my chat with the beeminder guy

I had the unexpected privilege of a long Facebook chat with Daniel Reeves, cofounder of Beeminder. I had attracted his attention by badmouthing the Beeminder customer service, which was totally uncalled for.

As you will see below they are actually total sweethearts in the Beehive and I don’t really know what I was thinking with that comment.

Anyway, Daniel spends basically his whole time thinking about the sort of thing I’ve been writing about in this blog, and helping cook up solutions. His product might not have worked for me, but he really knows what he’s talking about and is super interested in learning about different ideas and perspectives as you’ll see.

Somewhat edited and posted with permission, obviously.

Me (original blog post): When Beeminder fell apart for me it cost quite a lot of money. Needless to say, they weren’t terribly sympathetic about it either.

DR (reply on blog): Eek, can you say more about that? That doesn’t sound like us. I mean, I guess in one way it does. We’re pretty serious about the commitment device aspect — it would defeat the point if you could change your mind and get out of the commitment. Like you said, you’re setting up a program with yourself as the slave who has to faithfully execute it. On the other hand we have a whole list of safeguards to minimize any regret:

  1. It’s free at first and the amounts at risk go up gently (at first) and you can cap them at whatever amount of money you find sufficiently motivating.
  2. You can change your mind and end a goal or change how much you’re committing to do with one week notice at any time. So if you do change your mind you just have to stick it out for another week. (Or less if you can manage to maintain some safety buffer on your graph.)
  3. We’ll never charge you if there’s any kind of technicality, our fault or not, that led to the derailment. We make it super easy to cry foul if something does go wrong.
  4. There’s even a deadman switch so we won’t just keep charging you forever if you go completely AWOL (or die).

Me (reply on blog): Oh yeah! Beeminder and their staff totally aren’t like I just made them sound, and I’m really sorry about that. A lot of happy people use Beeminder and is very customer-focused. I definitely got interested in it because it was both run and used by people a lot like me. Points 1-4 are all completely accurate.

They didn’t unfortunately quite cover what happened with me, and I can explain why. Points 3 & 4 are out, because it wasn’t a technicality and I wasn’t awol. I had genuinely fucked up in the exact way I was trying to incentivize myself not to. Point 1 applied in the sense that the amounts involved in each goal were fairly small – I don’t remember how much I got burned by in the end, it was enough to really sting and make me not want to use it again but nowhere near the amounts of money I’ve lost on stupidity in other ways.

Point 2 didn’t apply because somehow I didn’t realize what was happening until it was already too late. I had got to the point where I had lost any intention of following through with any of my goals, but somehow imagined the tomorrow version of me would get it back, or something. If I could somehow sense a crash was coming, and not lie to myself about that, then I could have scheduled the canceling of all the goals and spent the next week trying to keep the graphs above water as best I could. As it was, I had to panic-cancel them all and watch that last week play out in pure frustration.

[we say hi over Facebook and decide to continue the discussion there]

DR: I was going to say that I think this failure mode is *pretty* rare but also probably more common than we’re aware of — see — so it’s really good for us to really understand the psychology of that failure mode

Me: I was gonna bring that up, like I could have told you how I felt at any point but didn’t, so I’m one of the 90% of cockroaches running around in the hidden places behind the sink.

Or, was.

Somewhat embarrassed to look over old emails, it looks like it happened in April 2015 though.

Me [quoting old email from DR] “Hi Giles, I can’t tell the intonation of the “seriously?” but either way it sounds like there’s hugely valuable feedback behind it. Can we hear more? Anything non-legit at all about the derailment? [1] Reply tonight or in the morning if possible so we have time to stop the charge! Really sorry about the frustration, whether beeminder-related or not (and if so, again, we’ll be so grateful to learn about how/why/what!)” <– that was you and it certainly doesn’t sound like the “not terribly     sympathetic” that I was accusing you of in the blog post.

DR: We’re certainly grateful for that.

So, yeah, almost 2 years ago

Me: Anything that I said so far didn’t seem to make sense? Or, like, really atypical of beeminders?

DR: Nothing unclear or blatantly atypical… Do you still have instances of what could be characterized as egregiously akratic behavior?

Me: Egregiously akratic? yeah certainly.

