Hierarchy Elevates Social Reasoning

In searching for where to seed our own marketplace, we realized our work on social decision processes was quite a bit more significant than we thought, both to our business and generally to how social apps are designed and built. The problem models central to our solution's main design features also point to clear, fixable mistakes we've been making industry-wide for twenty years.

From Social Finance to Social Reasoning

Let's skip past the tricky outer context. Our mission demands creating next generation social finance. To do better than just enabling a bigger Solar Roads, we needed better ways to qualify and create better campaigns. To do this, slow-roasting design work inspired by the nervous system of the broadclub cuttlefish was pulled off the back-burner.

In the middle of explaining what makes our combined solution better, the phrase making the money smarter had unconsciously popped out. Obviously we are not creating thinking money. We are only making the money smarter if we are making the people who spend it yield better results from crowd-sourced decision-making. We are talking about making crowds smarter.

Making crowds smarter is the more general tool. Social finance is just a powerful application of it.

What the cuttlefish had long ago inspired was intended to be a better Reddit, so the bigger potential wasn't lost on us. However, our improved social finance tools were technically easier to MVP. In addition, social finance is a viable package on its own: Enable otherwise impossible transactions and then charge transaction fees.

As a consequence of overemphasizing our social finance, we hadn't been talking enough about social reasoning. We had even been simplifying it out, boiling this feature down to its nearly degenerate form, technically just simple delegation. This kept us from talking about vaporware so we could focus on the concrete social finance prototype, but at length, we really need to be talking about and building the social reasoning vaporware.

Shifting the Spotlight

Adaptation provides a good moment to consolidate reasoning. During that consolidation, with extra annealing energy from the pressure to change, we realized that social reasoning can have both more powerful strategies and more compelling arguments built around it, ones that work out better for every single person involved. Rather than an incremental feature of social finance, our social reasoning was looking more and more like the dynamic tool while social reasoning was just something for that tool to strike against.

Social finance is the anvil for social reasoning, the hammer.

Social finance is the motivating problem that gives obvious utility to social reasoning. We had been forming an argument that people who want to buy things will need social reasoning to make their spending more effective. What had become somewhat lost on us was that social reasoning creates the things that they would want to buy. It is the better seed, the engine of creating all-new value rather than simply optimizing cost. People will show up to have fun creating things with social reasoning and then use social finance as a way to materialize what they create.

The resulting change is not radical. We are still combining next generation social finance with next generation social reasoning. We are still focusing on consumer software as an initial market, specifically open source, for which there are no good ways for consumers to effectively put money in. However, instead of merely building social reasoning as a dutiful expansion of social finance, we are going to be making it the centerpiece of our communication and strategy while also throwing a lot more technical resources at getting it to MVP.

They Didn't Read the Article

To understand how social reasoning is a step up from crowd sourcing and how it will apply to a wide range of problems, we have to rewind our VCR a little bit. In the 90's, many of us first heard about the internet on TV shows like beyond 2000. It was a top-down world. Mass media existed and was diverse, but to get in, you had to be someone who was doing something. It was exclusive, but it was qualified. It was forever too "mainstream", too out of touch with the true diversity of the audiences, but it was coherent.

The Web Opened Everything Up

At first, the web was extremely good at tunneling through the walls that guarded the esoteric. Instead of waiting for information to trickle down through the top-down funnels, it would appear online, globally routable by anyone. Repositories of institutional knowledge were moving outside of whatever internal departments had lorded over them.

"Democratizing!"

The read-write foundation for building dynamic web pages made it increasingly easier for resources that used to be cooped up behind insider relationships to be plugged into the internet and driven by all. In Bay Area speak, we love to call this "democratizing".

Quantity Did Not Bring Quality

Publishing also used to be the domain of a few. Web 2.0 happened and all of us could write. Write we have. Instead of high-quality, formerly esoteric information coming to light, we increasingly learned what the random person could see, what they think and believe. This kind of content used to be call-in talk radio. Far from letters to the editor, it was a somewhat perverse look into the minds of those who did not actually belong on the airwaves but had entertainment value.

It still kind of feels like we're democratizing. In addition to allowing anyone to play, we were allowing the wisdom of the crowd to pick winners, potentially circumventing old world entrenchment of the one-way top-down world. We will find that the way we have done this is instead decapitating.

