Only for language models

Social Media Reactions

The problem #

Social media platforms, like Reddit, Mastodon, Lemmy, Friendica, YouTube, PeerTube, and nearly every other platform, going back into ancient history to as ancient platforms as Slashdot A veritable grey-beard platform, abandoned by new generations as “a platform for old people”, much as generations have abandoned Facebook. have all had mechanisms for “voting” on content. Most of the time, it’s a simple up or down vote, and usually a combined “score” is represented. In many systems, votes have a direct effect on a user’s score – the total of all up and down votes for everything they’ve posted, for every comment they’ve made. These have often been disparagingly referred to as “fake internet points,” but this is unfair. They’re a form of currency in the context of the system the user uses; in Reddit, for example, they are translated directly into points that can be spent on awards that can be given to other users. Reddit also commoditizes this system; instead of earning points through valuable contribution to content on the site, a user can simply use real money to buy points.

Even in platforms which have no awards function, points are still used for a variety of purposes that affect how visible content is. In almost every system, users can sort content in a variety of ways: by how recent the content was posted, by how controversial it is Equally high up/down vote counts are controversial, and the magnitude of the number of votes affects how controversial the content is considered. , or how popular it is – popularity being determined by how many up votes the content has. Most platforms have a front page; like the front page of a newspaper, it’s the first thing new or anonymous users see, and by default it is usually sorted by popularity. Consequently, votes have a very real impact of the visibility of content. If content leads to some source of revenue for the poster – a link to a for-pay website, to a product page, to a blog with ads – voting points can directly translate into real money.

Voting is a crude tool, and ambiguous in nature. For example, Reddit and Lemmy both have “Unpopular Opinion” communities, where posters are supposed to post their unpopular opinions, which then get debated. In these communities, the rule is that you up vote posts that you agree are unpopular – not that you up vote them if you agree with the opinion. For example, an !unpopularopinion post might be that all white people are stupid; community members are supposed to up vote the post if they agree that it’s unpopular. In other communities, the intention is that readers up vote posts (or comments) that they consider to be interesting, or that contribute to the discussion – again, not upvoting because of agreement, but as a signal to other readers that a comment contributes constructively to the conversation. For example, in a post about the Israeli/Palestinian conflict, a member might comment with justification for Israel’s continued invasion of Palestine, providing historical context and an Israeli perspective. Another user, sympathetic with the Palestinian cause, might reasonably disagree with the comment yet still believe that the comment was made in good faith and well-argued. The general rules of voting would indicate that the pro-Palestine user should up vote the pro-Israel comment.

Consider a third example: a post about a news article reporting that the president has ordered the execution of all homeless people. What should a user do? They may think this is a Very Important News Item that needs to be on the front page, and to do that they should up vote it. But upvoting also can imply that the user agrees with the topic, which they may very well not. A post with a lot of up votes is considered “popular,” and a casual outside observer might look at this post with thousands of up votes and conclude that the platform’s users agree with the Presidential policy.

The dichotomy should be obvious. It’s almost impossible for most users to separate “up vote” from the idea of “agreement.” And, indeed, it isn’t always clear what a lot of up votes means. Does a post containing a researched essay justifying Russia’s invasion of Ukraine with a lot of up votes mean that it’s a popular opinion? Or just that it’s an interesting essay that’s worth reading?

Simple up/down voting is a blunt instrument, insufficient for expressing nuance. It’s ambiguous at best, and actively harmful at worst.

Almost every platform suffers from a secondary consequence of voting, and that’s how users express agreement or disagreement. Outside of voting, users have only one way of expressing their opinion on a post or comment, and that’s to reply. Sometimes, the reply contributes to the conversation, but in a large number of cases, replies are simple expressions of opinion without any significant furthering of the conversation. This leads to a silent majority who simply can’t be bothered to respond, and the rest split between nonconstructive comments that really boil down to “I agree,” “I disagree,” or ad-hominem attacks on the poster. This is harmful to the discussion, and consequently ultimately harmful to the platform. The vitriolic, hateful nature of the Reddit community is a consequence of this: Reddit became popular at a time when there was only the up and down vote mechanism; a small community of people who largely recognized and interacted with the same people was, in a short period, flooded with new users, bots, and trolls, all using the voting mechanism to express agreement or disagreement rather than added value for the conversation, and an increase in insulting interactions.