DR: So there’s an interesting forum discussion going on right now — at the very end of that thread I make a claim about the strongest argument against beeminder. I’m curious if that at all captures the reason you feel it doesn’t work for you. about being truly incorrigible, and maybe about the want-can-will test not applying

Me: I am not asserting that beeminder could never work for me btw, just a lot of other things I’d want to try before giving it another go plus I feel like I’m somehow missing a mental cog that beeminder needs, even if I can’t quite articulate what it might be.

Reading the thread now. My first thought is obviously that I don’t care if I cheat at     willpower, I just want stuff to get done somehow.

The “staying well above the road” trick is a good point (in that forum thread)

[I post a reply on the Beeminder forum]

Me: Also, I totally didn’t read or respond to the want-can-will, I will look at that now.

Yeah I don’t remember exactly how the goals I put into beeminder were defined, but I think they were mostly in the want-can-won’t category.

I don’t remember feeling I physically wasn’t up to [one of my goals], or that it somehow no longer seemed a good idea to be doing that.

DR: It sounds like the crash involved like an overall rationality failure?

Me: Yeah, sure.

DR: Like beyond just akrasia?

Me: You mean like an epistemic rationality fail? I pretty much view akrasia as constant rationality failure in any case.

DR: That might be a key assumption we’re making, that you’ll always respond rationally at least to immediate incentives if not to distant incentives.

So akrasia is irrationality wrt to longer timescales.

Me: right, and I don’t think that I do always respond that way with money.

Like I can just not pay a bill even though there’s nothing stopping me and it can’t be     described as inconvenient in any meaningful way, and I know they will start making me pay more if I don’t. There’s one sitting there now. I should probably take a look at that.

See that took like 90 seconds on online banking.

DR: I do that a lot. but in that case I view it as “tomorrow is no worse than today and I don’t feel like doing it *today*”

a “one more day won’t matter” slippery slope of sloth.

Me: …And yet I often don’t do it. Whatever’s going on there, it’s something a little weird. I remember Luke Muehlhauser writing an “akrasia equation” that involve this plus this divided by that plus that, and thinking it totally didn’t help me at all. Probably some people’s akrasia is like that, but it order to explain not paying a bill that is totally     convenient and doesn’t even make me feel bad, you’d need to make one of the numbers so ridiculously large it would imply I’d be a complete vegetable or something.

DR: …Cuz i’ll repeat that reasoning every day ad infinitum. and each individual day it’s technically true. but today vs next month has huge negative consequences

Me: I sort of buy into your “one more day won’t matter argument” though when I acquire sufficient gumption to actually write a to-do list that I have some intention of following through with, I usually mark as most urgent the things which are technically not urgent by that reasoning but that I have already put off for too long.

i.e. I am consciously aware that I do that, and that at least seems like it ought to be somewhat self-correcting.

I wouldn’t ever walk away from that line of reasoning feeling good about it, unless there really was something more urgent that came up in the meantime.

DR: That all sounds sane. Saner than I am about to-do lists, I think.

Me: Well. It doesn’t mean they get done.

Usually a small amount of progress gets made, enough to just about keep my head     above water on the genuinely urgent things, and then I have one of the cyclic crashes that I was talking about.

There is at least one other factor involved. I’m pretty sure the things I put off each have a shade of “there is some much greater and more complex and ughy thing underlying this that also needs attention at some point”


internet bill: I could probably be paying much less for internet if I shop around properly, which is something I totally hate. It doesn’t affect whether or not this particular bill should be paid, but I’m reminded of it when I open it and see the number.

or: I nearly got into trouble over my Canadian work permit, which could have been avoided if I’d talked to the right people sooner. The problem was, the right people were my employer, and at that time I was having really quite a hard time at work and was questioning whether I really wanted to stay with them.

DR: All these things are super familiar for me.

I think it’s the more severe rationality breakdown you seem to describe that I don’t fully understand yet. Like beeminder is sending notifications saying “you have to do 1 more to-do by 5pm or pay $90” and (consciously?) deciding not to.

Me: Hmm, let me think about that.

DR: That’s the difference (for me) with paying bills or doing the market research or talking to your employer — there’s no crisp deadline. so i just procrastinate indefinitely until I’ve dug myself in a horrible hole.

Me: Right. So… not responding correctly to the eep reminders, it’s not about not having a crisp deadline and it’s not about the other thing I mentioned of it opening up some other can of general psychological turmoil, because that didn’t really apply to any of the goals I set there.

So I need to think of some other excuse.

