Blockchains Never Forget

Just three years ago, in 2014, I wrote a little short story set in a future where most work is organized around blockchains. That story was set sometime past the 2120s, but it appears we'll get there a century earlier than I thought. The idea of organizing work through smart contracts on blockchains has been moving ahead at a breathtaking pace. Over the last few weeks, I had my first hands-on immersive experience of this particular piece of the unevenly distributed future. I'll share more about the specifics of this experience, and lessons learned, but mainly I want to enter my first serious attempt at blockchain punditry into the public record: the blockchain is irreversible social computing.  The message of the medium is this: blockchains never forget. By providing an extra-institutional base layer of irreversibly settling collective memories that cannot be erased, blockchains create a foundation for fundamentally different institutional and technological landscapes. Ones based, as I will argue, on a notion of artificial forgiveness. Work on the Blockchain To ground what is going to be a very philosophical, perhaps even theological, discussion, let me share some details about the Great Ribbonfarm Blockchain Experiment of 2017. On the surface, there's not much difference between organizing work on the blockchain as opposed to traditional ways. Take for example, the specific example of Kanban, a workflow model commonly used in both software and manufacturing settings. Here is the colony.io* interface for one of the boards in the ribbonfarm colony. This is the blockchain way of doing kanban. And here is one of the boards in my private Trello account, one of the popular incumbent tools for organizing Kanban workflows. As you can see, there is not much of a visible difference. If you've used kanban, either on a whiteboard with post-its, or with a digital tool, there isn't much that is radically different about kanban on the blockchain. It's all the same process management work: defining meaningful boards and sensible workflow columns like to do, doing, done, assigning tasks to people, and tracking work by moving tasks from column to column as they evolve. The only thing you do differently is that you assign a budget to each task in tokens (in the case of colony, these tokens live on a private blockchain, but in principle they could live anywhere, including public blockchains like ethereum*). These tokens get paid to the task assignee once the task is completed. You can use this token economy to create and run a task market. You can go beyond and do things that measure the value of these tokens in the open economy, relative to other assets. In my case, as a first experiment, I used the colony platform to organize reviews for the upcoming ribbonfarm longform blogging course. We received 77 applications. I grouped these into bundles of 3 each, set the offer price for reviewing a bundle to 1 token, and invited a bunch of past ribbonfarm contributors and long-time readers to help with the reviewing. The idea was to have each application reviewed by at least 2 people, with Sarah Perry and I, who will be teaching the course, making final selections from among applications that received two Yes votes (with some discretion to veto or make exceptions). As with any market, the price had to move to get the job done. I had to raise the offer price twice, from 1 token to 1.5, and then to 2. And I had to offer incentives for finishing more than one bundle. Initially, I offered a 10-token bonus for doing 5 bundles, and then increased it to 12. Our little ribbonfarm market wasn't quite as crazy as the public crypto markets over the same period, but by the time the market cleared, we had spent just over 140 tokens to get all 77 applications properly reviewed. Since each application was reviewed twice, the cost came to approximately 1 token per review on average. Here's a snapshot of the activity (some bits redacted). As you can see, Tim Roy reviewed two bundles, but got paid 3T overall. Okay, fun stuff, and I certainly enjoyed playing banker, exercising authoritah, and moving prices around to clear the market. But even this doesn't seem particularly special, does it? It's a complex, but not mysterious, kind of marketplace feature. One you could obviously engineer into any task management product. The feature is conceptually simple. It's just book-keeping after all. Except for one aspect that makes all the difference. The tokens also represent smart contracts that live on the blockchain. When a task is marked complete and the tokens transferred, it is done done. Not just enterprise-grade. This is Pharaoh-grade so let it be written, so let it be done done. There is no going back. There is no rewriting history. At best you can agree to disagree and fork off a parallel universe where the historical record of consensus is different (this is basically what happened in the aftermath of the DAO hack last year, which created the two parallel universes of Ethereum (ETH) and Ethereum Classic (ETC)). The Wheel and the Odometer The difference between a blockchain-based technology and a similar non-blockchain technology is something like the difference between the odometer in your car and all the other meters. They're all dials or digital displays. All work on similar principles of sensing, measurement, calibration, and display. But unlike all the other instruments on the dashboard, the odometer cannot ordinarily be reset or dialed back. There are engineering and legal protections to ensure that. Because the value of a car is a strong function of the number of miles it has been driven, this is