Akrasia Is Mainly Error Assignment Error
Akrasia is a long-standing unsolved problem: Why is it so difficult to get yourself to do what you plan to do? It seems like a strange thing, given that what we plan to do is often what we really wish we would do, and not merely what we signal to others.
It is not immediately obvious how bad one's experience of akrasia is relative to others. After all, human civilization carries on, all made up of people who suffer akrasia. But we can know how bad akrasia is relative to our own expectations, and that is usually how akrasia is defined, anyway.
Immediately from this we can deduce that akrasia may be something of an expectation problem. But obviously, it seems unfair that we don't get to be as ambitious as we want to be! What is not obvious, and what this piece will mainly be about, is that having more reasonable expectations will actually increase performance, thus enabling one to still remain as ambitious as they want to be. Your expectations about the final outcome get to remain the same.
It may not be obvious to you that it's fine for you to think "it's fine that I did not start my project today, I'll probably start it soon." It may not be obvious that it's okay to think, "I'm not sure how many hours I will work today. In expectation, probably somewhat more than I have on average the past several weeks."
It's probably not obvious that it's okay to think those things because the word "probably" in the context in which it is used may provoke an association in our minds to procrastination, which carries negative connotations. In fact, akrasia is often considered to be almost synonymous with procrastination. This, I suspect, is one of the main reasons this solution is difficult to see.
It all comes down to how you label the outcomes of your actions. If things fall below your expectations, do you declare that a failure? You should pay very close attention to what happens when you do this. If you declare an action a failure, it drastically lowers the probability you will perform that action again, and increases the probability you will perform an action that is very different from the old, penalized action. So you should only do this if you are sure that this is the correct action. But how can you be sure that penalizing an action you took is correct, given there remains so much uncertainty in the choice of the action?
I advise caution with taking actions that are as certain as penalizing an action because the outcome fell below your expectations. If your expectations could have been in error, then that lowers the probability that your action fell far enough below an arbitrary threshold that it warrants taking a step back and contemplating a completely new action.
If both your expectations and the action were in error, then it is more likely that the action taken was close to an action that would have met a possible expectation.
The human body is more like a ship, and your mind like the captain, than the whole body united as one. You can give orders to pieces of it, which then have a tendency to try and do things their own way. This, by itself, should not cause us much alarm, as it is the default situation. Our alarm may come from comparisons: We expect to do as well as what we see around us. However, I argue we have little to fear here, either.
One must understand that one can't completely carry over every single observation they make about every person they see word for word. Many of those things will consist of statements or images made by other people. In fact, the only observations they should carry over are measurable outcomes. I highly advise actually doing the work to determine what the measurable outcomes truly are, and placing yourself in the correct reference class. Then, be ambitious.
Akrasia itself is often described as the inability to do what one should do: That is, what one should do is determined first, set in stone, and then actions are carried out. When those actions fail to meet their target, the actions are deemed a failure. But you'll notice that what one "should" do is typically not assigned any error value whatsoever.
That may be because it is somewhat difficult to change what one "should" do. That being said, what you imagine yourself doing in the future is tied to what you want to do. What you want to do and what you should do are also tied (and I would argue, of course, that these are supposed to be the same).
I am going to request of the reader that they at least take it as a given that what they want to do and what they should to do might overlap considerably in ways that are difficult to disentangle. From that, it's quite a bit easier to deduce the method that I'm proposing. At least try this frame for a while and observe the effects it has (I think you'll like it a lot, actually).
Here is a somewhat recent list of potential solutions from LessWrong:
I don't think the solution to akrasia is going to consist of a "bag-of-tricks" in an ultimate sense, but feel free to try any you come across. Discouraging you from doing so would be against the spirit of this post. Ones you like and find useful might contain hints in them which point to a more general solution. I expect that if my solution is close to the general solution, then such tricks would make sense in the context of what I'm saying.
LessWrong also notes that two main approaches are willpower control and resolving internal disagreements. I think it’s probably possible to see the latter approach in terms of what I’m suggesting here and vice-versa. Though, I would add that disagreements in your whole self might consist of verbalized beliefs on the one hand, and subconscious or learned tendencies on the other hand. We have a lot more immediate, direct control over the former, but more slow, incremental, and probabilistic control over the latter.
But to be frank, I tend to think you have plenty of willpower already.
Why should I feel so confident that this approach is the “general solution” I’m looking for?
I think a “general solution” to akrasia would have to have at least the following properties:
To be a “solution” it has to make the problem disappear in either the sense that it shows it to be an illusion, or that it actually measurably results in an improvement according to the original problem’s parameters. (Ours is a “mysterious third thing” approach).
