From Personal to Social Knowledge Graphs: a vision statement
update: this June 2022 article formed part of the basis for a set of 3 interconnected articles published on 1/1/2023, summarised here:
My chapter for an upcoming book on Personal Knowledge Graphs (PKG) will encompass each user’s PKG, the Social Knowledge Graph created by networking them together via the Fediverse, Solid hosting, AI writing tools and Decentralised Autonomous Organisations. This post provides a first draft of its Introduction. Help me make it better.
Provisionally entitled “A decentralised social network of Solid-based second brains”, the chapter also describes the future of MyHub.ai, the publishing platform alpha-launched in 2020. Given the book’s subject, my chapter will focus on the Personal Knowledge Graph (PKG) at the heart of each Hub, and the Social Knowledge Graph created by networking them together.
if a Personal Knowledge Graph is to be useful it cannot exist in a vacuum
My thesis, however, is that if a Personal Knowledge Graph is to be useful it cannot exist in a vacuum. The chapter’s Abstract and Introduction, provided below, must therefore also describe the other components of the MyHub.ai ecosystem, show how they fit together and add value to the above Graphs.
Warning: this post is a “brain dump first draft” of my chapter’s Abstract and Introduction, so it’s long. Some of this content will probably transfer to other parts of the chapter in later drafts. I aim to publish subsequent posts over the summer as I explore each ecosystem component in more depth.
I’d love your comments and suggestions. If you don’t have a Medium account, post me your thoughts on Twitter (and soon, Mastodon).
Every “second brain” built on a Personal Knowledge Graph (PKG) makes the same promise: that each user’s confusing hairball of notes, ideas, bookmarks and draft posts will magically and effortlessly resolve into new insights and publication-ready articles.
PKGs to date, however, do not have a “grain”, or design principle, specifically supporting the creative flow from idea to research to draft to publication. As a result, that magic fails to materialise for all bar the few obsessive productivity geeks prepared to invest in massively customising PKGs into something actually useful for writing and thinking. For a longer explanation of why, please see Justin Murphy:
Personal Knowledge Management is Bullshit
Personal Knowledge Management ("PKM") is a trendy new term for techniques and applications designed to manage…
This explains why today’s PKGs are usually just one of an array of tools which thinkers, particularly writers, must cobble together for discovering, bookmarking and reading content; developing ideas in private and discussing them in public; writing, publishing and sharing.
PKGs are usually just one of an array of tools which thinkers, particularly writers, must cobble together
This paper describes an ecosystem to support all of this and more, bringing the benefits of PKGs to more people. The ecosystem will simultaneously provide its users with an AI-powered, PKG-based thinking tool, a publishing platform, and a decentralised social network, enabling distributed Social Knowledge Graphs on the Fediverse. This ecosystem, moreover, will be published on Solid to ensure each user’s privacy, financed by an innovative, ad-free business model and managed (maybe) by a DAO in which all users will stakes in, proportionate to their activity.
My first braindump of the chapter’s Introduction follows
Ecosystem at a Glance
I launched the alpha version of MyHub.ai as the world went into lockdown. There was no point trying to compete with COVID-19 so I hunkered down, kicking the tyres of my Hub and blogging what I found.
That brought me new networks and ideas, leading to the much more ambitious vision summarised in this chapter:
Figure 1: Ecosystem at a glance
Figure 1 includes several key elements, each explored in more detail later:
- Thinking Management Tool (centre): each is a Personal Knowledge Graphs (PKG), and is either hosted on MyHub.ai or self-hosted using the MyHub open-source toolkit.
- Each PKG is designed around a content pipeline, and provided dedicated writing productivity tools to help its Editor transform ideas, notes, bookmarks — all the messy content of a Personal Knowledge Graph — into published articles.
- Published articles appear on the PKG’s public-facing edge: the Editor’s Hub, blog, newsletter or other personal website
- Both the public sites and selected content from the private Libraries are networked together on the Fediverse using the ActivityPub standard, creating a personalised, decentralised Social Knowledge Graph of both Followers and Friends
- The Thinking Management Tool also provides natural language processing services from AI algorithms, which operate both on individual PKGs and across each Editor’s Social Knowledge Graph
- All data is stored on Solid Pods to maximise each Editor’s privacy and control over their data
- The ecosystem is driven by an innovative, ad-free business model, and managed (maybe) by a DAO.
While the first tweaks to the public Hub template are underway, this chapter sets out how the above concepts — PKGs, Hubs, Fediverse, AI, thinking tools, Solid and DAOs — fit together to create something bigger than their sum.