a case where a relatively tamper-proof record of history has economic value. Now imagine that idea applied to everything. What if it were easy and cheap to ensure that every piece of technology had a generalized memory of its own operational history? What if speedometer maintained an unalterable history of how fast the car had been driven? What if the fuel or battery meter preserved complete histories of filling/emptying cycles or charging/discharging? And all done in silicon with mathematics. No expensive and slow carving-on-stone required. No wax seals, no blood-fingerprints. What if civilization itself had an eidetic memory of history? What if all the photos and videos we are taking today were instantly, irrevocably saved to a blockchain, creating a credible virtual past we could visit via a virtual-reality time-machine? Not all histories are equally important or interesting, but the point is, history and memories that are laid down irreversibly, and are available to be fed back into present actions, alter the fundamental character of a technology. How you think about engineering starts to change. The odometer is the archetypal artifact for an entire kind of engineering. It is to blockchain-style engineering as the wheel is to regular engineering. A wheel can roll backwards or forwards. An odometer can only tick upwards, inexorably driving the depreciation of your car and the accumulation of a history. Which means, among other things, that when mistakes are made, you can't simply hit an undo button, or revert changes. In our experiment, mistakes were made. People reviewed the wrong bundle of applications by accident, for instance, so some applications got 3 reviews instead of 2, while others got only 1. To fix such errors, I had to assign rework tasks, a common enough situation. But since by design, there is no direct way to reverse transactions, on the couple of occasions when I paid out tokens before verifying the work, people got paid a token or two for work they didn't actually do. This is of course the well-known chargeback problem. With a normal credit card, chargeback transactions are conceptually simple logical reversals of computations. Additions are as easy as subtractions. On the blockchain, there's no direct way to reverse anything. You need what I call artificial forgiveness. We'll get to that. Fortunately, I ran a pretty tight shop, and though my process won't win any six-sigma certifications, it was pretty darn clean, with a very low procedural error rate. Okay, enough with these impractical, precious snowflake details. Let's get to the practical philosophy and theology around the key detail in the account above: the irreversibility of certain actions. Reversibility as Feature and Bug It is important to emphasize that we are talking about social irreversibility here, a phenomenon that takes a good deal of engineering to achieve. At a physics level, computing of course follows the second law of thermodynamics like everything else. It turns out that permanently deleting a bit -- setting a bit to 0 whether it is at 1 or 0 -- is an entropy-increasing action, and is the irreversible (and irreversibly physical, ie embodied in atoms) aspect of digital computing. The aspect where bits meet atoms and digital dualism breaks down. There is a sort of pseudo-reversible paradigm of reversible computing out there but we can ignore that for our purposes here. That sort of reversibility/irreversibility isn't particularly useful for social engineering, except to the extent that it enables mechanisms like bitcoin mining. The physical irreversibility of digital computing has historically been irrelevant for practical purposes. The entropy increase from even the most complex computation running across the world's data centers -- including all the bitcoin mining -- is minor compared to (for example) the entropy costs of running the planet's transportation on fossil fuel engines. And more importantly, it makes no practical difference to the meaning of the computation. Writing and reading bits off a magnetic disk might involve more net entropy increase than off an SSD, and you could perhaps refactor code to involve fewer bit deletions and more efficient kinds of memory, but it makes no real difference to the meaning of the computation. In regular computing of the sort that is at the basis of basically all the software you use, undoing actions is always conceptually possible even if the ability is artificially restricted. You need certain institutionally delimited powers to approve chargebacks on a credit card for example. In fact that has been the defining distinction between bits and atoms. In the world of atoms, a glass that falls and shatters cannot magically put itself together again. All the king's horses, and all the king's men cannot put Humpty-Dumpty together again. At best they can make an omelet. That's the second law in action. In the digital world, the broken glass can get rewound to its pristine state. Flow back into the unbroken glass, spilt milk can, if on-screen Yoda is. In the world of bits, so long as you keep perfect records and your storage media don't get corrupted, in principle everything can be engineered to be logically reversible, even if it is not thermodynamically reversible. Sure, there's bit rot at larger scales, but there's ways to design redundancy around that kind of thing. In virtual worlds, you can always add undo buttons. And that's a problem. Undo versus Wabi-Sabi Buttons The reversibility of computing is not just the main difference between atoms and bits, it is also arguably the primary selling point of the digital world, ahead of such selling points