Self Improvement

From: Peter de Blanc (
Date: Fri Jan 27 2006 - 12:10:19 MST

I've been thinking about self-improvement methods available to humans
right now, and particularly for Seed AI Wannabes. What are people's
thoughts on the value of time spent improving "general" intelligence and
rationality versus time spent on the specific fields relevant to Seed
AI? (Of course, you can often do both at once)

In particular, I think we can use LOGI to classify different methods of
self-improvement. Here's a look at some stuff that's cheaply available

Code level: nothing.

Modality level: reading, speed reading (how effective is this,
anybody?), foreign languages, and also various sports, martial arts, and
"twitch" video games.

Concept level: I think that most people do not often generate new
concepts except as children. Concepts can be serialized into text or
speech, and then re-encoded by another human, but the transfer is lossy.
Because of this, it is usually not possible to teach someone a complex
concept unless you can express it concisely in terms of simpler concepts
that they already know.

A person who is experienced at generating new concepts should be able to
"repair" concepts that have been mangled by the process of serialization
and re-encoding, and so should be able to learn concepts faster with
fewer supporting concepts. Most training available today (in any
specialized subject) seems to focus on conveying a large number of

What is out there that can help you practice generating new concepts?
Well, there are induction games like Zendo and and Eleusis, which
(depending on who you are playing with) can lead to many new concepts,
but these new concepts are usually discarded after the game ends. It has
been suggested to me that the process of learning games like Go involves
generating concepts, but this seems very slow to me. Writing science
fiction seems like it fits here too.

Thought and deliberation levels: "Critical thinking," the scientific
method, mathematical logic, and probability theory all go here, and so
do things like meditation. These are different approaches to thought
that can be learned, but seem to be very sensitive to natural ability.
High schools and universities attempt (or claim to attempt) to teach
"critical thinking," but are they teaching anything, or only weeding out
those who don't do it already?

Heuristics and biases belong at the thought level. Learning about
heuristics and biases can help you spot systematic errors, and
(hopefully) learn to stop making them.

Something fun you could try is to sit down with a calculator and Google,
and optionally a friend, and play Bayescraft: think of a question with N
possible answers, assign a probability to each, and check the answer.
Multiply your cumulative score by the probability you assigned to the
correct answer. Alternatively, one player can write down a sequence in
binary, and the other players can predict the next value.

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:55 MDT