There’s no simple path through this highly interconnected set of components, so to introduce them I’ll start with the social network you’ve probably never heard of.
What’s the Fediverse?
Fediverse: a universe of decentralised, federated social media apps, including Twitter-like Mastodon, the YouTube-like PeerTube and the Facebook-like Friendica.
Unlike today’s walled social gardens, however, Fediverse apps all use the same open W3C standard (ActivityPub). So what?
- any user on any Fediverse app can follow and interact with users on any other Fediverse app, on any other server. This would be like if my Twitter account followed your Facebook account, yet we could still interact seamlessly, from inside our preferred app.
- that’s because Fediverse apps are not owned by surveillance capitalism corporations who want to trap you in a walled garden and feed your privacy to advertisers: anyone can launch a Fediverse server, just like anyone can set up an email server and send an email to anyone else, on any other email server, using the open email standard.
- while users on each server can talk to anyone on any server, each server can have its own rules (privacy, hate speech, etc), rather than enforcing one content moderation standard on 3 billion people. And if you don’t like your server, you can move to another one without losing touch with your friends, so noone’s locked in.
Point (1) means the Fediverse is one single large network, while (2) & (3) allow Small is Beautiful communities within it (more: my public overview on the Fediverse).
the Fediverse is one single large network, but allows Small is Beautiful communities within it
These and other advantages make the Fediverse a valuable alternative to today’s social media… if anyone would join it. But it hasn’t taken off, perhaps because most Fediverse apps imitate existing social platforms, while offering fewer people. What the Fediverse needs is something original.
Which brings me to MyHub.ai.
Today’s alpha version
My Hub.ai is currently a platform where you build your personal Hub — a new sort of personal website.
Your Hub brings together everything you Like (the stuff you’d recommend your friends and followers should read), everything you Think (your own blogposts and other articles, wherever they’re published) and everything you Do (your personal portfolio).
Everything is tagged, and the natural language search interface makes it easy to create Collections like: “The Best Stuff I Think tagged #social media” or “Everything I Like, Think and Do tagged #creativity and #productivity”.
Tomorrow’s open-source toolkit
Today, of course, only the alpha version of MyHub.ai exists. In the future, as set out below, a Hub will just be one of several templates available, along with blogs, Substack-style newsletters, traditional websites and more.
All of these sites will be managed using a writer-friendly, PKG-based thinking tool, hosted on Solid to maximise the Editors’ data ownership, and networked onto the Fediverse.
The templates and thinking tool, along with the connectors to Solid Pods and the Fediverse, will be open-sourced, creating a MyHub.ai toolkit which anyone can use and build on, creating the ecosystem described in the Abstract.
The Fediverse, Solid and the alpha version of MyHub.ai already exist. The rest of this chapter therefore focuses on what remains to be built.
Your Hub’s CMS: a Thinking Management System for writers
Each Hub’s content management system (CMS) is actually a “Thinking Management System”: a thinking tool based on a Personal Knowledge Graph (PKG) which is custom-designed to support thinking and writing.
A thinking tool for your CMS
Why not make your website a seamless extension of your Second Brain?
There are as many thinking tools (Obsidian, LogSeq, Roam, Evernote) as there are names for this software class: some call them thinking tools, Tiago Forte calls them Second Brains, Cory Doctorow prefers Memex, others note-taking apps or digital gardens (more: my public outline on thinking tools).
Whatever tool they use, creators are currently supposed to do their thinking in them, as well as perhaps prepare an initial draft of whatever they intend publishing. But then they have to finalise that draft and publish it using a completely separate tool — their website’s Content Management System (CMS). And then they need to share the published piece via yet more tool s— their social media platforms of choice.
make your public website a seamless extension of your private, AI-powered thinking tool
But why not make your public website — and hence the writing on it — a seamless extension of your private thinking tool? By turning your Hub’s CMS into an AI-powered Thinking Management System (TMS), you can manage everything you have — your bookmarks, notes, ideas, draft and polished posts — in one place.
Set any content to ‘Public’ and it appears on your Hub… and in the Inbox of any Hub (or anyone else) Following you via the Fediverse. And as we’ll explore later, you can also set some content to ‘Friends’, allowing your social network to access selected content — e.g., your half-finished thoughts, or subscriber-only content.
A PKG which supports writing
Today’s PKGs do not have a “grain”, or design principle, specifically supporting the creative flow from idea to research to draft to publication.