such as computational speed advantage over pen-and-paper. It is a selling point that has historically constituted the biggest competitive advantage of the digital world over the physical world. The biggest difference between a word processor and a typewriter is that the former has effectively unlimited undo/redo capability. The biggest difference between Grand Theft Auto and physical roads is that you can uncrash a car in the former. The biggest difference between  an online poll and paper ballots (or electronic voting machines that print out a paper record at the time the vote is cast) is that you can uncast a vote in the former. We've gotten so used to reversibility as a self-evidently useful feature -- who doesn't like unlimited do-overs after all -- it is easy to miss the most important thing about the blockchain: it treats the logical reversibility of ordinary computing as a bug to be fixed. The blockchain is an invention rooted in the insight that engineering irreversibility into computing is of practical value. That the odometer is as important as the wheel. The way this is done is to couple irreversibility of the physical sort (computing takes energy because deleting bits increases entropy) to irreversibility of a mathematical sort: hard-to-invert cryptographic functions. Neither physical, nor mathematical irreversibility by itself is enough to create deep irreversibility in computing of the kind we're talking about. Mathematical almost-irreversibility would be useless if computing were free in energy terms. Physical irreversibility would be useless without a way to sneak it into algorithms via mathematics. But when both are possible, you can implement irreversibility mechanisms such as proof-of-work. You can build irreversible social computing systems: systems that have the very possibility of an undo button engineered out of them, using mathematics and physics. What the blockchain enables isn't the shallow social irreversibility of the sort any competent programmer could code into a piece of software without resorting to Merkle trees, cryptographic hashes and proof-of-work algorithms. Shallow social irreversibility is an illusion that an administrator with the right permissions can break simply undoing transactions and rewinding events. Or to put it more precisely, this kind of shallow irreversibility is always relative to the permissions architecture that defines an organization. In shallow irreversibility, the idea of administrator precedes the idea of reversible in the ontology (if you'll excuse a geek joke: you decide what can be done or undone, I'll decide who can sudo or not sudo). Shallow social irreversibility is relative social irreversibility. The irreversibility goes only as deep as the powers-that-be allow it to, and it never penetrates past the foundations of the organization itself, to find roots in the open environment. Irreversibility that threatens power structures can be contained within institutional design a priori, by giving the right people the ability to rewrite history. Recording an event on the blockchain on the other hand (through crypotgraphic-physical irreversibility mechanisms) is actually irreversible for practical purposes. It does not matter whether you have Senior God level privileges within some institution that lives on the blockchain. It does not even matter if you're on a private blockchain entirely owned by a single entity. As you've probably learned by now, unlike traditional passwords, which are recoverable or at least resettable by administrators of computer systems, the "wallet passwords" you use to manage blockchain activity, if lost, are gone for ever. No administrator can help you unlock the lost assets (you could still try a brute force unlocking, but the point is, no human can override the mathematics by fiat). You're working with the digital equivalent of spillable milk that cannot be unspilled. Only cried over. Or more precisely, this deeper notion of reversibility precedes notions of authority and permissions that go into the design of institutions. It is absolute social irreversibility. It is not a set of constraints imposed on certain mutable roles concerning what can and cannot be done. It is a property, the property, of a memory. Irreversibility is the sine qua non of the blockchain as a system. Engineering on the blockchain is wabi-sabi engineering. A mode of design where you take irreversibility into account, where history acquires technological salience. Where breaks can only be healed, not reversed. In the Japanese aesthetic mode of wabi-sabi, a broken object is lovingly repaired (often with precious materials like gold), restoring it, not to a state of pristine it-never-happenedness (such as what an insurance company in the US will pay for after a car accident), but a healed state that has restored functionality but also a trace of preserved memory. The crack in the cup becomes a thin vein of gold, a permanent record of the break. One can imagine a generalized wabi-sabi button as a replacement for the undo button in any engineered artifact based on blockchain technology. A button that, when pushed, causes healing, forgiving events that would make the Buddha smile. What can you do with this powerful property? Social Irreversibility All of the common metaphors for understanding the blockchain -- as a digital currency with double-spending protections, as a general digital ledger, as a decentralized public database that resists monopolies of truth, as a protocol economy -- understate or obscure the essential feature of the blockchain: absolute, pre-institutional, social irreversibility. Let me offer a working definition:

An event is socially irreversible if agents who observe it cannot unobserve it and return to the information state they were in before the event.

A socially irreversible event, once it is in the historical record, is potentially there for all eternity. Special steps must be taken to prevent the memory from affecting future realities; steps analogous to the golden repair veins on wabi-sabi cups. There is a word we use to characterize such steps. That word is forgiveness. In computational terms, forgiveness is a decision to construct a justification for actions that is at odds with the justification supplied by history, leading to increased freedom of action. To forgive is to not be a prisoner of of history, without having to destroy it through erasure or rewriting. A temporal non-zero-sum game. To forget, on the other hand, is to destroy the past to free the present, a much weaker mode based on a temporal zero-sum game. The act of forgiveness recalls to mind the famous Viktor Frankl quote: "Between stimulus and response there is a space. In that space is our power to choose our response. In our response lies our growth and our freedom." There is perhaps no more powerful force in human affairs than the power to ignore the instinctive imperatives of history and choose forgiveness. To be able to engineer digital artifacts that can work with this force is to have power comparable to that of religious figures. Forgiveness, as we will see, is the central phenomenon we must deal with in engineering irreversible social computing systems, just as the deletion of bits is the central phenomenon in computational irreversibility. The connection between the idea of forgiving a debt by "erasing" it, and erasing bits, is not coincidental. When faced with social irreversibility on the blockchain, at best you can create and record additional events that reverse some of the effects of a previous event. But you cannot undo, or unremember past events. On colony.io, if I paid you 1 token by mistake, you could create an empty task with a budget of 1 token called here's your damn token back, and assign it to me, and I could accept it. This is a no-harm-no-foul transaction, where the element of forgiveness is minor. But what if the amount were non-trivial, or one or both parties made commitments that they'd have to renegotiate on the basis of having or not having that token? What if that token lived on the public blockchain, valued in ether, and the price of ether jumped between an error and a fix? To heal the situation, one of us would have to forgive the other for a mistake that affected both of us. Unlike a unilateral chargeback, or a rewind achieved via administrative fiat, this pair of actions requires a forgiveness computation to occur between two humans. One has to ask for forgiveness, the other has to grant it. But we would never be able to forget. The pair of transactions would partially cancel out, in nominal token terms, but both would be recorded and remembered on the blockchain. What is significant about this? To understand that, we have to back up and consider the history of social memory. Or the history of history itself. The History of History Human collective memory is chock-full of what one might call ritualized forgetting. Among other things, we say things like:

That never happened.

Forget I said that.

Can we start again?

I'm just going to pretend you didn't just say that.

Even our rituals around remembering invoke or incorporate irreversibility. You break it, you bought it, we say. The examples reveal a blindingly obvious, but easily missed fact about social computing with human brains: it is irreversible by default, forgotten by accident, convention, or exception. A great deal of social technology was historically invented precisely to counteract or govern the effects of this default irreversibility. The human story is on the permanent record by default, simply by virtue of evolving primarily in human brains in the form of living memories and oral traditions. Brains, like blockchains, but unlike traditional institutions, aren't designed to forget or rewrite history arbitrarily. Biological memories derive their irreversibility from a different physical mechanism than blockchains, but the effect is similar. The idea that you cannot forget is so completely fundamental to everything else that it becomes a strong forcing function driving all design. The human brain, like the blockchain, is a spiritual descendent of the odometer, rather than the wheel. If the computer was a bicycle for a mind, the blockchain is an odometer for the hivemind. Our memories are designed by evolution to be imperfect, limited, noisy, delusional, wishful, and even outright false, but they aren't designed to be un-remembered. Many of the cognitive biases that afflict our thinking (particularly ones like confirmation bias and solution aversion) exist to mitigate the more painful effects of remembering-by-default. The structural boundaries between conscious, subconscious, and unconscious follow the contours they do in part to govern the effects of unalterable memories. Sure, what is recorded is the version of history with the most adaptive utility rather than the version with the closest correspondence to material reality (which, for the record, I believe exists, unlike some). But the point is, though we distort memories in the process of forming them, and go through cognitive contortions to bring present actions in alignment with them, our brains don't rewrite or forget wholesale. And certainly offer no consciously accessible mechanism for intentional forgetting that we can just use at will. If evolution discovered adaptive advantage to the rule don't forget it; repress it under layers of denial over simple forgetting, there is probably reason to suspect it's a good feature. Short of severe brain damage, or uncontrolled entropic decay, if you experienced it, it's there in your brain, seared in by felt emotion, the equivalent of proof-of-work (it is easier to actually forget, or rather, fail-to-remember, things you never cared strongly about in the first place). The blockchain is the same way. The only thing you can do about a memory that is close to unremembering it, once it is in your brain, is to forgive somebody about something. To forgive is to rob a piece of history of its power to shape the future. It is as close as we can get to a neural undo button. Our brains are in a constant state of wabi-sabi evolution, acquiring scars, and repairs. Artificial Forgiveness The emergent architecture of the human condition, as Hannah Arendt has argued, is based on three basic, primitive actions of making promises, breaking them, and forgiving such breaks. These are kinda like the NOT, AND, and OR primitive operations of the human social operating system. One of the three is not like the other two: unlike making and breaking promises, which can be modeled in ordinary, logically reversible computing, forgiving only makes sense with respect to irreversible social computing. You can only forgive what you cannot forget. If you can truly forget things, the concept of forgiveness is moot. Forgiveness is a consequential behavior only when there are things you cannot un-remember. Computationally, you can think of forgiveness as the adoption of beliefs that resolve the conflict between a historical imperative to act one way, and justifications required to act in a different way. It is more expensive because you now have to hold two sets of beliefs relative to the same action: memories (likely painful) that you will not act on, and an alternative set of beliefs that you will act on. Contrast the following pair of history-updates:

1. You betrayed me.
2. You could have not betrayed me.
3. I forgive you.
4. I will deal with you in the future as though you did behave honorably.

Contrast that with:

1. You betrayed me.
2. I will forget the fact and rewrite history.
3. You were true to me.
4. Therefore I will trust you.

The second history update leads to a net simpler state of historical consciousness. One with fewer beliefs. The former not only has extra beliefs and assertions recorded, it also includes a volitional decision to believe. Forgiving is fundamentally a piece of chaos magic, a decision to believe things you cannot justify, and live life on the alternative reality fork where it might have been true, as the bold-face counterfactual clause in the first list demonstrates. In The Human Condition, Hannah Arendt argued that not only is forgiveness a uniquely human capacity, but that Christianity, by institutionalizing the behaviors of trespassing and forgiving within its theology ("forgive us our trespasses, as we forgive those who trespass against us"), fundamentally rewrote the human social operating system. To forgive is to say, I cannot choose to forget, but I can choose to not let my memories affect my actions. Arendt argued that forgiveness is both an essentially human phenomenon, and a social invention that had to be specifically imagined into existence in a break from history. The blockchain provides a foundation for artificial forgiveness, just as computing itself provided a basis for artificial intelligence. What's unique about the blockchain-way of handling errors such as our 1 token correction is this:  you can fix some of the effects of errors but you cannot rewrite the record of transactions itself. That's there for as long as the blockchain itself is in existence, a constant reminder that once, one of us caused a minor 1T inconvenience to the other, that needed to be forgiven and partly fixed. In traditional computing, you can delete old versions of a spreadsheet that represent a solution to the same arrears problem. You can create a state description of our relationship that has no recorded memory of the error and its correction. You can't do that if there's a blockchain in the picture. Here's why this is a profound thing: institutions, the original moral computers of humanity-at-scale, have historically lacked the ability to meaningfully forgive. Institutions Never Forgive If blockchains never forget, traditional institutions never forgive. Not because they are vengeful gods, but because they have, and are designed around, the ability to truly forget. This gives them the ability to arbitrarily rewrite history as required for justifying actions, instead of living on the forgiveness fork of a multiverse of possible histories. My favorite exposition of this principle is to be found in an episode of Yes, Prime Minister, where Sir Humphrey Appleby expounds on the advantages of treating officially recorded history as the true record of what actually happened:

Ah, Prime Minister... It is characteristic of all committee discussions and decisions that every member has a vivid recollection of them and that every member's recollection of them differs violently from every other member's recollection. Consequently we accept the convention that the official decisions are those and only those which have officially recorded in the minutes by the officials, from which it emerges with an elegant inevitability that any decision which has been officially reached will have been officially recorded in the minutes by the officials and any decision which is not recorded in the minutes has not been officially reached even if one or more members believe they can recollect it, so in this particular case if the decision had been officially reached it would have been officially recorded in the minutes by the officials. And it isn't so it wasn't.