Explains what creates our sense of the problem to begin with, even if we do chalk it up to being an illusion, in the end.
Cannot resort to “bag-of-tricks”, or highly-specific algorithms for one to follow, as that would neither be general nor a likely reason for why some people do seem to experience akrasia and some do not.
We might posit that it’s possible to come across this solution via experience alone, perhaps unconsciously (as in not explicitly as a solution to this problem, per se).
Frames the problem in terms similar to reinforcement learning. And, does not resort to necessitation of externalized reward / punishment that is outside of your control.
We first start with the assumption that our solution cannot entirely be an illusion (I can’t simply choose to be happy with everything I am at exactly this moment and expect myself to remain that way), but that it may partially be.
I can choose (with my free will, to a large extent) to imagine what I want to be, and also intermediate states. I can also choose to consider myself in various degrees of “succeeding”, “not succeeding”, “on the way”, or “not even on the way.” It’s actually easier to change what you think the intermediate states should look like rather than the terminal states. I’m not asking you to change what you think the terminal states should be (your ultimate goals), which may not be possible anyway.
I’m mainly asking you to change what you consider to be “failing”, and set that label equal to “on the way” instead. Furthermore, I’m also asking you to consider “on the way” better at pretty much every step than the previous step.
One common theme I have mentioned in previous self-help-ish posts, which I will mention again here: The ideal strategy / solution should feel the most comfortable, even just to use, over the long-term period in which you have it.
One fact that may actually be difficult to realize is that your brain will automatically feel a positive feeling when it has made a correct prediction, and a negative one when it has made a wrong prediction (this part you don’t get to choose). So, in terms of reinforcement learning, you do get to choose your predictions, but not what you feel when you observe the outcome. Thus, positive reward signal is maximized not by trying to arbitrarily feel a certain way on any observed outcome, but by making predictions that are - in some sense - “healthily optimistic.”
In other words, you have to be quite kind to yourself. It feels a lot like being really kind to yourself, because it involves a lot of, “wow, good job doing incrementally more than you did yesterday! I bet you’ll continue to do at least the same thing or incrementally more tomorrow.”
In fact, I think this is sufficient. I don’t think you need to make commitments like “I will give myself extra ice cream if I succeed at that task.” Personally, I am not quite sure how that directly relates to the true reward signal we’re after, which is always correlated with success at the task in question. Besides that, it will also interfere with the reward pathways which are central to when you should eat ice cream, which is when you are hungry for it.
You should search for all genuine sources of reward signal from the environment, and this is much easier and more reliable than introducing arbitrary reward or externalized punishment. Making correct predictions is actually a pretty strong type of positive reward, and can happen frequently enough, and you don’t have to worry too much about “remembering” that it happened. But do consciously keep an eye out for it.
The environment punishes automatically too, so do not add additional stress to this process.
Part of healthy optimism is choosing to consider yourself successfully “on the path” to wherever you want to be, if there’s no evidence that things are getting worse. (Often, there really isn’t going to be evidence that things are getting worse!) Part of it will also mean placing more credence on the idea that what you are doing at any given moment might actually be what you should be doing, and that you’ll eventually give in to the pressure to do a little bit more of what your long-term desires say you should be doing.
The reason we’re allowed to search for a strategy that’s comfortable and providing the most fun for us to use is that these positive feelings are associated with consistently receiving positive reinforcement from successfully learning the task.
To give an example: When writing this essay, I usually have a notepad open which is given the name of the essay. This inserts the essay writing into the prediction stream. I don’t make hard predictions about how much I will write today, just that I will probably write something. Then, when I’m at my desk, and not sure what to do, I will unthinkingly open the notepad and start typing in a sentence or two, then close it, if nothing else comes out.
Eventually - as I have observed, not dictated - once the essay contains a handful of paragraphs, I usually continue it in longer bursts of work, the longest of which usually completes it. The whole process usually takes a handful of days of writing (although the topic and what I want to say about it floats in my brain for about a month).
It takes a good month, maybe longer, to generate something useful to say about a topic that seems worth writing down. Actually writing it down takes a lot less time. Again, these are not hard commitments, they are observations. The commitment comes in the form of “I will think of a good thing to write about, and eventually that will just happen.” When that does happen, I receive positive reward signal, and this behavior gets reinforced.
This process does seem reliable.
Finally, I also note that I consider whatever strategy I am currently in the process of learning probably resembles the final outcome to some degree. That is, I hope (and believe, given that I am doing incremental gradient-descent-like improvements to this strategy) that the ideal strategy has many of the key characteristics that I’ve thus far uncovered.