I mentioned a sort of self-destructivity in the forum post, but I’m trying to think how to unpack that a little

Like… did I feel I deserved the punishment? I’m not sure that was what happened or if it even makes sense.

DR: (oh, yes, I’ve heard the idea of a self-destructive streak before. […])

Me: Ok, if it’s something you’ve never really felt […] I can try and explain what it’s like.

(also, other people probably experience it a lot more severely than I do, and whatever I     have to say may not generalize in other ways either)

Well, this gives a quick intro but probably isn’t enough for you.

Somewhat like akrasia, but with the time dimension very different. If I’m akratic, e.g. not exercising, I know that I’m harming my future self but somehow am able to make myself not care about that. It’s not that I don’t care about future me at all, it’s that if I picture future me along certain dimensions then I only care about some of the dimensions. I care that he’s still in employment and a member of the EA community but not that he has health problems. Apparently?

Self-destructive urges are sort of that but with the time dimension collapsed into the immediate future. Somehow I discount how annoying something is actually going to feel, even though it’s going to happen to me next?

But I’m still able to picture other aspects of the future, like oh hey I won’t be worrying about this beeminder goal any more because I’ll have already failed it. Or something like that.

On the other hand it’s easy to make up psychology like this so I would take this with a grain of salt. I mean I’m sure I don’t need to tell you that.

DR: mulling this

Maybe it’s as simple as different people having something like akrasia at different timescales.

Me: Yeah not sure if that is right. Akrasia seems very complex to me. Some things work great against it for some people and don’t work at all for others. (Although most things that have to do with the brain are like that)

Thinking of it as ignoring one dimension of future-you’s utility function is maybe only one of the things going on with akrasia. And probably only one of the things going on with self-destruction also.

And self-destruction is probably only a small part of what was going on with me back then with the beeminders anyway, so wow am I totally not generating the insights tonight.

DR: Your forum post sounds important. I’m still thinking about that, and how to reply.

Me: …also as with most problems with my own psychology I feel like if I really had a good idea what was going on I could probably just fix it.

Lol sounding important wasn’t quite my goal.

DR: I mean important for us to understand. Or an important caveat for a set of potential beeminder users.

Me: Well. n=1

DR: I’d like to figure out how to turn it into something more concrete, like “don’t use beeminder if…”

Me: yeah. Your attempts to figure out people’s psychology and work with it really seem very impressive. It’s like you guys read the entire literature on cognitive biases and then just turned whatever they said into a business model or something?

And I know there’s a balance between trying to make sure the basically happy beeminderers carry on having a great time and figuring out what’s going on with the weird ones that don’t quite fit.

DR: we try! not sure how much of the blog you’ve read but we’ve tagged the things we think are sufficiently rationality-relevant.

Me: still curious how common what I’m saying is. I mean, people must leave beeminder right? And have reasons?

DR: Yeah, it’s rare to get a good postmortem though! and I think almost always it’s a form of normal-style akrasia. Archiving stuff because of life changes and meaning to but failing to resume when things settle down. That sort of thing.

Sometimes people actually say they beemind something and it becomes sufficiently a habit that they don’t need beeminder anymore.

Me: Yeah ok. Those two exit modes totally make sense.

DR: [describes someone crashing out of Beeminder and then coming back]

So I never really understood the self-destructive aspect or why it worked better the second time around.

Me: Right. Well there can be a big interplay of factors that means sometimes it ends up just working and sometimes just not working, even if only some small part of it changed. That aspect doesn’t seem a mystery to me.

Like maybe I could start beeminding tomorrow and it would all go great forever.

It’s not worth the risk for me just yet but there seems at least a 20-25% chance that     could happen. I don’t understand what went wrong, not completely, so I can’t say for certain it wouldn’t have fixed itself.

DR: Maybe another failure mode, which actually sounds like what you described in the forum post, is a kind of catch-22 where the small amounts of money are just not motivating but any larger and they’re too stressful and scary, and there’s just no sweet spot in between.

Me: Oh yeah, I think you might be getting somewhere with that one.

Another thing: I mentioned elsewhere on my blog that I recently started experiencing things I’d normally associate as “akrasia” as actual anxiety.

I still haven’t figured out what’s going on there, but one hypothesis is that I was always more anxious than I realized and was somehow blocking that knowledge from myself.

If I had dialed up beeminder into the scary amounts of money then it wouldn’t play well with any anxiety, because whenever it got to the motivating point it would be actual panic.