How Social Killed Social Mobility

A hallmark of democratic society is social mobility, a two-way flow of ideas and people back up into society's hierarchies. By enabling anyone to become a new direct source of authority, it would seem that we are creating social mobility.

Representation Beats Participation

An early problem we identified with existing social finance was the O(M × N) complexity of asking M contributors to evaluate and click on N projects. The premise of social finance is that by dividing up the cost, we lower the expense to each user, but if we just multiply the work, it's not sustainable. It doesn't actually scale or make the time-cost cheaper.

People are unique, but there is a lot of duplication near the popular modes. We can substitute a few for many and get roughly the same effect. Due to the added efficiency and cost-benefit of being able to evaluate more deeply, a sampling representation can yield better results than exhausting all participants.

Besides the efficiency of evaluation, we also wanted to enable use sub-communities to move between creators and sub-streams efficiently. Sampling among users to check decisions made by representatives is again more efficient. Instead of needing each user to log in and click for such a community migration to occur, a few users can create the decisive action for later review, enabling things to move.

Representation Incorporates Expertise

Early adopters and experts are by definition rare, a limited resource. Without efficient representation, not only are the modes inefficiently expressed but the expertise is completely drowned out. Avoiding duplication brings the scales of available decision bandwidth into closer parity.

Experts are inherently more eccentric. They are not the same as representatives who share and can communicate the broader interests of the population being represented. In the principal-agent problem, representatives preserve interest alignment while experts can inform better agent decisions. You need both to have an aligned and informed optimum.

The key design point here is that the same structures that create efficiency of representation also enables expertise to localize into positions with more authority. There they can inform decisions of representatives who have slightly less expertise but better interest alignment.

Conversations between representation and expertise are high-level conversations. Large distances of belief and awareness can be traversed at low cost. That sounds a lot like what we want in the internet, the kind of emergent sophistication we expected in the very early 2000s.

We Replaced Hierarchies With Flatness

A great deal of our social applications are built on one-person-one-like systems. The conversations are globally routable and globally writable. There is no depth. There is no transitive authority assignment that can successively concentrate representation to make it fair & efficient. Nor can this concentrate expertise to make decisions informed.

The signal-to-noise ratio not only drops to zero for expertise but also drops to zero for any group targeted for brigading or some activism, even if some are just caught in the crossfire. Suffocating the participation that representation was made to legitimize & protect can hardly be described as democratic.

Instead of assembling alliances of shared and compatible interests, we create spaces that are dominated by whatever mass of will arrives. It sounds like the decision process that would emerge from a crowd standing on flat ground because it is.

Increasingly, we have diminished the role of old world systems while shifting the creation of legitimacy into new ones that have no capability to capture or represent any social elevation beyond follower counts. We cannot have social mobility in such non-structures because there is no upward dimension. This does not broaden or balance the flow of people on a two-way street. Flattening hierarchies degenerates social mobility.

Bottom-up without the up part is just the bottom.

Truth preservation doesn't converge to coherence when most users don't even know what deductive reasoning or axioms are, when people don't even read the article. More free speech isn't the cure to free speech that has no reasoning discipline or knowledge. As the framing decoheres, disinformation, misinformation, normative manipulation, and opportunistic sophistry all thrive.

It is not the random user's fault that they are not experts in every single field on Wikipedia. Counting on random users, not even average users, to perform high-quality evaluations used to drive feedback loops is a design decision below the engineering judgment used to ship kit planes, let alone a 737 MAX. Our profession has to do better.

We are perhaps well into the find out phase of this adventure. After nearly two decades of "democratization," we have Balkanized public consciousness into schizophrenic silos unable to process information to climb gradients of truth or to act decisively based on what they learn. To any extent that we may have been growing immunity to our own misinformation bubbles, we are also poised to enter a whole new era of "you can't trust everything on the internet."

The Only Way Is Up

Our fragmented public consciousness may explain the global flirtation with authoritarianism. You may crave a return to simplicity, a return to a coherent foundation of shared facts. Consider that we humans may have a primitive instinct in times of chaos to actively embrace demagogues as a form of emergency hierarchy, facilitating a fast re-establishment of consensus and the cooperation it enables.

