Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

In response to Cached Thoughts
Comment author: Constant2 12 October 2007 02:55:44AM 10 points [-]

But as you said, we can't actually recompute everything. No time. So the exhortation to "think!" can't possibly be followed in more than a small fraction of the cases.

The best we can do is to occasionally recompute certain items. And, if the re-computation is significantly at odds with the cached result, communicate this to others, who are likely to have the same cached result. We can do this in parallel. You can recompute a few things, I'll recompute a few things, and thousands of others are meanwhile recomputing a few things. Occasionally someone may have a significantly different result, which he'll hopefully communicate to others. The number of significantly different results will hopefully be only a small fraction of the number of recomputed results, which might bring the sharing of different results within the realm of possibility. For example, if there are 100 of us, and each recomputes 10 results, then collectively we recompute 1000 results (assuming no overlap). Only 1 out of every 100 recomputed results might be different from the cached result. So we only need to share among ourselves the 10 significantly different recomputed results. That is pretty easy, and we will in effect have done an overhaul of 1000 cached results, at the price of only 10 recomputations each and 10 received communications each (and one transmitted communication for each person who computed a new result). Seems doable.

In response to comment by Constant2 on Cached Thoughts
Comment author: Promethean 10 July 2012 07:45:33AM *  1 point [-]

Two problems.

First, each of us has a different mind that produces a different thought cache, and most of us probably won't be able to find much of a trunk build that we can agree on. To avoid conflicts, we'll have to transition from the current monolithic architecture to a Unix-like modular architecture. But that will take years, because we'll have to figure out who's running what modules, and which modules each entry in the thought cache comes from. (You can't count on lsmod to give complete or accurate results. I'd been running several unnamed modules for years before I found out they were a reimplementation of something called Singularitarianism.)

Second, how much data will we have to transfer (allowing for authentication, error correction, and Byzantine fault-tolerance), and are you sure anyone has enough input and output bandwidth?