In theory, the PKG-based thinking tools mentioned above might appear to support writers “off the shelf”. Unfortunately, they do not for all bar the few obsessive productivity geeks prepared to invest in massively customising their PKG into something actually useful for writing and thinking.
The reason is simple: PKGs deliberately mimic how (their creators believe) the mind works, providing an environment where every note is linked to many others without any sense of hierarchy. As others have observed, however:
“everything in your mind is already … a graph… big and complicated, with way too many connections everywhere. There’s no good in replicating that in digital form… the greatest authors have always … cut through the knowledge graph with a bold and forceful line”
- Personal Knowledge Management is Bullshit (Justin Murphy)
The thing to remember is this: at the end of the day, the only person who can read your knowledge graph is you. For everyone else you must provide a linear text, not a messy collection of unstructured notes.
The only person who can read your knowledge graph is you… But getting from a messy graph of interlinked notes to a linear text is hard — it’s called writing
But getting from a messy graph of interlinked notes to a linear text is hard — it’s called writing. A writing tool which presents you with an unadorned “hairball” of notes in a PKG is not much help.
The PKG-based Thinking Management System underpinning Hubs — and any other site built using the MyHub.ai toolkit — will be designed specifically to solve the above problem by supporting Murphy’s “bold and forceful line”.
Content pipeline: providing the through-line for writers
Pipelines are common metaphors in the professional world. One way the Hub’s Thinking Management Tool helps you transform notes and ideas into fully published texts is by supporting each stage of the Content Pipeline.
Like any pipeline, it illustrates a process, with inputs coming in from the left, and outputs leaving to the right.
Figure 2: Content Pipeline
A complete explanation is provided later in the chapter, but for the purposes of this Introduction I need to explain the following pipeline elements:
- public content (right) appears on your Hub or other public site.
Not shown: While Editors can of course share wherever they wish, all content is automatically shared to all Fediverse accounts (Hubs, Mastodon, etc.) following the Hub. Each Hubs also has an enewsletter, and automatically creates an RSS feed for every collection of content.
- The rest is private content. Most of it resides in your Library: both your own notes, overviews and drafts, as well as high-quality content you have curated from elsewhere, with your notes. Everything is tagged.
- Some of that curated content comes from your Reading Queue — stuff you’ve identified as worth reading. You store it in your Library by tagging and (ideally) annotating it in your words.
- Some of content in your Reading Queue, moreover, comes from your Inbox. Content appears there because it was published by one of your Priority Sources (not shown): an external content stream (eg enewsletter, Twitter list, Fediverse account) which the Editor has prioritised because it provides particularly valuable, relevant content to an area of interest (eg, a subject the Editor is researching and writing).
- Other content also gets into your Library by curating it direct from the source website, using your Hub’s bookmarklet (see MyHub FAQ).
- The entire process is managed using the Thinking Management Tool, which provides a range of supporting AI algorithms (not shown).
The Thinking Management Tool thus helps Editors ‘push’ the most valuable content from their Priority Sources through the pipeline from left to right. At each stage, more value is added to it: it is filtered, annotated, tagged, combined with other content (eg original notes, other relevant curated notes) and eventually transformed it into the Editor’s own published articles.
A Solid Social Knowledge Graph
By networked these second brains together via the Fediverse, the MyHub.ai toolkit creates a distributed Social Knowledge Graph for each Hub Editor.
Figure 3: My Social Knowledge Graph
Friends and Followers
In the previous section I mentioned that any Fediverse account can be a “Priority Source” — ie, if you Follow a Mastodon account, all toots (or, optionally, just the links) published by that account will appear in your Reading Queue.
As each Hub has a Fediverse account, your Hub (“Hub A”) can Follow Hub B and Hub C, so their outputs (what they publish on their Hub) become your inputs, appearing in your Reading Queue. This Follow relationship is, like Twitter’s, one-sided: A may Follow B, but B doesn’t necessarily Follow A.
But there’s also a two-sided, more trust-driven Friends relationship, like Facebook’s, where you make some of your Private content accessible to your Friends, and vice versa. This could be your jotted ideas, article drafts (i.e., “working with the garage door up” — Andy Matuschak), your notes on other people’s content, subscriber-only content, or all of the above — you select what you share with your Friends, and they select what they share with you.
