Less Wrong: Open Thread, September 2010

matt

This thread is for the discussion of Less Wrong topics that have not appeared in recent posts. If a discussion gets unwieldy, celebrate by turning it into a top-level post.

I'm writing a post on systems to govern resource allocation, is anyone interested in having any input into it or just proof reading it?

This is the intro/summary:

How do we know what we know? This is an important question, however there is another question which in some ways is more fundamental, why did we choose to devote resources to knowing those things in the first place?

As a physical entity the production of knowledge take resources that could be used for other things, so the problem expands to how to use resources in general. This I'll call the resource allocation problem (RAP). This problem is widespread and occurs in the design of organisations as well as computer systems.

The problem is this, we want to allocate resources in such a fashion that enables us to achieve our goals. What makes the problem interesting is that making a decision about how to allocate resources takes resources itself. This makes the formalisation of optimal solutions to this problem seemingly impossible.

However you can formalise potential near optimality. That is look how to design systems that can change the amount of resources allocated to the different activities of the system with the minimum of overhead.

Upvoted for importance of subject - looking forward to the post. Have you read up on Information Foraging?

4Oscar_Cunningham16y

I was having similar thoughts the other day while watching a reality TV show where designers competed for a job from Philippe Starck. Some of them spent ages trying to think of a suitable project, and then didn't have enough time to complete it; some of them launched into the first plan they had and it turned out rubbish. Clearly they needed some meta-planning. But how much? Well, they'll need to do some meta-meta planning... I'd be happy to give your post a read through. ETA: The buck stops immediately, of course.

7Snowyowl16y

This sounds interesting and relevant. Here's my input: I read this back in 2008 and I am summarising it from memory, so I may make a few factual errors. But I read that one of the problem facing large Internet companies like Google is the size of their server farms, which need cooling, power, space, etc. Optimising the algorithms used can help enormously. A particular program was responsible for allocating system resources so that the systems which were operating were operating at near full capacity, and the rest could be powered down to save energy. Unfortunately, this program was executed many times a second, to the point where the savings it created were much less than the power it used. The fix was simply to execute it less often. Running the program took about the same amount of time no mater how many inefficiencies it detected, so it was not worth checking the entire system for new problems if you only expected to find one or two. My point: To reduce resources spent on decision-making, make bigger decisions but make them less often. Small problems can be ignored fairly safely, and they may be rendered irrelevant once you solve the big ones.