Second Law of Technodynamics

Left without counter-action, coordination control concentrates

Physics has laws. The social world has, at best, tendencies. Drop a stone, and Newton tells you where it lands. Drop an interest rate, and economists will offer you a range and later a retrospective explanation why the range was wrong.

Auguste Comte hoped sociology would close the gap and called it initially “social physics.” That hope keeps returning under different names such as econophysics, computational social science, cliodynamics, and the less-known recent world machines theory (to which I made some modest contributions).

Messy as it is, society shows some regularities. Zipf’s law holds across word frequencies, Little’s Law holds for any queue, and scaling laws predict the rate of crime and innovations by city size.

And here’s yet another social law. It is a law of technodynamics. Technodynamics is about the dynamics of coordination in socio-technical systems (more generally, the dynamics of cohesion). It’s about how coordination control arises, moves and concentrates.

This essay is part of the Autonomy and Cohesion series.

Technodynamics studies human socio-technical systems. What distinguishes human and animal coordination technologies is reflexivity. Language is coordination of coordinations working through its history of reflexive and recursive corrections. Writing extends the loop across generations. Print stabilizes and scales writing. Software makes the loop operational: programmers coordinate, in language and code, to update the coordination code itself.

Animal coordination technologies have no such loop. (You can read more about this distinction in the previous essay.) So the domain of technodynamics is socio-technical systems whose coordination technology is reflexive.

Coordination control

To revise a coordination rule using the coordination’s own means, the rule must be something you can point at, quote and change. From being a disposition inside the participants, it has to become an artifact outside them: a grammar, a statute, a price list, a codebase, a ranking algorithm. Participation terms are written in that artifact.

A term has to be revisable, held and binding. Where the coordination cannot be revised, as with the ant pheromone trail, nobody can insert a gate into it.

The second condition is that the term has to be held. In other words, somebody has to have asymmetrical privilege to change the term. No ant strumbling upon a pheromone trail left by another ant will find its terms of service updated, making it ambiguous unless a food-finder fee is paid.

The third condition is that the term has to be binding. A term is binding when the participant cannot readily replace the coordination.

In 2010, Oracle acquired Sun Microsystems and its rights to OpenOffice.org, then the leading free office suite. This was a case of the term not being effectively held due to the lack of exclusivity. The code was forked, and this is how the independent LibreOffice was born.

In 1996, the German-speaking states agreed a reform of German orthography. There was a campaign against the reform, and in 1998 the Federal Constitutional Court confirmed that outside schools and public administration, people may spell as they like. And some major newspapers did. They kept using their own in-house rules even after the Council for German Orthography removed the most controversial changes from the reform in 2006. That’s an example of when the term is not binding.

We went from reflexivity to coordination terms and then to the capacity to revise, hold and make the terms binding. Call such capacity coordination control:

Coordination control is the capacity to set and revise the terms on which others participate in a coordination that they depend on and cannot readily replace.

How is coordination control created?

One way to create coordination control is by enclosure, by turning free autonomy into leased autonomy. Facebook assembled its social graph from relations that already existed.

Coordination control can also be created by coupling with a newly introduced capability, as Apple did when it opened its App Store in 2008. Unlike Facebook, nothing was enclosed since the market of native iPhone apps did not exist.

A third way is by layer insertion. Open protocols don’t allow any coordination control, but real-world coordination regimes evolve beyond elementary coordination technologies and use technology stacks. Take queuing, HTTP and Git. Airline priority boarding is adding a layer on top of the queuing protocol. Facebook and Amazon add a service layer on top of HTTP, and we get the Technofeudalism of the platforms. GitHub creates coordination control by inserting a layer of collaboration and automation on top of the Git protocol. It is a valuable layer that easily creates dependency because it includes issues, pull requests and their review threads, CI runs, permissions, notifications, and the discovery that brings contributors to the project in the first place.

A fourth way is by using what Zuboff calls the behavioral surplus, which is extra data about people’s behavior that is collected beyond what is necessary to provide a service.