Multilevel social knowledge graph
In both cases, you will access this content via your Thinking Management Tool — Figure 3 shows:
- your Hub (centre)
- content published by your Priority Sources — Hubs and other Fediverse accounts you Follow, RSS feeds, Twitter accounts and lists, enewsletters — entering your Inbox (left)
- the content you publish entering your Followers’ Inboxes (right)
- centre: from inside your Library, you can also access relevant content inside your Friends’ Libraries which they selected to share with their Friends, and vice versa.
When you go into your TMS, therefore, you can choose to search and use just your Library’s content, and/or the published content of the Hubs you Follow, and/or the notes and ideas your Friends share with you. The latter includes both their own original thoughts, and their notes on other peoples’ content.
your personal Knowledge Graph, connected to the Knowledge Graphs of those whose judgement you respect via a writing-oriented thinking tool
The MyHub.ai Social Knowledge Graph is thus your personal Knowledge Graph, connected to the Knowledge Graphs of those whose opinion and judgement you respect via a writing-oriented thinking tool.
‘With “tools like Notion, Airtable, and Readwise … people are aggregating content … reviving the curated web. But these are mostly solo affairs… fragmented, poorly indexed…[while] the websites at the top of Google are not necessarily the highest-quality … but rather the ones that put the most effort into SEO”. Rather than try to “organize the world’s information… [we should] organize the world’s trustworthy information” — which is more likely when you combine the curation efforts of millions of people whose own credibility depends on the content they curate.’
A data union, built on Solid foundations
The entire decentralised network of Hubs, finally, is hosted in Solid pods.
Solid is a (relatively) new W3C standard that “lets people store their data securely in decentralized data stores called Pods… When data is stored in someone’s Pod, they control which people and applications can access it” (solidproject.org).
Building the MyHub ecosystem with Solid turns Hubs into a decentralised social network where users control their data.
Moreover, they can choose to share some of it to both benefit from and train the ecosystem’s AI engines. This means that they join a data union, where they pool their expertise and effort for mutual benefit. Which brings me to…
Where’s the AI?
Each Hub’s Thinking Management Tool accesses — and, in the process, trains - a suite of AI algorithms to help Editors manage their content. But the real magic happens when the AI tools operate across a user’s Social Knowledge Graph.
The following video, from 2020, focuses on an autoclassification (“auto-tagging”) algorithm:
Autoclassification, however, is the tip of the iceberg. Because the same autoclassifier will help tag content created by different users, content discovery will be immensely powerful, with recommender engines unearthing useful content for each user from acriss their social knowledge graph as they work in their thinking management system.
Once a user has selected relevant content, autosummarisers could help them make sense of it, creating a short briefing of all content relevant to the user’s needs. These briefings could also be provided as an auto-created, customised subscription service — eg “Send me a weekly summary of everything relevant to X and Y added to my Reading Queue and Library, the Libraries of my Friends, and those I Follow.”
Other ideas developed within months of launching MyHub.ai include integrating factchecking APIs to rate content trustworthiness and spot disinformation, machine translation to help content find new audiences and create multilingual communities, filter bubble analyses to help Users break out of their echo chambers, text network visualisations to help users explore links between content across multiple Hubs, supporting techniques like progressive summarisation and spaced repetition to help users better learn, and many more. These are explored in more detail here:
Imagining new MyHub.ai features as the pilot Hubs launch
This month the first pilot Hubs launch, so it’s time to imagine what comes next: AI auto-tagging? Filter-bubble…
But it’s what I’ve discovered since then that has me most excited. I’ll be developing my notes into the rest of this chapter over the summer.
AI as a service
Most of these algorithmic services are trained as the Editors use them. And highly trained algorithms are valuable.
The 2020 video, above, gives one example of this: the auto-classify algorithm is trained both by the Hub Editors as they use it and (due to the unique Hub interface) the Hub visitors as they explore content. Trained by millions of Hub Editors and many more visitors every month for free, such an autoclassifier could be monetised as a Software as a Service (SaaS).
But, again, this is the iceberg’s tip: auto-summarise services, for example, could be refined by watching how Hub Editors summarise the content they manually curate and how they use both the auto-summarise and progressive summarisation services. A spaced repetition service, on the other hand, would create a fascinatingly valuable decentralised database of Q&As on a huge variety of topics.
no need to mine users’ data or sell ads
These and other products and services can be monetised as SaaS, so the MyHub ecosystem doesn’t need to mine users’ personal data to sell ads.
