but I felt no motivation, so I wasn't surprised or really even disappointed when they didn't work out. The difference this time was that I really wanted to interact with these people, on the object level. I wanted it, but I couldn't do it.

I noticed I was confused. If the source of my social difficulties was a deep desire to not interact with other humans, then why, when that desire went away, did the problems remain?

The answer was very obvious when I finally asked myself the question with the usual self narrative out of the way. My main symptoms: Intense fear of interacting with strangers, especially in unstructured ways. Fear of situations in which I may be judged. Worrying about embarrassing or humiliating myself (mostly by looking stupid). Fear that others will notice that I look anxious. Having to fight to make eye contact. Intense fear of tests. Extremely inconveniencing myself to avoid socialization. Panic attacks that include trouble breathing, tachycardia, shaking, derealization, dissociation, and belief that I am dying. Hatred of humans does not cause things like this. But phobias do.


I struggled with this realization. I was in the middle of a massive paradigm shift that led me to consider suddenly changing course and devoting my life to existential risk reduction rather than academia - right after receiving a five year fellowship from my top choice philosophy program. That was a scary dilemma in itself, but on top of that I now understood that I had a crippling psychological disorder that I could only survive from inside the academy.

The discussion in my head went something like this.

System 1: "We've finally gotten really good at the academia thing. We're about to start getting paid to study philosophy. Charging into the chaotic outside world is completely insane!" 
System 2: "The future of humanity is probably in extreme danger, and you're proposing we do nothing about it... because we're scared. You think that's not insane?"
System 1: "Since when do we care about other people? We study logic because it's pretty, remember? Humans are ugly."
System 2: "Chapter 45 of Harry Potter and the Methods of Rationality made us cry. Lots. Given that we have social anxiety, that seems like pretty good evidence that we've been lying to ourselves about hating people to protect ourselves from having to change."
System 1: "Ok fine. Look, we specialize in an unusual kind of logic very few people study. It's really likely AI researchers will eventually need it - they'll definitely need it - and if we're the world's top expert they'll come to us, and we'll have advanced the field enough to meet their needs. So we can study philosophy and still save the world. Obviously."
System 2: "That is the worst bit of motivated reasoning we have ever attempted. What are the real odds based on our current knowledge that Friendly Artificial Intelligence requires advances in intuitionism specifically? Pretty damn small. Especially compared to the things we know it needs, like funding. Look, I just emailed the FAI guy and he agrees with me on this. Shut up and calculate."
System 1: "I don't wanna you can't make me la la la not listening. *falls on the ground and throws a fit*"
System 2: "Calm down, this is really simple. All we have to do is cure our social anxiety."
System 2: "Woah. I... think we may have found the problem. Listen. We won't want to not interact with people after we cure our social anxiety. It won't be scary. That is the point."
System 1: "Um... I... but..."
System 2: "Yes?"
System 1: "I know there's got to be something wrong with this. Just gimme a minute..." 
System 2: "*sigh* You know, to be honest, I'm not sure we could do this even if we tried."
System 1: "Hey. You take that back. We can do anything."
System 2: "No, I don't think so. We don't even have a plan."
System 1: "What the hell? Since when does that stop us?"
System 2: "I don't think we can cure social anxiety. We'll just have to hide in academia forever and never save the world, let alone achieve our full potential."
System 1: "Oh HELL no. We can totally cure social anxiety. That's not even close to impossible."
System 2: "Oh yeah? Prove it."

And so it began.

I moved into a group house/startup where self-improvement and extensive strategizing were encouraged and supported. Yes, a group house, with around twelve people, constant collaboration, and nothing but a large closet to myself due to overcrowding while the housing situation was in flux. I moved in on purpose. This should tell you something about how much awesome Leverage Research is made of. (It was also the quickest way back to the Bay Area.)

It was surprisingly non-horrible for a little while, likely because of the extremely focused, academia-like atmosphere. I was grateful to be there. But the relative calm didn’t last. The anxiety, stress, and subsequent depression compounded day after day, and my ability to solve difficult problems diminished proportionally.

But I was able to carry out parts of my developing plan. It was, after all, the perfect environment in which to study my reactions to social interaction under extreme stress. Furthermore, some of my housemates specialized in a certain kind of guided introspection that led me to form several testable hypotheses about the root cause of my condition. Through a bit of experimentation and diligent documentation, I learned more precise details of my symptoms, and disconfirmed a few plausible hypotheses. For example, it doesn't seem to be the case that I'm being constantly punished in social interactions by people's negative body language in response to subtle incorrect social signals I display.

