Early thoughts on Cyborgs & Mentats

Do you tie yourself to the mast, or do you learn to not feel the craving to jump?

While talking with Dreev, the creator of Beeminder, I realized we had two different approaches to fighting akrasia and achieving our goals. One focused on changing your incentives and your environment (the Cyborg), and one focused on investing conflict and reaching self alignment (the Mentat)

---

The Cyborg uses Obsidian, calendars, reminders, bets, Beeminder. External systems that constrain their actions so they take the right action.

The Mentat practices meditation, the CFAR handbook, self work, and general introspection, so that he resolves his internal conflicts and becomes a person who just naturally takes the best action towards their goal.

Mentat is Lucius Malfoy in HPMOR, burning the paper on which he wrote his thoughts, so his enemies can’t use it against him.

Cyborg is Thomas Frank, Tiago Forte, building elaborate external systems to manage their memories and schedule.

Both are present in the self-improvement world.

Both are present in the rationalist world.

Rationalists tend to be more Mentat, wanting to become better through improving their mind.

Or not really, some are going real hard in the external augmentation side, with the Second Brain.

There is a difference between intellectual augmentation and agency augmentation.

Rationalists can be both cyborgs and mentats for intellectual augmentation, but they’re way more mentat for agency, wanting the agency to be a part of them, not an external system.

Inkhaven is a Cyborg technique: strapping yourself into a container that forces you to act a certain way by changing your incentive landscape.

CFAR workshops try to turn people into Mentats, to yank people into a new trajectory, in which they have the tools to slowly become more aligned with themselves and solve their problems and achieve their goals.

Akrasia is the thing I care about. How to solve akrasia. It’s the perennial rationalist question about applied rationality. Nearly all of applied rationality is investigating and developing techniques to solve akrasia.

Claim (strongly worded, weakly held): If you look at any instance of akrasia, the causal chain leading to it will contain at least one node of the type “actually part of me does not want to do that”, and that changing all nodes of this type to “actually this part does want to do that” is sufficient for changing the action taken and solving the akrasia.

→ The mentat solves their akrasia by aligning their parts, making them all want the thing.

Claim (strongly worded, weakly held): If you look at any instance of akrasia, there is at least one node in the causal chain causing it that is a property of the environment, that could be changed to a value that would be sufficient to cause the action to change, and solve the akrasia.

→ The cyborgs solve their akrasia by changing their environment

An issue I have with the Cyborg approach, is that a lot of their techniques are agnostic to your actual extrapolated volition. For example: putting yourself into a situation where you will lose something important, like money, if you don’t do X. Such a setting will reliably make you more motivated to do X, no matter the value of X.

I guess a lot of the CFAR techniques can also be used in a way that stomps down on parts of you and makes you take the wrong decisions according to your extrapolated volition, but I feel like it and the other mental practices are still gesturing at a direction of “listen to every part of yourself, and notice when you’re trying to ignore a part of you”. Once you get good at it, you don’t really risk forcing yourself into bad decisions anymore, or bad external systems too I guess.

Let’s take a classic example of Akrasia: systematically going to sleep later than you endorse.

The Cyborg could approach the problem like this:

• My system 2 can consider the future and realize that it is worth it to go to sleep on time to not be tired tomorrow

• My system 1 cannot, because of hyperbolic discounting, and so will be more moved by the immediate pain of stopping whatever activity I’m doing and going to sleep, compared to the larger future pain of being more tired the whole day tomorrow

• So, to solve this issue, I will use a time when I have distance from the problem and willpower (i.e. a time when my system 2 can direct my actions), to set up a system or change my environment, such that the immediate incentive gradient actually pushes me to sleep in the evening. For example:

  • Setting up a script on my computer to turn it off automatically at midnight

  • Setting up an app blocker that only my friend can unlock

  • Not going to this cool party, as me-at-the-party won’t want to go to sleep, but me-at-home will want to.

  • Asking a friend to check in on me to see if I’m in bed by midnight, and if not, I have to pay them $20

  • Setting up a Beeminder that charges me $20 each time my smartwatch report that I was not in bed by midnight

The Mentat could approach the problem like this:

• One part of me thinks it would be good to go to sleep earlier. This part of me is part of my higher self/inner narrator, and is good at verbally justifying its belief

• Another part of me thinks it would be bad to go to sleep earlier. This part of me is less verbal, but can place higher bids than the other for its action to happen

• Both of those parts fundamentally care about my wellbeing, and given the opportunity to communicate, they could resolve their conflict and find a policy that would satisfy them both.

• So, to solve this issue, I do some self work practice (CBT, IFS, meditation, internal double crux, non-naive trust dance, etc.) to set up a dialogue between those parts, which we run until they reach an agreement. The process could look like:

  • Eliciting the first part’s goal: making sure you finish your work by the deadline. It’s currently worried this won’t happen, because you’ve been so tired at work

  • Eliciting the second part’s goal: protecting your dignity and humanity, by securing time for yourself to play and learn and have agency. It’s currently worried that the work is sucking the life out of you and that you’re becoming just a cog in a machine

  • Both parts acknowledge that the other’s goal is also important. It would be terrible to become a soulless cog, and also terrible to become destitute because you lost your job

  • Finding a potential change that would be an improvement over the status quo for both parts. For example, not doing overtime twice per week, which would unlock higher quality self-directed time on those days, and going to sleep earlier, which would make the whole work week more productive.

  • Both parts find this change better than the status quo, no other parts object, so I am motivated to do it and don’t need external pressure to go to sleep on time anymore.