Promise Theory: the ethics of algorithmic news feeds and chronological timelines
The algorithmic news feed is the source of misinformation, addiction, and filter bubbles. But why did any of this happen? I believe there’s a simple answer, and that promise theory can direct us to philosophical heart of the problem. Once we analyze algorithmic news feeds from the perspective of promise theory, it becomes easier to reason about the ethics involved, and to propose a better way of handling algorithmic news feeds.
Promise Theory states that an agent can only make promises about itself. In the case of the algorithmic news feed, we’re concerned with the Subscribe button. The promise that the Subscribe button makes is quite complex. It goes something like this:
When I click on the Subscribe button —
- I am agreeing to receive ALL of the Updates from the Source.
- I am agreeing to allow you to include all of the Updates from this Source in chronological order in a Reader.
Subscriptions aren’t a new idea. Let’s briefly explore how a promise is kept with a newspaper subscription in the year 1948. The concept is simple: You provide me a newspaper every day, and in exchange I give you money. As long as you keep giving me newspapers, I’m going to keep giving you money. And the reverse holds true as well, as long as I keep giving you money, you’re going to keep bringing me newspapers.
There’s a few problems with this system:
- Physical requirement: You need to physically deliver the newspaper to my front door.
- Non-atomic information delivery: The smallest, atomic piece of information that makes sense to read is an article, but the smallest that makes sense to deliver to you is an entire newspaper. It would be nonsense to show up at your door 15 times a day with tiny slivers of paper each time a story was published!
- Delivery time: The earliest I can get updates is once every 24 hours. And due to the process of creating, editing, and processing the newspaper, information may be several days old before I receive it.
This equation changes a bit when you introduce computers and the internet as a medium of delivery and exchange. The thing that’s getting delivered is much smaller now — now we care about the article itself, not the newspaper. Someone doesn’t have to be present to deliver a physical newspaper, so the delivery mechanism can update at the fastest speed that reading device cares to check or at the fastest speed that the server can push a notification. And I no longer need to give you money directly for the article you delivered — now this economic exchange is abstracted to advertising revenue, or voluntary support (Patreon or direct donations) or both.
One note on this equation before we move forward: There is still an atomic limit to the information that a subscriber wishes to receive. For example, if I suddenly started tagging you in Tweets, and each Tweet I sent was only a single character long, and when you combined the characters they spelled out a sentence, and then a paragraph — then you’d Unsubscribe (Unfollow) me quickly. Sending you such small bits of information is just nonsense, and it’s a waste of your time and attention. Twitter and the atomic limit of information consumption itself is an interesting topic to talk about on its own, but that’s a topic for another article. I’ll say this much: Twitter originally was limited to 140 characters, but has since bumped up the limit to 240 characters.
All promises can be broken. In promise theory we can enumerate this as the criteria for not keeping the promise. Not keeping the Subscribe button’s promise looks like this:
I will consider this promise broken if —
- Updates are not delivered in a timely fashion. I expect delivery of information to happen in minutes, not hours. (so from the point of view of the Publisher, eventual consistency is OK)
- Updates from a different Source are delivered.
- Updates are not delivered in chronological order.
- Updates are transformed, broken, or incomplete.
- Updates are stopped without my consent (without clicking an “Unsubscribe” button).
- I am not allowed to unsubscribe from updates.
- The update channel changes without my consent.
Let’s break this down a bit, because there a few agents in this equation:
Subscriber: The agent clicking the Subscribe button.
Source: Content made by the creator and hosted on the Publisher’s platform.
Creator: The agent that is creating new content and pushing it to the Publisher’s platform.
Publisher: The agent who owns the platform that hosts the Source’s content. The Publisher is also the one who determines the update mechanism and delivers the updates for the Source.
Reader: The agent which receives the updates that the Subscriber agreed to receive.
Subscriber/User: The agent which uses the Reader to consume content from the Source.