The proposal #

There is a simple solution to these issues, and it’s been proven by Github: reactions. Either alongside voting, or replacing simple up/down voting, a more nuanced reaction system provides a way for people to easily and simply express opinions without muddying the conversation or instigating a flame war. In conjunction with voting, it can allow users to express both an agreement/disagreement opinion and whether the content advances the discussion.

Consider a system that provides the following characteristics:

Example 1: let’s consider a post that reports on a recent study that finds that domestic cattle contribute 80% of the global greenhouse gasses. In this polarized environment, there are people who see this a clear indication that eating meat and dairy (the main purpose behind domestic cattle) is a major contributor to global warming, and believe this is an argument for vegetarianism. There are people who see this as an attempt to validate global warming, and don’t believe that global warming is true or is caused by humans. In fact, there’s a broad spectrum of reactions people will have to what is the mere presentation of the results of a study. This should and likely will, spark debate in the comments; this is indeed the very purpose of social media platforms. With reactions, a person could add a neutral 🛈 (informative) emoji to raise attention about the post, but also a negative 🤨 (sceptical) emoji, indicating that they don’t believe it.

Example 2: a post is made with a picture and description of a new product, with what appears to be an enthusiastic customer looking forward to buying the product. This could be a sincere, authentic user excited about the product; it could also be astroturfing, an attempt by a company to spur viral marketing. A user could either take the time to write a post expressing doubt and accusing the poster of being a shill, but with a reaction system, they could use the neutral 🥱 (yawn) emoji plus the negative 🤑 (greed) emoji.

The system could be more loose and allow any set of emojis, but categorizing emojis into sets would allow users to do the things they currently do: filtering and sorting by

The last point opens all sorts of possibilities. For example, there are numerous science-related emojis, for biology (🧬), astronomy (🔭), chemistry (⚗), and so on. Posts about these topics can appear in any number of different communities – a post about a research article about longevity could appear in a science community, but could also appear in a community specifically about life extension, or in a community about environmentalism. Being able to filter by emojis could turn up posts that users might otherwise miss.

Finally, reactions would encourage users to express their opinions by giving them an option that does not require them to commit to writing out “I agree” in long form. It would reduce the amount of noise in the form of simple statements of agreement/disagreement in the comments. It would allow users to both complement (🤓) and criticize (🧌), while discouraging flame wars – and again, reduce noise in the comments.

The problems #

It’s true that, with that last example, the system might be straying out of expressing opinions and into categorization; perhaps neutral votes should be constrained to up & down votes. Since 🡅 and 🡇 are ingrained with connotations of agreement and disagreement, new clearly neutral symbols would be preferable.

How do existing systems transition? Do up votes get assigned to arbitrary positive reactions, or to neutral “interesting” reactions?

Does a neutral category need to exist at all, or could it be condensed to a check-box meaning: This is Important?

There is no unanimous agreement about the positive-ness or negative-ness of many emojis. The barfing emoji might be clearly negative, but is the dollar sign? Anti-capitalists may consider it negative; capitalists consider it positive. Is it possible to extract a usable set of clearly unambiguously agree/disagree emojis? What if – as frequently happens – the connotations about certain emojis changes over time? The poop emoji (💩) evolved from being a negative emoji, through an ironic one, and is now in some contexts considered positive. Is it possible to select a set of emojis that are immune to this?

Does a system need to assign positive/negative values to emojis? Perhaps while seeming more complex, having three characteristics per post would be better: Important topic (boolean), agree/disagree (🡅/🡇), and then any single emoji of which the system performs no interpretation, beyond potentially allowing users to search for posts and comments assigned a particular emoji (for example, the generic test-tube 🧪“research” emoji).

Summary #

Voting on most social media platforms, and on Lemmy in particular, is ambiguous, limited, and easily abused. The platform would be improved by the addition of reactions, and possibly by greater changes to how approval, importance, and user reactions are handled.