RS 161 - Tom Griffiths and Brian Christian on "Algorithms to Live By"

Release date: June 12th, 2016

Tom Griffiths and Brian Christian

Julia chats with the authors of "Algorithms to Live By", about how to apply key algorithms from computer science to our real life problems. For example, deciding which apartment to rent, planning your career, and prioritizing your projects.

In the process, they discuss the assumptions that underlie those algorithms (and what to do about the fact that those assumptions are inevitably violated by the messy real world), and why procrastination might actually be the right algorithm for the wrong problem.

Brian and Tom's Book: "Algorithms to Live By"

Brian's Pick: "Finite and Infinite Games" by James Carse

Tom's Pick: "Do The Right Thing" by Stuart Russell and Eric Wefald

Podcast edited by Brent Silk

Full Transcripts 

Comments

[faraz]

wew what an algo!

[Max]

About procrastination, the guests suggested weighing tasks by importance, but operating systems also steadily increase a task's importance while it waits its turn so that it won't be neglected forever. So the longer you put off cleaning out your closet or having children because you have more important things to do, the more urgent those neglected things will become. That's pretty intuitive.

[Max]

I have a solution to the Repugnant Conclusion.
Suppose you have limited resources R and population n, and each person gets R/n resources.
Model a person's utility as log(R/n), so doubling the person's resources increases his or her utility by log(2).
Then, total utility is n*log(R/n) = n*log(R)-n*log(n)
Now find the n that maximizes total utility by setting the derivative to zero.
log(R)-log(n)-1=0
log(R/n)=1
R/n=e
n=R/e
So the optimal population is R/e, and each person gets e resources.
You can modify the model for different wealth distributions and utility functions. The point is to make a mathematical model instead of just hand-waving.

[arfi]

Reads the article, I felt much better.

[Jameson]

This episode has at least three really useful techniques:

The Rule of 37: Spend 37% of available time in explore mode before moving on to exploitation.

Multi Step Tasks: Evaluate all the steps before beginning them (take the midway pail of water down half the hall)

Have Reasonable Expectations: Just because your strategy does not yield an optimal result does not prove you have a less than optimal strategy since even the most optimal strategy only guarantees the highest probability of success, rather than absolute success.

Julia Galef may not want you to think like a computer, but she apparently does want people to think much more rationality.

[Ernesto]

Good insights to learn about. a very interesting topic to help us with Time management and such. thx