Fucking Around and Finding Out

On Elon, Elections and Emergence.

Richard Feynman once said that if all of scientific knowledge were to be erased, and he could only preserve one sentence, it would be: “that all things are made of atoms — little particles that move around in perpetual motion, attracting each other when they are a little distance apart, but repelling upon being squeezed into one another.”

Feynman’s sentence is predicated on the same idea that animates the XKCD comic above: that science is fundamentally a reductionist project. It’s about taking big things, breaking them down into small things and understanding how those work.  That idea is taken as an article of faith in most of science.  But there’s an alternative view that’s been gaining currency for the last few decades.

In 1972, Nobel Laureate and Physicist PW Anderson published, “More is Different.”  Anderson’s argument was that there are things that happen to big systems which we could not, even in principle, predict just by looking at the properties of their parts.  This idea that the sum is literally greater than its parts is called “emergence.”  To cite a few canonical examples: we can’t predict the wetness of water from hydrogen or oxygen atoms, we can’t predict life from carbon atoms, etc. 

Wait.  Why are we talking about physics?  

Well – because the implication of “more is different” is very relevant to human systems.  We cannot, even in principle, predict what is going to happen when you get millions of people together online and ask them to communicate, trade or govern themselves.  As a system gets progressively more complicated, theory ceases to accurately describe systems.  The only thing to do is, to put it bluntly, fuck around and find out.

And this seems like a good week to talk about this because we are witnessing two attempts to govern complex, emergent systems with their own kind of feedback loops. First, there’s Elon who has been fucking around and finding out on Twitter.  Second, there’s the American midterm elections which offer a report card on the fucking-arounds of Joe Biden and the Democrats’ government. (Ed note: And though we won’t talk about it here, it seems SBF did his own bit of costly fucking around and finding out…)

The problem with both loops – a speed running autocracy and a messy democratic election – is that they’re broken.  Good feedback loops – the kind needed to experiment in and course-correct emergent systems– are pretty easy to describe.  

You need: 

1. The ability to accurately measure an outcome (Did this change result in more revenue?);

2. The ability to isolate the impact of your initiative on that outcome (Was it really our project that made us that revenue?);

3. The ability to do this on the smallest, viable timeframe (How long will it take to learn this?);

4. The ability to iterate if things aren’t working; (Do we have accountability/control to change course?)

They’re just very difficult to implement.  Getting all four criteria for a feedback loop right is subtle, delicate work.  But then: Elon.

Elon and Unaccountable, Unmeasured Speed Running

A few weeks ago, the House GOP Judiciary Committee tweeted this (for reasons surpassing understanding and further proving we live in the strangest timeline):

House Judiciary GOP @JudiciaryGOP

Kanye. Elon. Trump.

 

It’s a weird tweet (for a number of reasons).  But it might also be kind of insightful?

Trump would throw out half-baked thoughts. He would then see how the blowback erupted. It was a destabilizing and exhausting way to run a government. But it was an extremely efficient way to get feedback.  

Elon is playing the same game. Twitter was an infamously slow company. A decision on broadening blue-checkmarks or what to charge for them would have taken months – maybe years to ship.  It might still have missed the mark.  But Musk’s philosophy for learning and guiding emergent behavior is simple: dumb and fast beats smart and slow.

And that’s how we get user research happening in real time with best-selling horror author Stephen King:

This is product management by natural selection.  After all, natural selection is just nature testing random shit and seeing what works.  And that gave us Keanu Reeves.  So it has something to recommend it.  

But lest you think this is just an Elon-fan-post, I don’t for-a-second believe that Musk’s “damn the torpedos” attitude is going to let him solve problems that lots of smart people have tried and failed to address. And that’s for one simple reason: his feedback loop is also busted. 

In natural selection, genes combine and mutate to test our new traits.  New, better combinations win out as nature weans itself through the competition to survive. As a result, nature learns and improves. 

Can Elon? The problem with being a dictator surrounded by loyalist fanboys is that no one tells you you’re wrong. 

Elon is smart. He wants signal. I think that’s why he is so active on Twitter. But he’s also prone to the same mistakes of motivated reasoning and cognitive bias that all of us are.  Without external forces to hold him to account, he’s probably going to fuck things up badly. 

And that’s why - to govern giant complex systems - we have designed other accountability mechanisms – like, say, elections. 

The Worst System Except for All The Others?

By the time this is published on Wednesday, it will either have been a good, bad or inconclusive night for President Joe Biden.  

But what will this election tell us about what the American people want? If Democrats win, is it a referendum on saving democracy? Is it about gas prices falling? Is it about the overturning of Roe v. Wade?