Advertiser: The agent which pays money to the Publisher to run marketing campaigns on the platform created by the Publisher.
Here is the main problem. When a single entity controls the Publisher and the Reader, and the Publisher makes its revenue through advertising, bad things happen. Why does this happen? I think it’s because the Reader is a proxy for the Subscriber, and is the agent in this equation that is responsible for declaring if the subscribe promise has been kept or not. But if the Publisher owns and controls the Reader, then they force the person behind the Reader to consume content in a way that maximizes economic value for the Advertiser instead of maximizing value for the person using the Reader. In other words, if the Subscriber doesn’t control the Reader, there’s no way to keep the Publisher honest.
If the Publisher owns the Reader —
- The Publisher can change the terms of the original subscribe promise, even if the promise was made in the past without consequences.
- The Publisher can decide to not keep the promise and force the Reader to accept behavior not originally intended.
- The Reader can be changed in a way where it’s indistinguishable from the Publisher.
It’s unsurprising that Facebook, Instagram, Twitter, YouTube, Reddit, and every other Publisher on the planet are mucking around with the Reader. It’s also completely unsurprising that all of the UIs for the Reader component is converging on a single design. After all, if the Publisher controls the Reader, it makes economic sense for the Publisher to maximize the amount of money they’re making. In other words, there are strong economic incentives at play to force you to consume content that the Publisher desires. Ultimately, since Publisher platforms generate revenue through advertising, the Publisher is economically incentivized to present a face to advertisers that maximize the revenue that marketing campaigns can earn through their platforms.
The next conclusion we reach should also be unsurprising: violating the promises the Reader presents in order to maximize revenue for marketing campaigns is a deception (or greed) optimized strategy. This system design inherently rewards deception driven strategies with money. Call it what you want: “click-bait” or “fake news” — this is the obvious side-effects of a system designed to to violate the contract between Reader and Publisher.
The Publisher wants to cut you, the Subscriber, out of the equation so they can maximize their own economic value at the cost of your privacy and control of the Reader. In order to cut you out, two agents have to agree to cede control of their data: the Creator has to agree to push their data to the Publisher, and the Subscriber has to agree to use the subscription mechanism provided by the Publisher.
Which data is more valuable? Is it the content uploaded by the Creator? Or is it the data generated by the Reader while viewing the content? Since Publishers are in the business of selling advertising, it is access to the data generated by the Reader that is economically valuable. After all, at the end of the day, these are the eyeballs the advertisers are trying to get on their own content!
There’s one caveat here: Advertisers do care significantly about brand perception. Advertisers do not want their marketing campaigns featured next to content out of “style” with their brand. In extreme cases, where your brand is at risk of being associated with content such as extreme violence, racism, radical politics, terrorism, and so on, an Advertiser may consider stopping marketing campaigns entirely.
The algorithmic news feed is designed to maximize eyeballs on advertisements. Its construction not only violates the promises given by the Subscription button, but it leads to other nasty behavior.
Addiction: Since the goal of the algorithmic news feed is to keep your eyeballs on as many advertisements as possible, the algorithmic news feed is designed to give you little hits of dopamine and get you addicted, much in the same way as gambling.
Filter Bubble: The algorithmic news feed can’t risk upsetting you by violating your core ideals because you might stop reading. This makes it difficult to challenge your worldview or give you new ideas. Note that this is different than making you angry or giving you pleasure through someone else’s misfortune (schadenfreude).
Extreme Reactions: The algorithmic news feed needs to keep things interesting. So over time, it will feed you more extreme articles, whether this makes you happy or sad or angry, this doesn’t matter — it keeps you reading and keeps your eyes on advertisements. Social network companies even have a term for this phenomenon: “engagement”. Note that this isn’t at odds with the filter bubble, as long as these “engagements” don’t violate your core ideals.