This understanding of the function of meeting minutes in an organization is much more than just formal structure serving as myth and ceremony: it is historical memory being designed and deployed as in instrument of power, to sustain a particular course of promises made and broken, broken and forgiven. In particular, if it didn't happen according to the official record, it requires no forgiveness. By choosing what to remember, you choose what needs forgiveness. And given sufficient scale, the ability of institutions to truly forget and rewrite history can overwhelm the smaller-scale inability of humans to unremember. That smaller-scale feature of humanity operates at the level of living individuals and small tribes that make-do with oral histories. In other words, the inability of humans to forget extends to Gemeinschaft, but not to Gesellschaft. Usages like forget I said that  represent forgiveness operating in Gemeinschaft, or community. The linguistic convention does not correspond to the literal action alluded to, since you cannot actually forget I said something through force of volition alone. You'd need some brain damage, induced by alcohol perhaps. In Gesellschaft, you could just have the local Humphrey Appleby record history as it is most useful to remember, after the fact, after utilities and implications have become clear. This is the default. When institutions need to actually remember, they use one of two major mechanisms to imperfectly imitate forgiveness. The institution of confession in Cathlolicism is of the archetype for the first mechanism: turking. If you think about it, that isn't actually impersonal institutions doing any forgiving. Instead, the human capacity to forgive is leveraged. Particular (in theory) worthy human beings -- priests -- are mythologized as gatekeepers of heaven. Their ability to forgive is institutionally extended to cover transgressions that are not specifically against them, and may not in fact be forgivable in other contexts. The priestly power to forgive is a power to forgive on behalf of humanity at large. You could say that priests are the mechanical turks of artificial forgiving. Homunculi within machine-like, rule-bound, outside-of-time atemporal institutions that do the one thing no set of rules and procedures can: forgive another human being. And remember what that means: to accept a memory but forbear from action on the basis of that memory. Institutions that are not built around priestly figures generally rely on the second mechanism: using the ability to officially forget and remember to produce a more accurate record of history, rather than an Appleby-like fabrication. Structured, procedural more-true rewrites modulating the impact of unstructured, natural remembering. The most familiar example of course is to be found in courtroom proceedings:

The jury is instructed to ignore that remark.

Strike that last remark.

Let the records show...

The court ordered those records sealed/unsealed 

This isn't forgiveness, but nor is it arbitrary historical fabrication. It is consensual rewrite of history through an adversarial process that is hopefully more trustworthy than the fabrications of an internally cooperating bureaucracy. The advantage of the priestly method of turking forgiveness is that it can handle even the most ambiguous and messy record of events. Good priests can hear an account of your worst crimes, feel the pain your victims felt through empathy, and grant you absolution. Priests are theoretically ideal virtuous beings, worthy of being appointed the gatekeepers of heaven, precisely because they are supposed to be immune to the moral hazard of not actually being the ones transgressed against. So they can be trusted to dish out forgiveness on behalf of others. In theory. The advantage of the secular, judicial approach is that it relies on policing the act of record-keeping rather than directly dealing in promises and forgiveness. Judge and jury are algorithms trained to run on the official record of precedents and admissible evidence. They are supposed to be above the passions of conflict and produce procedurally fair outcomes and records. In theory. With both approaches, in practice, we have two disadvantages. A small one, and a big one. Blood Money The small disadvantage is that both turking and official forgetting involve reposing a great deal of trust in fallible individual humans. Neither priests, nor judges and juries, are generally deserving of the power to fake forgiveness on behalf of others. The big disadvantage is that neither actually solve the problem of artificial, institutional forgiveness. They merely sweep it under the rug of Gemeinschaft through acts of symbolic book-keeping. The limitations of Gesellschaft turn into the hidden wounds and shadow debts of Gemeinschaft. Consider:

• A marauding colonial conquerer enslaves parts of an entire continent, and a guy in a box grants him absolution on the basis of a commitment to say a few prayers. But the survivors of the genocide neither forget nor forgive.
• Rules of admissible evidence within the world of judicial due process let a murderer go free, but the community cannot forget or forgive, and simmers for decades, until all concerned are dead.
• A thoughtless juvenile commits a crime, and the records are officially sealed.  But Manipulative CIA Guy blackmails Former Juvenile Delinquent into doing a bit of spying based on blackmail using "sealed" records.