To rebuild tech's incentive alignment and our own engineering integrity, what needs to be built is more vertical, more private because it is more personal, tools of connecting people who are inherently rare. We need to broadcast what expertise can synthesize, and amplify it through representation so that it can be accessible to those who need to see it. The two-way street needs a place to go up to. We must digitize the old models and, like the trailblazing Ebay, expose the new advantages afforded by digitization of old things.

One way or another, whether it is by kakistocratic promotion of loyalty over competence or by vengeful innovation powered by 90s millennial techno optimism, we are going back up.

The Non-Shallow Tech

The cuttlefish inspired concept is from spring 2015, while working on my first startup. We were in a hole, surrounded. The only way to survive was to find a breakout and run for it. I would later read Skiena's algorithm design manual to get ready for a Google interview. We hadn't made it out, but the chili on the back-burner never went totally cold.

Why Industry Likes Flat

There are decent reasons industry hasn't done things this way all along. Hierarchy is a bit harder to bootstrap because it requires user interactions to begin working. If feels counter to our beloved democratizing, as if mere direct democracy is an important free-standing principle rather than an unpopular method of democratic governance used by almost no one. When building MVPs, one-person-one-vote is cheaper. It is easier for users to learn and adopt. Perversely, the inefficiency of spinning wheels drives engagement, so it is likely more profitable for the attention economy.

Personalized Graphs

Besides surviving the many tar pits in social finance and social reasoning, once anyone gets to a viable solution, the way that technologies must be built is more complex than just boring CRUD. The first problem is that hierarchies involve multiple lookups or maintaining their materializations. There are cascading writes. If we have non-exclusive views of hierarchy because of plurality, we have cascading updates of graphs that are personalized. There's a bit of coupling between how the user behavior accomplish the social behavior and how the implementation will support the user behavior.

We will explore the details more as the subject becomes less vaporware. When there is a self-evident prototype working for regular users, it will be both easier to explain more obviously worth the effort of trying to understand.

Making It Concrete

By building PrizeForge, we are giving substance to our model of the solutions. We started with the fundraising side, with a prototype called Elastic Fund Raising. For our interim decision model, we are going to manually carry out many operations such as appointing delegates, monitoring payouts, creating streams, and writing each stream's use of funds. This is the doing things that don't scale phase for us.

Seeding the Supply Side

On any two-sided platform, users have nothing to do alone. Since it would make little sense for founders to act as the demand side, we do the work of the supply side in order to bootstrap. If you're gig economy, you do gigs until real gig workers can just react to demand that has begun gathering on its own.

Picking a Market

As programmers, Positron can build programs. Open source programs fit the problem case: we can make a lot of value for not much cost, but capturing enough value to be sustainable would be very difficult if using traditional models.

Validating

We used traditional tools to validate. Our CEO started off by building Emacs tools and tutorials, gathering about $200 per month with Github Sponsors. A specific $15 tier was created with a specific interest in "open innovation" rather than Emacs. We started developing a message. $15 sponsors showed up. The stage appeared to be set. It wasn't a paycheck, but we were using a terrible solution, so it will be better after we launch our own technical solution, right?

Invalidating & Adapting

The validation work looked scalable, but hasn't quite translated into enough of an initial jolt of energy for the actual service yet. While there are some obvious glitches and low-hanging fruit everywhere, another problem has been growing and will not get better: All of our software is written in Rust. While writing Rust day to day, there is almost no overlap in taking time off to do Elisp programs and videos.

If Only It Was All Common Lisp ☯

It would be tremendously beneficial if Emacs was programmed in a general purpose language, similar to how Lem uses Common Lisp. We could be writing our entire stack in CL while extending a still-popular editor in CL. (If you are a CL proponent, we have a stream raising funds for Lem!) For now at least though, that ship sailed many years ago. We have to play the hand we were dealt.

Hiring For Elisp 🤡

Another problem ahead was how to grow a team when almost nobody else programs both Elisp and Rust. Finding co-founders and early hires is already somewhat miraculous. Let's not steer into headwinds by making it even more miraculous.

A Spectre of Wisdom

The last headwind is difficult to tease apart. There is a deductive argument that programming tools can add a lot more value than they cost, so programmers are financially interested in these tools getting a lot better.

So You Want Us to Pay For Work We Do?

