Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio.
This is: Make your training useful, published by AnnaSalamon on the AI Alignment Forum.
As Tom slips on the ice puddle, his arm automatically pulls back to slap the ground. He’s been taking Jiu-Jitsu for only a month, but, already, he’s practiced falling hundreds of times. Tom’s training keeps him from getting hurt.
By contrast, Sandra is in her second year of university mathematics. She got an “A” in calculus and in several more advanced courses, and she can easily recite that “derivatives” are “rates of change”. But when she goes on her afternoon walk and stares at the local businesses, she doesn’t see derivatives.
For many of us, rationality is more like Sandra’s calculus than Tom’s martial arts. You may think “overconfidence” when you hear an explicit probability (“It’s 99% likely I’ll make it to Boston on Tuesday”). But when no probability is mentioned -- or, worse, when you act on a belief without noticing that belief at all -- your training has little impact.
Learn error patterns ahead of time
If you want to notice errors while you’re making them, think ahead of time about what your errors might look like. List the circumstances in which to watch out and the alternative action to try then.
Here's an example of what your lists might look like. A bunch of visiting fellows generated this list at one of our rationality trainings last summer; I’m including their list here (with some edits) because I found the specific suggestions useful, and because you may be able to use it as a model for your own lists.
Action ideas, for three related biases:
A. How does it help to know about overconfidence[1]? What can you do differently, once you know your impressions are unreliable?
Action ideas:
Try many things, including things you “know” won’t work. Try cheap ones.
Don’t be so sure you can’t do things.
Don’t be so sure that the things you are doing, are working:
If a given “necessary” task is using a large portion of your week, test what happens if you skip that task.
Ask others whether your efforts are working, and what you might try instead. Test their suggestions.
Ask how you’ll know if you hit your goal: what specific observables will be different? (Not “I’ll know calculus” but “I’ll be able to solve all the problems on the AP calculus test”. Not “I’ll be happier” but “I’ll improve my score on the Beck Depression Inventory”). Track these observables.
Be suspicious of received wisdom, since others are also overconfident. But don’t just ignore that wisdom in favor of your own error-prone impressions -- look for empirical tests.[2]
Your friends and family are weirder (more unlike your models) than you think they are. Try to notice how.
B. How does it help to know about the conjunction fallacy? What can you do differently, once you know specific stories are less likely than we generally expect?
Action ideas:
Use simple or disjunctive plans:
Choose a (city/college/etc.) in which there are many promising possibilities, not one with a single, highly promising scenario.[3]
Apply for many jobs, in many sectors of the economy.
Gather re-purposable resources, such as money, rationality, sanity, capable friends, math skill, reading speed, mental and physical fitness. Focus on fundamentals more than on situation-specific techniques.
Tell detailed stories when you want to convince someone:
Describe specific scenarios to angel investors, potential customers, etc.
Visualize specific scenarios, when you want convince the less verbal parts of yourself that your new (exercise plan / whatever) is worth the effort.
Don’t put all your caution into safeguarding one particular step. For example, don’t “ensure your start-up will succeed” by focusing only on the programming step, or only on the “where to sell it” step. Brainstorm many ways your plans can go wrong.
Realize that conjunction-ridden theories...
view more