One does not simply forget that something has happened. Thanks to the nature of  human memory, events are etched in the communal mind in the form of a collective, irreversible, historical consciousness. Memories are carved into this consciousness through experienced emotions, pain in particular being the important one. If proof-of-work and proof-of-stake supply the logic underlying the engineered irreversibility of the the blockchain, proof-of-pain supplies the logic underlying the natural irreversibility of social memory. Gemeinschaft remembers through pain what Gesellschaft conspires to forget by design. Here is a mnemonic to help you remember this principle: the most perfect money is blood money. It is informationally complete with respect to the agonies and ecstasies of history that it represents. To the extent that all institutions represent some form of social currency, whether or not that currency is made legible in the form of banknotes or tokens, they are imperfect to the extent they forget. And that is why, in a world that is not reversible, the ability to forget is a bug, and the inability to do so, a feature. The great promise of the blockchain then, is that it can enable the design of institutions that more accurately capture the natural human condition. In computer science terms, blockchain-based institutions will have the necessary expressivity to model the human condition in its full, bloody glory. Institutions that can officially forget are institutions that function as reversible computers, always under the threat of pwnage by superusers like Sir Humphrey Appleby with the power to monopolize history. They are institutions meant for idealized, timeless worlds, where remembering history is a nice-to-have feature rather than a critical one, and where the incentives are stacked against remembering the unpleasant parts. That is not a good thing when you are trying to inhabit a world that is not timeless, where the affairs of humans are not reversible. Why institutions have been able to get away with being built this way for so long has an interesting answer that I don't have room to get into in this post (it hinges on the the fact that commitment is often an acceptable substitute for consensus). But the point is, we no longer have to accept this state of affairs. In the post-blockchain world, we will have the ability to never forget in a very flexible form. And we will have to re-solve problems traditionally solved by forgetting in a new way. Forgive or Fork-it The historical record is an incredibly powerful source of agency and legitimacy. In even the worst wars, humans have sought to justify their actions with respect to the historical record, and the official record can often justify actions that collective memory would not, and prohibit actions that collective memory would permit. Transitions from peace to war, and war to peace, rest on acts of ritual and institutional promising-making, promise-breaking and promise-break-forgiving. And all three rest on the institutional ability to forget what humans naturally remember. You might even say that winning institutions are the editors of collective human memory. It's not just that winners write history; those who can write history tend to keep winning if they write it well.  It is hard to overstate the significance of this point. The lack of irreversibility is the root of utopianism in all institutional design. When you can rewrite or entirely forget your history, you can self-mythologize. You can make mistake after costly mistake, and keep wiping the slate clean. You can pretend uncomfortable things never happened. You can convert entire peoples to new religions and force them to adopt new names. You can cook the civilizational books. Humans live in an irreversible, messy world with inconveniently enduring memories, but human institutions seem to live in a sort of Groundhog Day space, where they get to evolve in a space of cost-free do-overs, where time doesn't matter unless you want it to. Where meeting minutes, rather than clock minutes, embody the kairos of human affairs. Fortunately for our sanity, it so happens that a convenient truth for you can be an inconvenient for me. So in a distributed system with the right kinds of digital wabi-sabi buttons and no undo buttons there is a possibility that something resembling the truth can emerge. And be irreversibly recorded outside of proof-of-pain human memories, despite every party to it having an incentive to distort it later with the benefit of hindsight. Cryptographically secured irreversible social computing, driven by autonomous agents with varied motivations and autonomous agency, is basically a reboot of history itself. Big History and Simple Commitments I remember, as a kid in school, my history teacher remarking, Alexander's invasion is the sheet anchor of Indian history. Indian history until that point had been a fanciful mix of myth-making, legend, and cherry-picked facts of convenience. With Alexander's invasion in 326 BC, a credible, time-stamped record of events, begins. With the invention of the blockchain, history, for the entire world, is about to be taken down a couple of notches in wishfulness. The shift from history-writing to blockchaining underway today will be as big as the shift from myth-making to history-writing that began at the dawn of the Axial Age around 800 BC. For the act of recording to create social realities with the force of history behind them, they must represent a meaningful current consensus among all of those who will, in the future, seek to justify their actions on the basis of the record. You will not easily accept a record of history that spells future doom for you, and in that reluctance, and agency to act on that