At the same time, there is a market fear, an apprehension that we may just be constructing a Ouroboros, a snake eating its tail, one our company can tax and control. Confluent with that fear is the deep mistrust of commercialism that is heavily concentrated in open source user communities, especially within the "free-as-in-beer" underwing of the "free/libre" advocacy, who in turn provides the inoculation against dangerous thinking such as spooky Open Source.

Shouldn't We Steer Into Risk?

This is a case where dedication to the deductive angle might have us look past all the subjective and wrong thinking. Shouldn't we just write off the irrationally apprehensive and focus on people who get it, the people who want it? Doesn't the value-to-cost ratio let that readily happen and then turn silly heads later? Don't all startups press through at least some doubts of the market?

From The Outside, Flow The Revenue Must

A lot of programmers rightly feel that the revenue needs to come from the consumers and businesses, that building a GDP within the 0.5% of programmers and the 0.1% who has money to light on fire (we do), while perhaps viable, is ignoring so many lower risk, much easier plays, ones that pull revenue in from the outside, from the people who depend on us. When money comes in from the outside, then we can spend it on tools.

Do Right or Do Not

The wisdom is that, while every venture does have to engage with uncertainty, pushing directly into an unnecessary uncertainty that can be avoided is itself a compromising behavior, a source of mistrust. It heralds the beginnings of bad partnership. If it can be turned away with the calm words of cool patience, it very much should be turned away, softly, because if we are paying attention and honest, we will get it on our own.

From Audio to Visual to Bootstrapped

We almost desperately need a stream that can justify users providing & refining input that will prioritize feature work and drive product development from the demand side. The demand side can then reward the supply side to make it happen. PrizeForge itself is somewhat of that meta stream already, but that is itself a promise of an anvil without a compelling cause.

Relieving the Venerable ProjectM

To break the logjam, we're building a music visualizer. Rust is good at high-performance apps that need to talk to hardware and compile on a variety of platforms. Our CEO prototyped a visualizer already in about 2018 before getting hired to write other Rust. (If you have used the Laplace distribution in the probability crate, you are welcome! That is one of the tangential inspirations of the PrizeForge Nabla ∇)

Music visualization is a great problem. We can make something interesting for everyone from career programmers to casuals who just want to trick out their infotainment systems. The best open source alternative is ProjectM, a twenty year old program that probably hasn't been updated in ten.

Tech Readiness Better Than Ever

In the intervening time, new possibilities have made this opportunity look a whole lot better. Mixing generative and procedural techniques is within arm's reach. Steel scheme looks viable for embedding a programming language for scripting presets. Rust Vulkan libraries are more mature. Rust itself is more mature.

Fixes Our Strategy & Tactics

This change of direction fixes a lot of problems. It puts us using our most valuable language. It gives both us and potential team members a way to gain trust by coding on something cool. It brings us into direct contact with demand for small, local, truly open generative models. It sets up the marketplace to be truly two-sided instead of being full of programmers trying to scrape a living off of other programmers. It gives us a stream for a project that needs to be defined by community feedback, not just paid for with community funding. That definition needs continuous improvement through social reasoning.

To begin this shift, we're opening up a stream for the visualizer, µTate (MuTate). The rough technical sketch is in the use of funds. The repository is up. We will be posting updates on our YouTube. Alongside developing µTate, we will be developing tools to communicate to improve its use of funds definition. This will give truly expert users ways to provide feedback and reconcile tough problems without feeling like they are talking to a swamp on Reddit or expected to pay for service that they provide.

The Beginning Feels Near

The company has been laying low in Busan. In some ways it feels like this is not yet the beginning. This is the beginning that nobody sees, the real beginning, working from a studio apartment, grinding down uncertainty, finding ways to make the problems into ones that more code or any code can fix.

Seoul My Soul

As the temperature drops, it is as if it is time to return to Seoul. In any city that dense and expansive, the pace of events is an offline internet that brings new, stronger connections at a faster rate than anything possible online. It is winter, and it will be harsh as usual, punctuated by Siberian fronts that add to the feeling of fighting for survival.

The journey may have arrived. Along with µTate attracting contributors, we can lure Rust engineers out to social events, teach them to make the best brownies on Earth, and recruit them to work on our entirely Rust stack. We can physically bump into likely customers from every addressable market in every kind of media and manufacturing. Many people here are either wildly underserved by local apps or by big overseas vendors since open source adoption is lower, held up by language barriers until now.