I eventually noticed that "understand the problem" wasn't getting very far (though I was making progress on "understand what the problem isn't") and decided it was time for a different approach. For some unknown reason, System 1 was behaving strangely. Actually, it was behaving remarkably like a previously abused dog I'd recently befriended.

So I tried imagining what would happen if I treated myself the way I treated the dog. Central would be compassion, patience, and generosity. I'd engineer a safe environment for experimentation and growth. I certainly wouldn't try to force myself to behave like a normal human. I'd find ways to show myself that I wouldn't be punished for behaving unusually, but I'd reward myself quickly and copiously for taking risks toward recovery. Operant conditioning would be the name of the game. And I'd need cooperation from others.

Techniques based on that line of thinking definitely caused clear progress, and I picked up ideas from other people along the way. Here are a few things that made startling differences.
  1. I made a special effort to spend what social energy I had on people who made me feel especially comfortable, happy, and fulfilled. (Thank you Katie's dog for being so easy to anthropomorphize.)
  2. I was completely and utterly honest about my project with just about everyone. I told them I was battling social anxiety, that I'd only like to schedule the date if we agreed I would be free to cancel at any time, that I was looking uncomfortable because I was scared of social interaction and not because of anything they'd said, and that they should keep on asking me to hang out even if I said no nine times in a row because by chance they'd eventually catch me on a really good day. I explained that certain kinds of socialization are worse for me than others, and that I'd respond better to proposals of goal-directed meetings than to proposals of free-form hangouts. Rather than indefinitely dodging their phone calls, I told them I have a strong preference for meeting in person or chatting through text. This hugely mitigated my fear that others would take my symptoms personally. (Thanks Mike B, Alexei, Leveragers, and everyone else who took me at my word in these situations.) YAY TELL CULTURE.
  3. I installed a habit of imagining a version of myself that wasn’t afraid whenever I needed to make an important policy decision, and I counted on my simulation of her to reason sensibly when I couldn’t. I predicted her actions and followed suit rather than deciding whether to socialize (thanks, Anna). Deciding, it turns out, automatically engages the affect heuristic in a way that predicting does not.
  4. At the suggestion of a book on cognitive behavioral therapy, I almost completely cut out caffeine, prioritized sufficient sleep, and replaced part of my usual meditation with progressive relaxation. This dramatically reduced the frequency and severity of full-blown, spontaneous panic attacks (which are different from the anxiety feedback loops described below).
  5. I installed a habit of distancing myself from my emotional reactions whenever I noticed that they were excessive or forming dangerous feedback loops.

    An anxiety feedback loop looks something like this: I would interpret a stimulus as anxiety inducing, which would cause anxiety. Then I’d take that experience of anxiety as evidence that the original stimulus was in fact anxiety inducing. My confidence that I would feel more anxious increased, and the anxiety itself increased in turn. The fear of greater panic also added epicycles on top of this process. It could only escalate so far since my attempts to calm down were eventually effective, but it could sustain itself for an hour or more at whatever level I reached before the calming effect kicked in.

    Distancing had long been in my toolkit, but somehow it had never occurred to me to apply it to this kind of experience (thanks, Val). I originally learned it while living at a Soto Zen temple, where meditation sessions are long and frequent. When you first begin a meditation practice, your muscles and joints are not prepared. It can be extremely painful early on if you sit for, say, an hour and a half at a time every single day. (I actually began with a week-long traditional Soto retreat, so make that four to six hours a day). The only way to get through it is to let the pain happen without suffering from it—without “attaching” to it, as Buddhists would say. You assume a mental posture that turns “I am hurting” into “there exists pain”. Distancing is the opposite of attachment or identification.

    With the application of distancing to social responses, I gained the incredibly satisfying ability to stop sudden panic cycles in their tracks a majority of the time. Watching a panic reaction as an outside observer breaks the connection between “evidence of anxiety” and “I am feeling anxious”. It didn't immediately end the panic, because my brain was still flooded with the first spike of stress hormones. But the physical response couldn't sustain itself without emotional engagement, so I could just ride out the aftereffects. There was a racing heart, a flash of heat, and an impulse to run and hide. But none of it was mine. None of it was me.