These methods of creating coordination control often work together.

Concentration

The second law of technodynamics states:

In any socio-technical system, isolated from counter-action, coordination control concentrates.

The parallel with thermodynamics is, of course, intentional. That parallel brings a bit more than mnemonics but not much more. Let’s examine it.

In thermodynamics, “isolated” is used literally. It means the system is closed completely and no energy or matter can enter the system. In technodynamics, “isolated” is used metaphorically and functionally. It is metaphorical because socio-technical systems are open and far from equilibrium due to the constant flow of energy. As long as the flow continues and there is no counter-action, the law holds. And it is used functionally because counter-action doesn’t need to come from outside, which is what “isolated” implies. It can come from inside the system.

The other parallel is the spontaneous direction. It is towards evenness in thermodynamics and towards concentration in technodynamics. The direction is the opposite, because the system classes differ, not because physics is inverted.

                  ┌───────────────────────────────────────┬───────────────────────────────────────┐
                  │             thermodynamics            │             technodynamics            │
┌─────────────────┼───────────────────────────────────────┼───────────────────────────────────────┤
│ system type     │ closed                                │ open                                  │
├─────────────────┼───────────────────────────────────────┼───────────────────────────────────────┤
│ isolated from   │ energy and matter (literally)         │ counter-action (metaphorically)       │
├─────────────────┼───────────────────────────────────────┼───────────────────────────────────────┤
│ moves towards   │ evenness                              │ concentration                         │
├─────────────────┼───────────────────────────────────────┼───────────────────────────────────────┤
│ what is cut off │ the work that would reverse the drift │ the work that would reverse the drift │
└─────────────────┴───────────────────────────────────────┴───────────────────────────────────────┘

Otherwise, a more direct parallel to physics would be the Bose-Einstein condensation.

So, coordination control concentrates. What drives that concentration and what can slow it down or disperse it? The next section reviews the causes of concentration and is followed by one about counter-actions.

Concentration Drivers

Concentration of coordination control changes a coordination regime. Coordination is an aspect of cohesion, but since it’s more evocative, I use it interchangeably and even more so after introducing the cohesion spectrum.

Cohesion is driven by forces and technologies. The initial split I introduced was forces and tools, but since tools are not the general case — plenty of coordination technologies do not involve tools — I corrected that in the essay about reflexivity. This distinction, after some sharpening, can be reused for the first split of concentration drivers.

The question creating the split is: does it run on its own?

A force is a standing condition which, once present, works on its own. While belonging and FOMO are cohesion forces, preferential attachment is a concentration force, dependent on, and working on top of them.

In the case of Facebook, the concentration force (preferential attachment) works on top of the cohesion force (you want to be where your people are), which drives the third concentration force, lock-in (once your social network, photos and history live only there, leaving is dear). All these forces are channeled and amplified by the use of concentration technologies, but they are not technologies by themselves: Facebook did not invent belonging or preferential attachment.

Concentration works even when the conditions are not favorable for concentration. Gnutella was a peer-to-peer network launched in 2000 with the intention of being as decentralized as possible. Yet it grew its own ultrapeers anyway, driven by scalability and efficiency pressures.

The dollar-clearing system, the SWIFT messaging network and the chip supply chain were each built by private firms chasing their interests. Yet they all concentrated into chokepoints.

A technology must be worked. It is often used intentionally toward concentration: buying a rival, forming a cartel, or utilizing personal data. A platform’s owner benefits from a network effect, but the effect concentrates whether or not somebody exploits it, so it is a force, not a technology. A referral scheme accelerates that concentration, but it is something put in place and maintained by somebody, so it’s a technology.

Just as the effect of a force that is exploited can attract a technology to accelerate it, a technology can activate a force. When a vendor creates a proprietary technology, then the switching cost effect operates as a byproduct, a force.

So the first split is based on the question: does it run on its own? If yes, force; if not, technology.

Now the second question: does it gain control or keep it?