I assume you’ve seen the cartoon with the immediate gratification monkey and the panic monster.

DR: Yeah, waitbutwhy.

Me: Yep it is a classic although I don’t think it completely categorizes akrasia correctly and I probably even complained about that fact on my blog somewhere.

DR: (oh, I didn’t mention the biggest failure mode of all which is just the whole site     being too confusing and the interface not being friendly or friction-free enough. That’s our main hurdle to growing at this point.)

Me: huh. I would have not expected that to be the one. I guess I am just way to used to arcane software by now. + math degree.

DR: Yeah, most of our users are *huge* nerds

Me: There were actually ux problems that had been tripping me up a few times     previously. [describes one]

DR: it’s going to be super embarrassing if you try again and the same issues are still unfixed…

(do tell us if so! hopefully the shame would light a fire under our butts. there are actually plenty of things we’ve known are sucky for ages that we still haven’t fixed

Me: Hah. You’re not luring me back quite that easily.

The next productivity hack I want to try again is actually Complice. Also something I’ve tried before and didn’t get to work but less scary.

DR: []

Me: One quote on that page resonates with me, it’s from Nick Winter. “…the real cost of failing a goal is not the loss of your Beeminder pledge money. It’s the loss of confidence that you will meet all future goals that you perceive as similar to the current goal. You will trust yourself less.”

I had already derailed on all the goals at least once I think, and so trusted the system less. It seemed to be working better right at the start when I hadn’t had a chance to derail yet and really felt it was possible to beat it.

DR: That reminds me of another forum discussion sparked by another blog post [Sinceriously].

Sounds different than your arguments, and that person I don’t think was arguing from direct experience with beeminder, if I recall.

Me: …One other thought: I remember I didn’t feel sufficient pressure to “escape” beeminder’s danger area. I get the vague impression there are two ways to use beeminder: the “panic monster” where you let it constantly be red or orange with a scarily high amount of money pledged, and spend your days constantly fighting those fires and are somehow able to make that work. I’ve said that’s not me because of cowardice and anxiety, but also because for whatever reason the threat only seems to be effective with say 80% probability. When I stayed in the danger zone too long I would keep crossing over.

The other way to use it is the Virtuous Bee, where you learn to apply enough willpower to keep it safe most of the time (while still keeping the slope of the graph reasonably aggressive). It will go red or orange occasionally, but when it does these people are able to not just save it day by day but actually apply significantly *more* than the average and get back into safety. I definitely remember not wanting to apply more than the average, and doing the literal bare minimum to avoid paying one day even though I knew that would mean a strictly greater amount of panic the next day

DR: Ah yeah, similar dichotomy posed here.

Me: It is related but my panic monsters vs. virtue bee axis is at a slight angle to black vs. yellow. (You can tell panic vs. virtue just by looking at the numbers – what percentage of the time are you in the red or orange – whereas the black vs. yellow is more about attitude and can’t be directly measured, you just have to talk to people to find out).

I also wouldn’t describe it as a dichotomy. It’s more like there are actually two separate forces that cause Beeminder to work. The first is the short-term emergency: I need to do some of this right now, else I pay. You can conceptually give each person a score on this, which is basically the probability that in that scenario they would actually do the right thing rather than giving up and paying anyway.

And then there’s the other force, the escape velocity. This would basically be measured as: on a good day, how much do you exceed the slope of the target graph? If you score well on the escape velocity (regardless of how you score on the emergency situations) then you will be a virtue bee. If you score badly on escape velocity but well on emergencies then you will be a panic monster. I score badly on both and end up crashing.

And obviously in this model, these two factors are not some fixed part of your personality but can vary according to any number of external or internal factors, which could lead to people switching from one pattern of behaviour to the other, or doing ok and then suddenly crashing like I did.

DR: Still thinking about all this… I think I’m definitely a panic monster myself…

Me: I just read the Sinceriously post from the forum thread you linked me to. […] I’ll add my thoughts.

The original post from Sinceriously seems very genuine but not completely coherently argued. The crux of the argument is in paragraph 4 “it may be because you only…” which contains the word “may”, so I read the whole rest of it feeling like “oh so we’re only supposed to use beeminder for the other 50% of things then?”