    From there, my mind could consider alternative hypotheses about the other person’s motivations, because I wasn’t busy engaging with the panic. Usually, some other mental state was obviously more likely, upon reflection, to have caused their behavior than whatever perceived state triggered my anxiety. So besides causing less suffering, the new freedom for my beliefs to grow more accurate made my interactions more effective.

    Distancing didn't do much for the constant low-grade anxiety, but it was a clear improvement nonetheless.
Each of the above techniques caused marginal improvement. Each made life just a little better. Even with all of them together, my phobia was still crippling. I’d solved about 15% of the problem, and I was running out of low-hanging fruit.

It turned out to be the process of solving that 15% that really mattered. Every new successful technique fed a much larger success spiral. The gradual discovery of one after another replaced the trapped and helpless feeling with powerful confidence in my ability to conquer my weaknesses, to do apparently impossible things, and to domain-generally self-modify.


This is where it gets seriously strange and awesome. But first, you’ll need a little background on hypnosis.

I’ve been playing with hypnosis recreationally for a few years. This isn't the place for details on that, because it’s mostly about sex and I don’t want to distract either of us. You’re welcome to ask me about it privately, or to try convincing me to write about it elsewhere. Anyway.

The relevant point is that I’ve dabbled as both a hypnotist and a subject (though much more the latter than the former). I therefore have considerably stronger priors for the reliability of hypnotic effects than mere academic research would justify given the current state of science on the matter (which is abysmal).

My bottom-up, gradual improvement approach to overcoming social anxiety wasn’t moving quickly enough (according to my standards). When I asked myself, “How can I cheat?” hypnosis was the most obvious thing to reach for. Why slowly shape through operant conditioning when you can access unconscious processes directly?

How exactly to use it, though, was not so obvious. I puzzled over that for at least a week, worrying that I might have to understand the root cause after all to devise a workable plan.

Then I encountered the Miracle Question. The Miracle Question goes like this. “Imagine that there’s a miracle overnight, and you wake up tomorrow morning to find that your problem has magically disappeared. What is the very first thing you encounter that is evidence of the change?”

For me, the answer was, “I think about a potential future social interaction, and I don’t feel anxious.” Even for extremely familiar interactions, there was always at least a tiny bit of anxiety. For example, I noticed at one point that I was consistently careless about cleaning things up in the kitchen because I knew that my housemate could walk in at any time, so I wanted to leave the communal space quickly. The first evidence of the Miracle would probably be a lack of anxiety on that level.

So I thought I might as well use that as a starting point. I played through the following strategy in my mind. First, I’d have my hypnotist friend put me very deeply into trance. He’d set up a clear trigger for “relax, calm, untroubled”. Then he’d have me begin to think about a social interaction. The moment I noticed the slightest hint of anxiety, I’d indicate that and he’d give me my “calm” trigger, causing me to feel completely untroubled. We would keep doing this until I could imagine social interactions while remaining calm, possibly over several sessions. Finally, he’d give me access to the “calm” trigger as a post hypnotic suggestion, so that I could activate it at the first sign of anxiety.

Note that I spent about three minutes developing this plan, and I was in my mental state for “creative problem solving” the whole time, which involves intense inward focus and devoting extra resources to my imagination. That’s probably important.

During the conversation in which I described my plan to him, we meandered to the topic of a meetup of professional hypnotists he’d recently attended. He told me they talked in passing about what it’s like to change their own behaviors. They all knew they could use a long, draw-out induction (or series of inductions and post-hypnotic suggestions) to self-modify if they wanted. But that takes time and energy, and it turns out that if you’re sufficiently confident it’ll work… you don’t have to bother with the hypnosis.

Think about that for a minute. They treated it as a perfectly normal, every-day occurrence. Basically they were saying, “Yeah, when I don’t like what System 1 is doing, I just tell it to do something else instead. No biggy.” They seem to have this available as a primitive action.

Initially, I said it sort of tongue-in-cheek: “Ha, well I guess we don’t really need that induction I described then!”