This question captures the two clauses of the law. To gain control is to produce the power position. It is the force or work of making the particular mode of coordination what it is and hard to substitute. To keep control is to prevent its reversal, to keep the system isolated from counter-action. Keeping presupposes gaining. Otherwise there is nothing to keep.

Crossing the two questions gives a matrix of four concentration drivers:

                                        ┌──────────────┬──────────────┐
                                        │ GAIN control │ KEEP control │
                         ┌──────────────┼──────────────┼──────────────┤
                         │ Forces       │ The Pull     │ The Hold     │
                         ├──────────────┼──────────────┼──────────────┤
                         │ Technologies │ The Grab     │ The Guard    │
                         └──────────────┴──────────────┴──────────────┘

The Pull

The Pull is the driver that gains control based on concentration forces. Participants tend to converge on whatever is better connected, larger, and cheaper.

The more are already connected, the higher the value of connecting. One fax machine is scrap, while the second makes both useful, and every machine added raises the worth of all the rest, a regularity known as Metcalfe’s law. And the more are connected, the more attractive it is for a newcomer to connect.

When a network is concentrated, it is cheaper for any node to reach another via the hub. There are also economies of scale, which attract new users, lower the cost per unit, and attract even more users.

Much more can be said about the Pull, but in the interest of space (there is a lot still to cover), I’ll stop here, and can recommend this essay by Gordon Brander.

The Pull concentrates on its own, but it needs a Keeper to turn concentration into coordination control. Before moving to Keepers, let’s first check the second Taker.

The Grab

The Grab produces a position through deliberate action. Where the Pull accretes, the Grab takes.

Once big, the shortest path to a stronger position is through acquisition. That can be an act of direct elimination of rivals. When Western Union bought the last remaining national rivals in 1866, it got 90% of the US telegraph market.

Not only acquisitions but also grants can work as a Grab. Queen Mary I granted a royal charter to the Stationers’ Company in 1557, and thanks to that, the company held a monopoly until the Copyright Act of 1710.

Sometimes neither purchase nor grant is needed. Rivals can stop being rivals by agreement, and we have plenty of cartel examples, not only in the US but also in Europe, the most prominent proven cases being the vitamin cartel and the truck cartel.

A fourth way is acquiring the lever over a position others have built, as the US acquired access to SWIFT’s records after 9/11.

And now the two Keepers: the Hold and the Guard.

The Hold

The Hold is a standing condition that makes leaving costly or futile. It holds whoever is inside with no one working it.

A participant can be free to leave and still stay because of the investment and the cohesion forces, as already mentioned for Facebook. That’s not only the case for platforms but also for protocols designed to prevent such effects. ActivityPub permits migration between instances, and yet the top three Mastodon instances hold 40%. Overall, the distribution of users in the ten thousand instances is heavy-tailed. Such a statistical signature is typical for many systems, and for some of them it approaches a power-law distribution. That signature is due to the Pull, but the Hold stops the tail from flattening out. In the current case with Mastodon, the coordination control is concentrated, but it’s not active. It’s closer to the dormant leased autonomy, as discussed some essays ago.

A stronger case is the vendor lock-in, and there is no better example than Microsoft. They have built proprietary technologies, file formats, coverage of nearly every business need and tight integration and complementarity between products. But the Hold is an emergent systemic effect. Once there are millions of documents, custom workflows, trained employees, internal compatibility and security policies, the switching costs are so high that Microsoft doesn’t need to do anything to trap its customers.

The Guard

The Guard works actively against replaceability.

For the merchants selling on Alibaba's platform, alternatives existed, but they were punished if they tried to use them. The forced exclusivity lasted until 2021, when Alibaba was fined heavily for breaking the Chinese Anti-Monopoly law.

Another popular case is Booking.com. Their contracts with accommodation providers carried price parity clauses, forbidding better prices elsewhere. It took two decades to change this practice in Europe. Outside Europe, the practice continues.