If Republicans win, is it about inflation? Is it a normal structural backlash to the in-power party? Is it a ringing endorsement of Trump and MAGA candidates?

Who knows!  

In 2012, the Republican Party was confident of victory.  When they lost, they commissioned a report to tell them what to do next.  And that report told them that the only path back to power was softening rhetoric on immigration, appealing to women and generally embracing a more diverse, progressive country.

So they ran and won with **checks notes**... Donald Trump.

Countries are definitely prone to emergent, unpredictable behavior. Democracy’s best characteristic is that it has peaceful means of accountability.  When something isn’t working, we can change course.  If it is working, we can double down.  

That works sometimes.  But our system has its own problems with feedback loops: its feedback is non-specific.  In the coming election, people will choose who to support based on a number of issues – inflation, health care policy, abortion, democratic values, crime etc.  A reasonable person might have some heterodox but completely logical views across this matrix of options.  They might think that spending causes inflation, that we need more health care support for needy families, be pro-life, anti-election-denier and pro-police.

How do they signal that the government should correct in this direction?  They can vote Blue which hits about half of their values. They can vote Red which hits the other.  

Neither decision will communicate their actual policy preferences to the government.  That means we don’t have any precision to our feedback loop.  The question that animates me is: could we do better? 

What would a better system look like?

It won’t surprise any of my readers or anyone who has heard me talk in the last year that I think DAOs provide a playground to improve the governance of complex systems.  And my reasoning for that is simple –  DAOs provide an opportunity to build better feedback loops than corporations, with more precise accountability than our electoral system.

In a 2013 blog post, Vitalik Buterin asked if we could build better organizations by eliminating human management altogether: 

But what if, with the power of modern information technology, we can encode the mission statement into code; that is, create an inviolable contract that generates revenue, pays people to perform some function, and finds hardware for itself to run on, all without any need for top-down human direction?

Put another way:  can’t the AI just solve this for us?  

Let me cut to the chase:  No, it can’t.  No GPT3 text prompt is going to tell you how to define an ideal government for a complex collection of people and needs.  But Vitalik’s instincts aren’t totally wrong – we can get a long way by doing a few things:

1. Clarifying the mission of an organization;

2. Empowering specific teams with clear rules on how to achieve those missions;

3. Letting those teams compete for rewards that will be assigned based on a vote by all stakeholders;

4. Measuring their progress against their stated goals and then deciding whether to extend or kill their workstreams.

Now, no DAO is doing this exactly today.  But the individual governance legos are starting to emerge.  You still have to write out your human mission the old-fashioned way, but if you do that, you can learn and iterate quickly and publicly using this set of tools.  And the organizations that adopt them will be able to out-compete the outmoded top-down speed of a corporation or the lossy representativeness of traditional electoral democracy.

1. Small, accountable organizations with Metropolis.  Metropolis is a set of tools that enable an organization to operate as independent pods.  These pods have their own rules, their own mission and their own funding.  They can operate collectively or independently to achieve a group’s mission.  They’re federalism in action.  

2. New mechanisms for incentivizing the right actions with PropHouse and Coordinape.  

   1. Prophouse by Nouns.  The Nouns community invented and has now publicly launched a tool called Prophouse.  It allows any group of people to set a prize pool (say 5 Ethereum), a theme (“What party should we throw?”) and then vote on how to allocate the prize amongst proposals made by members.  It’s a democratic way to choose specific policy ideas from a community.  The discrete ideas, and their proposers, can then be evaluated based on outcomes.  It keeps feedback loops tight and accountable.

   2. Coordinape.  Coordinape enables decentralized “peer bonuses” in a community.  Members decide on how much budget to allocate for the period (say a month).  Then each member gets to award “points” to the members they think did the most valuable work.  At the end of the month, payment is distributed to users proportionally based on the points that they received.  This mechanism encourages everyone to collaborate with everyone.  It’s a p2p way to encourage teamwork.

3. Public and open data to track progress with Dune. One of my favorite things about writing about communities in Web3 is that I can see **all** of their data – even when I’m not a member.  A simple trip to Dune.xyz reveals a world of dashboards measuring every community’s KPIs.  This open data allows anyone to assess what’s working. It is a source of real time truth that allows for quick iterations and improvements.
   

Independently, each of these tools is an interesting organizational innovation.  Taken together, they’re signals of the potential of a new type of organization that helps a community to harness the emergent intelligence of its network.  Allowing insights and leadership to emerge from anywhere enables a network to learn at the speed of its fastest nodes.

It’s governance that can actually keep pace with the governed. Feynman was right that the world is made up of individual pieces, but it's only through bringing those pieces together that we see the really interesting stuff.  More, it turns out, isn’t just different.  When it comes to feedback in complex systems: More is better.