System 2: Surely not. It can't be that simple. There’s just no way that will actually work. Nobody cures a life-long psychological disorder overnight. Don’t be ridiculous.
System 1: But it would be the best cheat code that ever happened. We have to try it. Pleeeeease?
System 2: …I guess it doesn’t really cost much. We just have to put off the explicit induction plan for a few more days. It might reduce our confidence in the longer term plan slightly, but not nearly enough to compete with the VOI we’re talking about here. Are you really really sure the explicit induction plan would work if we went through with it?
System 1: YES DEFINITELY. That’s exactly the kind of thing hypnosis can do given enough time. Plus, have you SEEN all the kick-ass self-modification we've been pulling off lately? I told you we could do anything, remember? You said, "Prove it." So let me do that.
System 2: That's a good point. We are in the middle of a success spiral. What the hell, let’s give it a try.
My friend agreed to wait. I’d watch for anxiety to hit, then snap my fingers as though the trigger already existed. That was the idea, anyway. I’m not sure how seriously he took my hypothesis. I’m not sure how seriously I took it. I suppose part of me must have been totally serious.

I went home, prepared for bed, and went to sleep. When I woke up, I remembered that I’d been invited to a dinner party that night. Perfect opportunity to test it. I waited for the first jolt of panic, fingers poised to snap, pleasantly excited by my curiosity even as I braced for the impact -

- but nothing happened.

There was no jolt of panic.

I kept waiting. I imagined going to the dinner party. I even imagined scenarios in which embarrassing things happened and everyone thought I was stupid and everything went horribly wrong. I reminded myself that the dinner party really was happening, and it really was tonight, and I really did have to go to it. Some of those awful scenarios were even plausible.


My observations strongly contradicted my model of the world. Psychology just doesn't work that way. I purposefully scheduled several historically uncomfortable types of social engagements throughout the week, trying to break whatever weird and presumably temporary coincidence was happening. I at least wanted to be able to test the trigger.

That was three months ago. I'm still waiting.


[Trigger warning for this section: Abstract math/logic concepts with virtually no explanation.]

I've thought a fair amount about how the hell I did what I did. It still seems completely crazy. I don't really understand it, but I have a favorite hypothesis.

Löb's theorem states that "If it's provable that (if it's provable that p then p), then it's provable that p." In addition to being a theorem of set theory with Peano arithmetic, it's also a theorem of modal logic. (There's a modal proof here.) 

A standard semantic framework for modal logic is epistemic logic, where provability here is just replaced by "knowledge" or "belief", and "belief" is defined in terms of possible worlds, so that you "believe" something if and only if there's no world accessible from your perspective in which the thing is false.

This is basically what's going on with placebos. (By the way, placebos work even when you know they're placebos.) 

Try this on for size: If I believe that (if I believe that this chocolate chip will cure my headache, then this chocolate chip will cure my headache), then I believe that this chocolate chip will cure my headache. 

Do you believe in the placebo effect? Do you really believe that believing that something ingested can cure a headache actually causes the headache to get better? If you do and you're right, then by Löb's theorem, you can now cure headaches with chocolate chips.

I know it sounds like a joke, but it really does work. I use this all the time now. For instance, suppose I have a meeting at 7:15 and I fear the planning fallacy. I just think, "If I believe that (if I believe I believe that the meeting's at seven, then I believe that the meeting's at seven) then I believe I believe that the meeting's at seven." (Easier said than done, maybe, but you get the hang of this particular convolution after a while.) Fortunately, it's actually true that if I believe I believe something, then I probably straight up believe it. Subsequently, I get to the meeting at 7:05 believing I'm late, and am relieved to discover that I'm actually ten minutes early. This is real. It's just too ridiculous for me not to laugh at it, even though it's clearly part of reality.

Now compare this to the social anxiety cure I described. "If I'm hypnotized such that (if I'm hypnotized such that I'm not socially anxious, then I'm not socially anxious) then I'm hypnotized such that I'm not socially anxious." So if it happens to be true that being hypnotized such that one isn't social anxious is sufficient for not being socially anxious (as I indeed believed wholeheartedly), then if hypnosis can be modeled similarly to doxastic phenomena, my instant anxiety cure is an instance of Löb's theorem.

(Please insert your favorite evil laughter here. Alternately, THIS IS SPARTAAAAAAA!!! But for realizies, like... woah.)

I recognize that probabilistic beliefs complicate this picture. I don't know whether probabilistic logics have a correlate of Löb's theorem. Dynamic doxastic Baysian systems, anyone? I'm afraid that's still over my head at the moment. But I take this as (very) weak evidence that they do.