We got the four drivers asking two questions: does it run on its own (yes — force, no — technology), and does it gain or keep concentration? There is another useful question: is it one-off or does it use a self-reinforcing feedback loop? This splits into one-shot and compounding drivers, and this is how we get to eight drivers.

Here’s a flat and static presentation:

                  ┌─────────────────────────┬────────────┬──────┬─────────────┐
                  │ Driver                  │ Kind       │ Role │ Feedback    │
                  ├─────────────────────────┼────────────┼──────┼─────────────┤
                  │ hub efficiency          │ force      │ gain │ one-shot    │
                  │ preferential attachment │ force      │ gain │ compounding │
                  │ acquisition             │ technology │ gain │ one-shot    │
                  │ roll-up                 │ technology │ gain │ compounding │
                  │ stranding               │ force      │ keep │ one-shot    │
                  │ lock-in                 │ force      │ keep │ compounding │
                  │ blocking                │ technology │ keep │ one-shot    │
                  │ attrition               │ technology │ keep │ compounding │
                  └─────────────────────────┴────────────┴──────┴─────────────┘

And here is a 3D interactive visualization, similar to the one I used for the coordination technologies and stacks.

Counter-actions

Counter-actions include keeping a coordination replaceable or forcibly dispersing concentrated coordination control.

If a dependent can substitute coordination, no terms bind, so no control concentrates. But what needs to be substituted is the full value of the coordinated system. In the case of a social network, not only the content needs to be portable and comparable features available from elsewhere, but also the network context and affects.

A counter-action can come from inside or outside. It can come from a change in the community’s own governance, a fork, a rival or a regulator.

In 1911, Standard Oil was broken up into 34 successor companies. In 1920, the “Big Five” meatpackers were subjected to a far-reaching consent decree that required them to divest many non-meat businesses and restricted further expansion into adjacent markets. In 1948, the Paramount antitrust decision forced the major Hollywood studios to divest their cinema chains, ending their vertically integrated control of film production, distribution, and exhibition. In 1984, the Bell System was split into seven regional operating companies, ending AT&T’s monopoly over the US telephone network. Sustained community pressure, combined with growing competition from other staking providers, reduced Lido – the largest decentralized staking pool on Ethereum – from a peak of 32.3% of all staked ether in late 2023 to about 24.4% by mid-2025. This moved it safely below the one-third threshold at which a single staking pool could prevent the network from finalizing new blocks.

Different counter-actions attack different drivers. Dissolving a trust attacks the Grab, building alternatives threatens the Hold, and maintaining interoperability weakens the Guard and the Hold.

But the drivers persist, and when a counter-action is reduced or withdrawn, the concentration resumes. Beef packing re-concentrated from a four-firm share of roughly 25% in 1977 to 70% within fifteen years of antitrust enforcement weakening, and stands near 85% today. The seven regional Bells created by the 1984 breakup gradually re-merged through consolidation into effectively two dominant firms. The studio decrees were terminated in 2020, with vertical integration now partially reassembling around streaming platforms.

Counter-actions can also bring unintended consequences, including helping the drivers of concentration. In the week after the GDPR took effect in 2018, websites’ use of third-party vendors fell by about 15% while concentration among the survivors rose by about 17%, with much of the gain, at least for a while, going to Google and Facebook. India’s public UPI rail grew a near-duopoly of private payment apps on top. The non-profit Let’s Encrypt became a roughly 63% dependency for publicly trusted TLS certificates, and a single bug in 2020 led to the planned revocation of about three million certificates on a day’s notice. Mozilla, steward of the last independent browser engine, derives about 85% of its revenue from Google, so the surviving concentrator alternative sits on the concentrator’s payroll.

The counter-actor can itself become the dominant player. OPEC was founded as producers’ counter-action against the market power of the major international oil companies and became the term-setting cartel itself. Britain’s 1989 Beer Orders broke the brewers’ tied houses, but non-brewing pub companies bought the freed estates and tied their tenants much as before.

Coda

Concentration of coordination control does not go on forever. It ends at some point.