reluctance, lies the freedom of truth in the future. From tribal talking sticks to Robert's Rules of Order for parliamentary proceedings, we have greatly increased the complexity and capability of our commitment-without-consensus technologies. But we've reached the end of the line. We cannot get to greater levels of of commitment and collaboration using technologies built on top of priestly turking or court-ordered history rewrites. Institutions designed on those premises are at their limit; at their breaking point. To get beyond those limits, we need consensus (and dissent) technologies and institutions based on true forgiveness at scale, grounded in extra-institutional technologies that never forget. The truth, as the saying goes, will set us free. Especially if it is crypotgraphically secured and independent of the history production processes of any truth monopolists. Better history is more painful to write, but creates more powerful potentialities for action than mere ungrounded commitment. Remembering and forgiving is better than straight-up forgetting because the truth is a more powerful force than coercion in the long term. Institutions that are built on these assumptions are likely to be fundamentally more powerful. Just as the heights of the tallest building in the world skyrocketed historically after the invention of steel-reinforced concrete, the scale and power of institutions based on truth-reinforced consensus will be the equivalent of skyscrapers in the space of beliefs. Such institutions will, hopefully, create histories that justify peace more easily than war. When you agree to agree on more things about the past, there are fewer reasons to go to war in the future. Or so we hope. As I write this, we observe Towel Day, and if there is one thing Douglas Adams taught us, it is that the truth is not necessarily a force for peace. But even if that turns out to be a false hope, we can say this much: if we're going to keep having terrible wars and violence anyway, let's have interesting new kinds, based on deep disagreements about great truths, rather than deep delusions about great lies. Either way, we are at the beginning of  a phase transformation in the nature of the institutional landscape and history itself. To snowclone the title of a seminal Big Data paper, bigger histories and simpler commitments beats smaller histories and more complex commitments. History as a Forgiveness Record History will likely classify the blockchain as a nick-of-time invention. It appeared just as the internal contradictions of industrial-age institutions, based on commitment-over-consensus, forgetting and utopian memory rewrite conventions, were about to cause systematic collapse in the institutional landscape. Kurt Vonnegut had a definition of history: it is simply a list of all the surprises so far. Paul Graham has a related definition of history: it is simply all the data we have so far. I have one that I think combines the insights in both: history is simply everything that has been forgiven so far. Surprises are data. Surprises are also what cause promises to be broken. But the human story begins when broken promises are forgiven, not when inconvenient truths are forgotten. History without forgiveness is just one damn thing after the other. A stream of surprising data that is all noise and fury signifying nothing. In a world that does not forget, there is only one way, besides an act of forgiveness, that history can be created: through the death of those who must ask for forgiveness, or those with the ability to grant it. The meaning of history can change when somebody dies. This, incidentally, gets at why only humans can seek and grant forgiveness at the moment. Rooted as it is in proof-of-pain irreversible memories, if events themselves cannot be erased from history, only death can change the equation. Things that merely experience wear and tear cannot experience pain, or participate in forgiveness interactions mediated by proof of pain. Your beautiful broken-and-mended wabi-sabi tea cup is, ultimately, just an object that experienced wear and tear. Aestheticizing the memories it embodies does not animate them to the level needed to require or grant forgiveness. Until artificial objects can feel pain, they cannot forgive or require forgiveness. But we're almost there with the blockchain, which can record the burden of work, the degree of skin in the game. And perhaps one day, pain. With the right kinds of artificial intelligence built on top of blockchains, perhaps we will get to artificial forgiveness within our lifetime. Which suggests the term smart contract is a misnomer. We should be using the term smart covenant, as in marriages and other death-do-us-part relationships based on irreversible processes of growth and decay, joining and breaking apart, asking for and granting forgiveness. What's the difference? A technology for smart contracts is yet another piece of banal enterprise technology that will be bought and sold by jaded middle-aged executives in bad suits conducting business as usual. A technology for engineering covenants on the other hand, whether or not they are smart, will lead to software eating the world all over again. And I for one, am willing to forgive the whatever we might break in moving fast in that direction. * Disclosure, I hold some Ether and Bitcoin, and am a paid advisor to Colony.io Thanks to everybody who participated in the first ribbonfarm colony experiment (there will hopefully be many more), especially Jordan Peacock, Kartik Agaram, Carlos Bueno and Mark Maxham, who all did a significant amount of non-trivial reviewing work and truly earned their blockchain stripes. Thanks also to Collin Vine of Colony.io for product support through the experiment.