Dumbing Down Discussions: People are complex creatures with strong wills, but it would be wrong to say that algorithmic news feeds don’t effect human behavior. The effect we’ve witnessed is that algorithmic news feeds tend to dumb down discussions. Casual discussions, idea generation, political improvement, and so on are casualties on any public forum that is driven by an algorithmic news feed. I think they also reach out into the real world and impact our discussions there too.
Misinformation: What is knowledge and where does it come from? Epistemology is completely erased from algorithmic news feeds. Tracing information back to its source, or the author, is intentionally hidden in order to prevent you from leaving the algorithmic news feed and stopping you from having your eyeballs on more advertisements.
Your Thoughts Become Your Destiny
Last year I was in Japan for software work, and a (very talented) coworker asked me, “Whats the big deal about giving up privacy for ads?” His opinion was that it wasn’t important because companies are just serving you ads with the data you give up, and that you can block ads anyway with NoScript, uBlock Origin, and so on. At that moment, I didn’t have a good answer for him — but I felt it was important. I’ve had some time to think about this, and now I think I have a decent answer.
First, I can’t emphasize how important it is to own your own data and the mechanism through which you consume your data. We are talking about everything you choose to read, hear, and think. By proxy, your second-to-second thoughts, your passions, your fears, your philosophies, and your destiny will be controlled by how you decide to own and consume your data. You might think, “Well that’s a stretch, algorithmic news feeds don’t control me.” But we’ve shown exactly that: You are NOT in control of what you read if you’re not in control of the Reader. If you don’t control that Subscribe button, the Publisher can do whatever they please! The Publisher can change the rules at any time. The Publisher controls what you are reading so that the Publisher can maximize advertising money. The Publisher wants you to maximize your time and your eyeballs on their Reader. Ultimately the conclusion I discovered was this: Social media platforms without strong promises between Publisher and Reader are a form of mind control. To be clear, this isn’t the “evil” kind of take-over-your-whole-body-and-make-you-rob-a-bank mind control. This is the gambling kind of mind control, except it is applied to your friendships and the news you read and the videos you watch. I don’t think it was created this way on purpose either. It happened because software engineers accidentally built mind control software to make money from advertising.
Watch your thoughts for they become words,
watch your words for they become actions,
watch your actions, for they become habits,
watch your habits for they become your character,
watch your character for it becomes your destiny.
— Frank Outlaw
Here I was thinking to myself, “Eventually we software engineers are going to build something horrible, like killer AI robots, or addictive BTL VR.” (Better Than Life Virtual Reality, which could be more addictive than any drug). But no, software engineers built mind control by accident to make more advertising money. That is stupid and depressing.
The dangerous thing Publishers have done is convince us that it’s a good idea that we give up on empowering the Reader. That we give up on third party developers, and strong, open APIs for creating Readers to empower Subscribers. Unfortunately that’s a trend that’s continuing.
User Interface as Information Freedom (or Mind Control)
In the equation of promises with the Subscribe button, the Reader is a user interface, and the data that powers it. So far we’ve discussed just the data aspect of this equation, because I think it’s far more important due to data gravity. Data gravity is term coined by Dave McCrory — it’s the idea that applications and services are attracted to platforms where data lives. But most people interact with user interfaces, so the reality is that the shape of the data and where it lives ends up shaping and forming the user interfaces created to interact with it. A user interface, then, can be a beacon for information freedom or for information mind control.
This is the YouTube homepage as it appears on May 30th 2018. First, let’s distill a few things about the videos on this page. The YouTube homepage is infinitely scrolling, so let’s just pay attention to the first two sections, Recommended and Recently Uploaded.
I’ve already watched most of the Recommended videos, and out of the 3 I haven’t watched, I’m not interested in any of them. The average age (mean) of these videos is 10.14 days.
On the other hand, I haven’t watched any of Recently Uploaded videos, and they look far more interesting to me. I might actually watch some of those. The average age (mean) of these videos is 2.75 days.
This is a real example, but what’s interesting to me is that the the algorithmic feed (Recommended) is relatively worthless compared to the time-based feed (Recently Uploaded).