Choosing the right goals and projects in the broader context of your own values and identity is super important though. It’s plausible that you guys might not emphasize it enough. Generally you’re pretty good in all your blog posts at talking about both the broader context of things and also your actual direct competitors, which is pretty good going for something that is still, basically, a business, so I wouldn’t worry about that one too much.

[I express some concern about using “revealed preference” as the be-all-and-end-all of what it is that you actually want]

On the other hand, “It’s better to limp in the right direction than run in the wrong one.” pretty much describes effective altruism for me perfectly.

Something else relevant to that [from my same previous post]. It has to do with allowing some trade between your near- and far-mode personas (although I didn’t put it quite like that). The context was not commitment contracts but simply committing to some kind of strategy for getting icky tasks done. My idea was not to make the system totally rigid, but allow the present self to make up some of the rules based on how it was feeling, with the idea that those feelings carried useful information. Not sure if it applies to Beeminder or to what I was saying about it, though. Obviously the point of commitment contracts is not to allow room for wiggle.

The point about using Beeminder to get information about yourself is super interesting.

mattepp’s comment sounds interesting because it sounds like me except I didn’t come back or mature or anything. hmm.

[a while later]

Me: Oh… I literally only just remembered that I was lying to beeminder with some of my numbers as well. Can’t believe I forgot about that. Seems like it might be relevant but will need to think about that.

DR: yeah, that’s an incredibly slippery slope that commonly destroys beeminder’s efficacy for people. Can’t believe I forgot to mention that when talking about ways beeminder fails for people. we have a couple blog post (of course) about this: and


3 thoughts on “my chat with the beeminder guy

  1. >oh so we’re only supposed to use beeminder for the other 50% of things then?

    In my experience, doing in some sense “your best” to find the X% of things you actually don’t want to do so you can stop doing them, and having suspiciously looked over the 100%-X% as part of that effort with sincere intent to stop doing them if they were, makes commitment mechanisms unnecessary.

    There are a huge number of subtle ways to screw it up though. “your best” means incorporating all the information contained in a bunch of little inner-sim runs. If part of you feels like something will work, or won’t really be that bad, it means the kind of state of intention which causes you to direct attention to that and expand that piece of belief as much as you can. To search into and through it because you are searching for the thing that that part of you executed that inner-sim run as a way of getting at. Search into and through contradictions between different pieces of belief as represented both explicitly and in inner-sim runs because of intent to get at the thing your brain just wants to get at. Going through the motions of taking that information seriously so you can “stop being akratic and do the thing you know you should” does not count.

    The long-term arc of my blog is that I’m trying to explain the thing that CFAR’s internal double crux technique is about, with the extra knowledge and speculative models I’ve gained in the course of actually making it work. I make no promises you can get it to work. I am pretty confident you could *in theory*, but I haven’t even laid out everything because…

    * All my concepts and techniques have way more dependencies than I can see at first glance
    * Some things have possibly become circular dependencies since they reinforce each other and I’m not sure how I bootstrapped into them.
    * Even when I iterate drafts on this topic several times with test readers I still get people coming away from posts on this topic saying it was profound and a really good point and I ask them what they got from it and they misinterpreted it as something completely different than what I meant.
    * Communication about goings-on in minds are very hard. Porting software between minds is even harder.
    * A big part of it is I need to explain an approach to metaethics that subsumes moral motivation into the “wanting” framework without it sounding like something that’s “evil”, and test reader reports indicate this is hard.
    * I don’t remember why I believe everything I believe.
    * Many of my best datapoints are private.

    Liked by 1 person

    1. OK, that’s really interesting! There is so much to this I hadn’t guessed from just glancing at one page of your blog and I *definitely* need to look at the rest now. Interested in how I can gently ask my brain why it doesn’t want to do the things that seem like they should be the right things to do. (And yes, contrasting this to the brute force approach of commitment contracts is valid: since I’ve struggled and failed with those myself I want to know what other kinds of approach are out there as well).

      Really interested in internal double cruxing. It didn’t exist when I attended a CFAR workshop, and I haven’t even had the opportunity to properly try regular double cruxing (i.e. with a debating partner). People mention internal double cruxing and the idea baffles me, which is why it’s interesting. If that’s a theme of your blog then I’m super looking forward to it.

      Other stuff here resonates too like the idea of using a wanting framework for ethics, without the conversation descending into morality=egoism=selfishness=bad. I’ll want to think about this a bit though and read your other stuff to put what you’re saying here in the proper context.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s