- Unsupervised Learning
- Posts
- How Twitter Should Work
How Twitter Should Work
Twitter’s central concept should be curation.
Curation should run throughout the product’s DNA like “quality results” does for search, and the reason for this is simple: people don’t have the time or energy to manage precisely what they want from the entire planet earth..
And that’s what Twitter is. It’s realtime humanity. All of it.
So let’s approach this exercise as if we were designing Twitter for the first time. If it were pitched properly, with the scope I’ve given above, what would it do for us?
We’re not fixing Twitter, in other words. We’re thinking about what it would do if it were capable of magic (which it is).
Starting with preferences
I think the first step might need to be something similar to what the music services are doing. You basically train Twitter into knowing what you like getting from the world.
Examples are best, so let’s take me:
I like infosec news
I like infosec commentary
I like infosec insights
I like political insights
I like philosophical discussion (but not very often)
I like political news (but very little, of high quality)
I sometimes like to be inspired by Twitter, where it exposes me to new art or writing
I sometimes like to be entertained by Twitter, where it shows me something hilarious that’s going on, or that someone said
I want to be informed by Twitter. I love the idea that Twitter can tell me if something just happened, and then transport me there through the eyes of the people experiencing it
Ok, I think this is pretty typical. People have interests and they have moods and they have requirements. The trick is knowing when to expose the user to which, and in which way.
And then building a UI that facilitates that.
This isn’t a UI issue, though. It’s an “understanding what the user wants from a magical pane of glass” problem.
It’s predicting, based on their preferences, what would delight them when they see it. And that brings us back squarely to curation.
We need a realtime interface
So the first thing we need is a “realtime” tab, or button, or pane, or whatever.
[ NOTE: If I were an Illustrator guy I’d try to mock it up. I’m not an Illustrator guy. ]
This section would walk the line between firehose and water faucet. You could drink from it, but it’d feel exhilarating to do so. Events are basically streaming in, which are already pre-selected based on what you’re likely to be interested in (with some overlap for new exposure), and you’d have the ability to explore each event in more detail.
You can imagine something like the WordCloud concept, where more important or trending topics are bigger, or bolder, or have more visual appeal in the GUI.
[ NOTE: This trending/focus feature needs to automatically factor in how loud and obnoxious the “OMG I can’t believe she’s wearing that.” effect, which by pure volume could dominate a stream. This is why the tuned preferences bit is so important. For a professor of Political Science at UCSB, trending topics should look way different than for someone who loves working at FootLocker because they see the new shoes before everyone else. This lack of personalization is severely limiting Twitter right now. ]
Adjusting curated exposure based on context and mood
This next bit is crucial. Users should be able to select a VIBE from any stream.
So if I’m watching, let’s say, a presidential debate, I want to have my stream popping in with a wide array of interesting content from that event—again, like the barely-drinkable firehose.
But within that GUI I should see options like “humorous”, “insightful”, “most popular”, and “controversial”. These are curated contexts that are associated with moods and vibes of people consuming the content.
Let’s say I’m trying to figure out if Trump really is serious, and if he really could lead our country. At that very moment I don’t want to hear about his hair. Or about all the outrageous things he says to get a reaction from the media.
What I want to hear then is concise wisdom from random people or paid pundits who are actually doing analysis on what’s being said.
And within a flash I could be turned off by that, or made tired by it more likely, and want to see people make fun of what he’s wearing.
Here’s the important bit: That context change should affect how I see everything else as well.
When you know what a user likes, and you understand moods, you can tag streams and content types based on how they will match those contexts.
So if there’s another event, say a sporting event that you’re watching at the same time, the larger context switch from “serious” to “playful” (or whatever Twitter refers to it internally and presents to the user) will determine what types of tweets it’s showing you about the game. There’s a big difference between,
Unsupervised Learning — Security, Tech, and AI in 10 minutes…
Get a weekly breakdown of what's happening in security and tech—and why it matters.
This is why their defense continues to dominate.”
and…
My cat could have made that play, and judging by the smell in the room—she just did.
And it’s tricky, because some guy in Egypt could be both a serious political blogger AND really like Hip Hop videos. So Twitter can’t just take everything he does and inject it into the [political/serious/blogger] context of realtime listeners.
No, they have to know the difference between content creation vibes/contexts and content receiving vibes/contexts. So they can match them.
Now that sounds hard. It is hard. But Twitter’s problem right now doesn’t seem to be how hard this is. Twitter’s problem is that it doesn’t realize this is the problem it should be working on.
Curation through and through
So that’s the main idea, illustrated through the realtime tab.
People have no idea what Twitter is, or what it should be. So they need to up their game with both their pitch (simple + expansive):
Humanity in realtime…
…and simultaneously delight users by showing them what they want to be looking at in the world without them having to think about it.
This means constantly learning more and more about the user. Foods, friends, favorite sports teams, favorite celebrities, favorite political topics, favorite actors and TV shows, etc.
So now when their favorite actor, from their favorite show, starts streaming live from on set, they can be notified of this as an event—if they’re not in the middle of an INSIGHTFUL context reading CSPAN’s coverage of internet security reforms.
Understand the user
Magically provide them the exact content they want, exactly when they want it
Delight is all we’re asking for. Actually, no. Nobody’s asking for it. Not the users anyway. Investors are, though, and this is what will deliver it.
Curation is the key, and you can’t have good creation without knowing who you’re curating for.
That’s how to fix Twitter.
Notes
Power users who want to manage their own lists and such should be allowed to do this to whatever degree they desire. The product shouldn’t limit people who wish to do this. It simply should not (and cannot) be a requirement.
The new Moments feature is definitely moving in the right direction. Unfortunately that only means “not backwards” right now, given how off-mark the product has been for the last few years.
If you’re upset about the information on the user that will need to be gathered in order to make this a good experience, prepare to be unhappy with the future of the entire internet.
Another idea that just came to me, which is related to my Universal Daemonization idea, is that services shouldn’t be independently collecting preference data. That should be part of ones’ ecosystem, e.g., Apple or Microsoft or Google. They know who you are, what you like, what your mood is, etc. And they can should be able to provide that information to services, with your permission, as required to improve said services. It should be done in a secure, standardized way so as to avoid the current, chaotic default of “let’s just start over and give this same content, in a whole new way, to a completely different service.” For privacy wonks both still suck. But welcome to curated technology and services. It’s not going away, so let’s do it right instead of wrong.