Endgames

Its endgame can be brought by a successful counter-action or by a transfer. Neither ends the coordination itself. A counter-action may disperse it, and a transfer only changes the party that holds it, as when Venice lost the spice trade to Portugal in 1498 when the Cape route was opened.

The third way a concentrated coordination control can end is a system collapse. Then not just the control but the coordination itself ends. I thought that such collapses were driven by concentration alone. And indeed, it looked quite plausible.

When coordination concentrates, it is not only isolated from counter-action but also from learning. The Keepers keep participants from leaving, so there is no departure that could send a corrective signal. Concentration also tends to reduce the reserves for both known risks and complete uncertainty, as discussed in Requisite Inefficiency. Another cause of collapse is over-extraction, when the coordination holder consumes what it lives on. And the most obvious candidate cause of collapse is that concentration creates a single point of failure. A concentrated network survives random node loss better, but in cases of deliberate removal of the largest nodes, it breaks down.

And yet, I found only collapses with multiple causes and not a single one due to over-concentration alone. Systems with concentrated coordination control take serious blows that confirm the vulnerabilities I described above, but survive and get stronger and more concentrated. The failures of concentration feed it, and the law keeps running.

First law

Now, you may ask, if there is a second law, where is the first? There was a first, the initial version of which, about the conservation of agency, I shared last year. It was well fitting my work on protocols, since if the coordination is done with a non-agent, a protocol, like it is in a roundabout, when compared with manual control or a traffic light, then participants have more agency and the overall coordination is smoother and safer. But I realized that “total agency” sums heterogeneous things, so I dropped it. Then after the second version of Cohesion Spectrum, I tried a similar hypothesis by changing the conserved quantity to “participatory autonomy,” but that failed too. The most recent attempt, when working on the second law, brought “coordination control” itself as a candidate for the first law. It was so promising that the initial title of this essay, as a follow-up to the previous, was “Technodynamics II: laws.” But that version of a first law did not survive the tests either. So the first law remains an open question.

In fact, the whole question of lawfulness is open too. And not only for social systems but also for physics.

What kind of law?

Joe Rosen, a theoretical physicist, argues in Lawless Universe that science works only on phenomena that are reproducible. When the same experiment run again yields the same result, the pattern generalizes into a law. This works for whatever comes in vast interchangeable quantities but not for the one physical system that is not a member of any population: the universe as a whole. While Lawless Universe concerns the scope of the laws, the recent book The Blind Spot goes after their source, the scientists, who invariably indulge in the fantasy of God’s-eye view. Science describes reality as it is in itself, independent of anyone describing it. Science relies on human experience in every step but then quietly deletes it from its self-image. QBism addresses the problem with excluded experience from inside quantum physics. For Fuchs and his collaborators, a quantum state is not a property of an electron, but a personal catalog of an agent’s expectations about what will happen if she acts on the world. QBism follows John Wheeler’s slogan “law without law.”

But even if the laws of physics are questionable, there is still something that puts the laws of social systems in a class of their own. When Newton described gravity, gravity did not change. But when Adam Smith described the market, Marx described the class struggle, and Hayek and Friedman described the price system and the role of money, these descriptions changed the social systems they described. Which brings us back to reflexivity, where we started when making the main distinction in technodynamics.

And yet, when a law does not assign roles, concern interests and point to levers for change, even if it describes an aspect of social life, as Zipf’s law does, like the laws of physics, it doesn’t change what it describes. It seems that some social laws are Marx-kind, others are Zipf-kind.

What kind of law is the Second Law of Technodynamics then?

It names roles, concerns interests and points at levers, so it qualifies as a Marx-kind of law. The levers it points to reach the technologies at best, but not the forces. And even the levers backfire because they are meant to: counter-action needs coordination, and coordination control concentrates. Which qualifies the Second Law as a Zipf-kind of law.

Bringing the two together, it turns out the Second Law of Technodynamics is a Marx-kind of law about a Zipf-kind predicament. It is a law you can act on, without ever finishing.