The great thing about YouTube is that it highlights the channel the video belongs to. In our promise theory equation, this is the Source— the epistemological source of the video. So I can instantly filter out sources I know I don’t trust (like Jim Sterling) and simply opt not to watch that content. The name of the channel usually will even give me an idea about tone of the content, or what the channel’s theme is about — all without having to watch the videos (and therefore watch advertisements). Brilliant!
Here is the (non-experimental) version of the YouTube subscriptions page, on May 30th 2018. This page is a chronological, time-based feed for the channels that I’m Subscribed to. Measuring the effectiveness of this feed is easy: I’ve watched every single one of these videos. I might have skimmed some of them, but I at least clicked on them and checked them out.
There are still a few problems here though — just like the Recommended feed, there’s no way to tell if I have watched any of these videos previously. Maybe instead of consuming them in chronological order, I just watched a few and left the rest for later? What if I just want to filter this feed by source — maybe I just want to quickly catch up on Linus Tech Tips?
One thing that YouTube gets right is the tools they give creators to tag videos with information about the source. This is important epistemological data that helps users identify and trust content, and overall just know what they’re getting themselves into. Larger channels get a checkmark to prove they are who they say they are (Twitter does this too). But still, there’s work to be done here — It’s often unknown who owns or runs the channel. For example Machinima is owned by Time Warner, and they also own YouTube channels Inside Gaming, ETC News, Realm Games, and Primr, but none of this information is disclosed or discoverable on YouTube itself.
RSS + YouTube = ❤
This is my RSS Reader, Reeder. RSS doesn’t contain any of the problems or constraints that consuming videos through YouTube has. It’s easy to filter by your subscription Sources, and it’s easy to know when something is read or unread. Most importantly, I own the subscription data — not YouTube, so YouTube doesn’t have the power to change what a “subscription” means, because I didn’t even use the Subscribe button to begin with.
One really important feature of RSS is that it’s easy to keep up with news from sources that don’t publish very often. Maybe you’re interested in news from a video game that is under development, but the developer isn’t going to post updates until there’s some significant news to announce. You don’t want to forget about the game either (you’d like to buy it and hear more news when it comes out). So subscribing via RSS to YouTube (or any other source) makes it easy to keep up, without missing out.
RSS isn’t perfect by any means. It’s difficult to network with others, it’s hard to archive old articles and old feeds, and sharing your favorite content is impossible without some other transport mechanism or sharing platform.
Reddit (Redesign, May 30 2018)
In stark contrast to YouTube, this screenshot of the Reddit (Redesigned) homepage is gross. These are all the defaults settings. You can instantly see the seven deadly sins of the algorithmic news feed in this one screenshot. Let’s review real quick:
✅ Addiction: The design of the site is infinitely scrolling, and it wants you to keep scrolling. There are no indicators of articles you’ve previously read, so there’s no way to know when you’re “done” or when you’re “full”.
✅ Filter Bubble: The default view sorts by “hot” (a filter bubble mechanic) and my IP Adresses’s physical location in the United States. I’m not even logged in or displaying any tracking information, but I’m still in the filter bubble!
✅ Dumbing Down Discussions: The design necessarily demotes the discourse on the topic to a comments section on another page, and that comments section is only a peanut galley in practice.
✅ Extreme Reactions: In this random sample of only 3 articles, we have both the maximum amount of outrage (Roseanne did something racist and got cancelled) and of maximum comfort (a gif of a dog snuggling a baby).
✅ Misinformation: The original source is not credited (there isn’t a place for the original source to be credited). Instead, the credit (and imaginary internet karma points) goes to the user that posted the image. The image isn’t linked to on the original site where it came from, and instead, was hosted on imgur.com, which is an image hosting site (with its own algorithmic feed).
But Reddit is driven by votes! It’s democratic!