Premium services and platform
Almost everything bar the AI algorithms themselves will be provided as an open-source toolkit, allowing anyone to set up their own Thinking Management System, public website, Solid hosting and Fediverse account on their own server. If they wish, they can connect these to the AI algorithms and other services, but that would be purely optional.
But there’s also room for a platform which takes care of everything for you for a small subscription fee. With a click it would roll out the basic Solid infrastructure, Thinking Tool, your choice of public website (Hub, enewsletter, blogs, collaboration workspaces, etc) and Fediverse presence, all integrated together with the relevant AI services. A platform marketplace would moreover allow developers and designers to market new apps and templates.
The Editors using those AI services, finally, will not only benefit from the AI services they help train, they should profit from it. But how will we reward Editors for their contribution? Read on…
What’s a DAO, and why’s there a cryptocurrency?
The final piece of the puzzle is a Decentralised Autonomous Organisation (DAO) and its cryptocurrency for managing the platform. I am still weighing up whether this is cost-effective.
Explaining DAOs and cryptocurrencies would need at least two more blog posts, so check out my public overview. Briefly:
- DAOs are decentralised software programmes run on the blockchain, which means they can carry out preprogrammed operations without human intervention. Among other things, this enables perfect strangers to work together in a climate of trust towards shared goals.
- Many of those operations involve manipulating an associated cryptocurrency: tokens, stored on the blockchain.
MyHub DAO will run our new platform, offering users:
- Solid hosting services and editing tools,
- basic & premium templates (Thinking Tools, Hubs, Blogs…)
- AI algorithmic services (auto-classification, auto-summary, etc.), trained by users and visitors as they use the platform, and sold to external clients.
My Hub tokens are the platform currency: basic templates cost a few tokens, more advanced templates and AI services cost more.
However, users can earn these tokens as they use the platform, rewarding them for the training they give the AI. Basic usage should earn enough tokens to pay for a basic template. Advanced users can also earn usage tokens by developing their own templates and selling them via the platform’s marketplace. And the Hub’s “subscribers only” option will let them sell access to exclusive posts, enewsletters and curated collections.
users can earn tokens as they use the platform
Apart from buying usage tokens to use the platform, users can also buy tokens to access the SaaS (i.e., use the DAO’s AI in their own systems). As the quality of the algorithms improve in a virtuous circle, token value will rise.
Figure 4: Ecosystem and revenue streams
The DAO thus has two real-world revenue streams: selling platform usage, and selling API access. It can invest that revenue in platform development, buy back value tokens to increase their value, or distribute it to tokenholders. Usage token holders will have a say in these decisions.
Conclusion: OpenWeb from A-to-Z
So that’s how you build a AI-powered, open-source publishing and thinking tool platform, networked on the Fediverse, underpinned by Solid, and managed by a DAO which shares decisionmaking and revenue with platform users.
With data stored using one W3C standard (Solid), and content shared using another (ActivityPub), the entire platform is OpenWeb. allowing anyone to use the publishing tools we develop to set up their own publishing system.
However, I’m betting most will use the platform, both for the convenience and for its added-value AI support.
While the platform will launch with just two “premium apps” (Hubs and Thinking Tool), this will be an open platform: the platform’s marketplace will not only allow developers to provide free and premium Hub templates and add-ons, but also entirely new apps: classical websites, blogs, enewsletter sites and other forms of public digital garden.
Contribute, connect and stay informed
The above is a first “brain dump” draft of the Introduction of the Chapter I’m writing for an upcoming book on Personal Knowledge Graphs (PKGs).
I’d love your comments and suggestions. If you don’t have a Medium account, post me your thoughts on Twitter (and soon, Mastodon). You can also:
Get a Hub: The first batch of Hub Editors will keep their Hub for life, although the premium features I have planned (AI support, Solid hosting, etc.) may require coins in the future. There are (as I publish this) a few dozen free places left, so sign up for a basic Hub and get it free for life.
Read more: start at everything I #LikeDoThink tagged #myhub, and refine from there, particularly:
- MyHub faqs are tagged #myhub and #faq
- what I’m thinking and doing about MyHub.ai is Stuff I Do or Think tagged #myhub
- of these, the work on the PKG book is tagged Stuff I Do or Think tagged #myhub and #PKGbook
- if you sign up for a Hub you will get the MyHub newsletter
- if you don’t, sign up for my personal newsletter. All its editions are published on the Hub, tagged #newsletter, which means that there’s also an RSS newsletter feed (here’s how you can get an RSS feed of any collection on my Hub)
- the PKG Book is @PkgBook on Twitter