Research Notes on Social Protocols

Research Notes on Social Protocolshttps://example.org/research-notes/Recent content in Research Notes on Social ProtocolsHugoen-us<link>https://example.org/research-notes/2024-02-05--top-note-algorithm/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-02-05--top-note-algorithm/</guid><description>Review of the Top Note Algorithm We are considering two posts: $A$ and $B$. $B$ is a reply to $A$. Then we are observing the following events: $u_A := \text{the event that a vote on A is an upvote}$ $u_B := \text{analogous}$ $s_B := \text{the user was shown post B}$ $\bar{s_B} := \text{the user was not shown post B}$ $s_{B*} := \text{the user was shown the top note on B}$ $\bar{s_{B*}} := \text{the user was not shown the top note on B}$ Then we define two quantities:</description></item><item><title/><link>https://example.org/research-notes/2024-02-06-informed-probability/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-02-06-informed-probability/</guid><description>Measuring Informed Probability Our scoring formula starts with an estimate of the probability that a user upvotes a post, and then observes how exposure to information in the note influences that probability.</description></item><item><title/><link>https://example.org/research-notes/2024-02-06-vote-tallies/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-02-06-vote-tallies/</guid><description>Computing Vote Tallies The notes here have been updated and moved to 2024-05-24-calculating-tallies.md</description></item><item><title/><link>https://example.org/research-notes/2024-02-10-brigading-solution-matrix-factorization/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-02-10-brigading-solution-matrix-factorization/</guid><description>Thoughts on “Brigading” The community notes algorithm, doesn’t solve the problem of “brigading”, or the “self-selection problem.” But here’s an idea how a variation of this algorithm could solve the problem.</description></item><item><title/><link>https://example.org/research-notes/2024-03-01-support-formula/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-03-01-support-formula/</guid><description>Derivation of the “Support Formula” The Problem Suppose we have an estimate of the effect of note B on post A, and the effect of subnote C on note B. Should we be able to predict the effect of subnote C on post A.</description></item><item><title/><link>https://example.org/research-notes/2024-03-05-support-formula-derivation/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-03-05-support-formula-derivation/</guid><description>More Formal Derivation of Support Formula In my notes on the support formula I used some fairly informal reasoning to derive the formula. Here are my thoughts on a somewhat more formal derivation.</description></item><item><title/><link>https://example.org/research-notes/2024-03-07-adjustment-for-informed-vote-selection-bias/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-03-07-adjustment-for-informed-vote-selection-bias/</guid><description>Adjustment for Informed Vote Selection Bias The informed upvote probability calculated from the informed tally suffers from a selection bias. People who choose to vote on the note may vote differently on the post from people who don’t vote on the note.</description></item><item><title/><link>https://example.org/research-notes/2024-03-07-improvements-to-algorithm/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-03-07-improvements-to-algorithm/</guid><description>Improvements to Scoring Algorithm Fix algorithm for 3-level arguments (post->note->subnote): estimate $P(A|C)$ using tallies for A given considered C. Better heuristic for top note: $informedVotes × relativeEntropy$ Adjustment for Informed Vote Selection Bias Adjustment for Entrenchment Bias: Users’ informed votes influenced by whether they voted on post before voting on note Implement/Improve support formula (use $P(note|subnote)$ to inform prior for $P(post|subnote)$: deal with probabilities below 0 or about 1 improvements based on notes in 2024-03-05-support-formula-derivation.</description></item><item><title/><link>https://example.org/research-notes/2024-03-27-questions/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-03-27-questions/</guid><description>Question Model Here’s a review of an idea we have had for adding questions to our model and the possible benefits. UI A post can be a question, and the UI enforces that it is worded like a question.</description></item><item><title/><link>https://example.org/research-notes/2024-04-09-feedback-deliberative-consensus-protocol/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-04-09-feedback-deliberative-consensus-protocol/</guid><description>2024-04-09: Feedback on “Deliberative Consensus Protocol” Link: https://social-protocols.org/deliberative-consensus-protocols/ Introduction: Scalable Group Decision-Making However, there are solutions to all these problems! That sounds like too enticing of a promise too early on in the essay.</description></item><item><title/><link>https://example.org/research-notes/2024-05-24-calculating-tallies/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-05-24-calculating-tallies/</guid><description>Calculation of Informed and Uninformed Tallies Background: Effects One of the most important concepts in the GlobalBrain algorithm is the “effect” of a reply on a target. We use the term “target” to refer to the post being affected.</description></item><item><title/><link>https://example.org/research-notes/2024-06-03-choosing-priors/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-06-03-choosing-priors/</guid><description>Choosing Prior Upvote Probability We currently use a prior of $Beta(0.25, .025)$ for the uninformed upvote probability. This was arrived at from two different directions: 1. The Empirical Prior The ‘Empirical Prior’ of $Beta(.</description></item><item><title/><link>https://example.org/research-notes/2024-06-10-transitive-support-formula-review/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-06-10-transitive-support-formula-review/</guid><description>Uses of the Transitive Support Formula This note briefly reviews the support formula, why we are not using it now, and how we could use it. The support formula allows us to estimate effects transitively: e.</description></item><item><title/><link>https://example.org/research-notes/2024-06-10-weighted-average-informed-probability/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-06-10-weighted-average-informed-probability/</guid><description>Proposed Change to Scoring Formula I think that we should change our scoring formula to calculate p not based on a single critical thread, but as a weighted average of all argument threads.</description></item><item><title/><link>https://example.org/research-notes/2024-07-16-rationales-on-weights-in-score-calculations/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-07-16-rationales-on-weights-in-score-calculations/</guid><description>[!NOTE] Moved here from src/lib/score.jl in GlobalBrain.jl. weight returns a score for determining the top comment for purposes of calculating the informed probability of the post. It is a measure of how much the critical thread that starts with that comment changes the probability of upvoting the post.</description></item><item><title/><link>https://example.org/research-notes/2024-07-19-notes-on-imbalance-score/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-07-19-notes-on-imbalance-score/</guid><description>Notes on 2024-07-19-thoughts-on-convincingness.md This note is a response to Johannes’s 2024-07-19-thoughts-on-convincingness.md “Imbalance” ≈ “Statistical Independence” It seems to me that “imbalance” as you are defining it boils down to statistical independence, or lack of correlation.</description></item><item><title/><link>https://example.org/research-notes/2024-07-19-thoughts-on-convincingness/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-07-19-thoughts-on-convincingness/</guid><description>Measuring Convincingness A lot of the Global Brain algorithm revolves around the concept of convincingness. We approached the concept of convincingness as how convincing a comment is with regard to the post it replies to, measuring it as how likely users are to change their vote on the target post, given that they considered the comment.</description></item><item><title/><link>https://example.org/research-notes/2024-10-03-factcheck-based-scoring/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-10-03-factcheck-based-scoring/</guid><description>Proposed New Factcheck-Based Algorithm So what if, under a fact checks or opinion poll, all replies are other fact checks. The system then chooses the top sub fact-check (initially, the one with the most upvotes).</description></item><item><title/><link>https://example.org/research-notes/2024-10-09-merging-community-notes-matrix-factorization/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://example.org/research-notes/2024-10-09-merging-community-notes-matrix-factorization/</guid><description>Merging Community Notes Matrix Factorization into the Global Brain The primary goal of the global algorithm is to estimate informed opinion – how users would vote on a item given they had considered all the comments on the item.</description></item></channel></rss>