This is possibly true of smaller subreddits and posts up to a certain voting threshold. But Reddit (and many other algorithmic news feed driven websites) is dominated by bots, and even marketing companies that make fake personas to create controversy and bend the discussion in their favor. Knowing the source of the article, image, or video is far more important than its popularity, and Reddit makes an effort to hide the Source. (In fact, many subreddits prohibit authors posting links to their own material entirely).
Reddit (Old Design May 30 2018)
One good thing about Reddit (old design) is that it at least gives you some ability to filter out content without consuming it entirely. While it still suffers from all of the same problems as the redesign above, you are not force-fed the content as you go. It’s easy to skim and scroll past the headlines you don’t care about. There are also buttons to expand and collapse content in-line so you don’t have have to stare at every image.
Twitter as a platform got many things right. I think the reason for this is simple: early on, Twitter gave developers wide open access to the API, and developers went wild creating their own Readers for Twitter. In other words, the promise between Publisher and Reader was enforced at an API level.
Twitter does an excellent job at epistemology. It’s easy to know where Tweets come from. Subscriptions on Twitter is called Following, and the Subscribe button is a Follow button, of course. When you follow an account, you get every tweet from that account, unfiltered, in a chronological timeline. It’s simple and predictable and useful.
The primary issue with Twitter is that the atomic limit of information was 140 characters (now 240 characters) for many years. While that’s great for information density, it makes more complex topics difficult to parse.
In practice, this means that people look to other solutions to generate long-form content (like Medium) or images (like Imgur) or videos (like Youtube), and Twitter becomes a commodity RSS-like reader. Twitter would be far more interesting if these kinds of features were integrated into the platform (much like Vine attempted to do before it was shut down). I don’t know if modern Twitter can change so radically to include long form content. Maybe that ship sailed a long time ago.
The key here is that when you have long-form content, it’s easy to place effective, ethical, and healthy advertisement on it. It’s easy to know your audience and what they’re likely to buy. This is what the big YouTubers do. Not only do they use the platform’s native advertising tools, but they also bring in sponsors that can speak to their audiences more directly. On YouTube, or on a blog, your advertisement is going to particular segment of people (and the Creator has the stats to prove it). You’re able to reach your audience because the audience already trusts the Creator of the long form content — and the audience is more likely to trust your product by proxy. In other words, on platforms with Creators that make long form content, you’re marketing budget is getting backed by the trust between the Creator and the audience. Anywhere else, and you’re just paying for “exposure”.
Twitter also isn’t immune from the problems other algorithmic news feeds have accumulated over the years. The robust third party developer community that Twitter started with is gone. Twitter seems focused on deprecating the existing streaming API which allows third party developers to run important features. Instead, they are replacing it with a Premium Account Activity API, which is very expensive, and will drive many developers out of business (many of whom publish their apps for free).
While I could spend additional time here dissecting Facebook, Instagram, Snapchat, and so on, all of these algorithmic news feeds are practically the same design. Many of them are ultimately the same flavor of business.
If you’re a software engineer or business owner at a media company, I’d encourage you to think deeply about the ethics involved with an algorithmic news feed from the perspective of promise theory. Make sure there’s a clear, sharp line between the Publisher and Reader — and communicate that relationship with your users, customers, and creators. Better yet, just use a chronological timeline for Subscriptions, and put your algorithms in a Discovery or Recommended section. Creators depend on you for their living. Advertisers depend on the trust creators establish with their audiences to reach eyeballs. And audiences depend on predictable mechanisms to get subscription updates from the sources they subscribe to. The last part is the most important — if the promise between Publisher and Reader is not kept, then your platform is a breeding ground for bad behavior. Finally, take extra care to ensure the integrity of epistemology in your platform so that users can identify the Source of the content. Give users the tools they need to understand who the Creator is and what their motivations might include. Allow users to verify the chain of trust themselves. I think if we can do that, and we can think more clearly about the Subscribe button, then we can make the world a better place.