>> law, technology, and the space between

All content by Kyle E. Mitchell, who is not your lawyer.

You can subscribe via RSS/Atom or e-mail and browse other blogs.

Emancipation by ReferenceI am just a copy of a copy of a copy. Everything I say you have read before.

This post is part of a series, Common Form.

I am a deals lawyer, and I write software. I am an “impending death of the legal profession” apostate. New technology—“expert systems”, “AI”, “machine learning”, “deep neural networks”—will not “eat” lawyerdom, at least not as deals lawyers know it. Deals lawyering has always lived, and taken responsibility, out ahead of what rote systems can administer. Rote systems become problem solving table stakes. And the lawyer who’s defined the client’s problem has nine-tenths of the job done.

If you asked me to spec out what any artificial intelligence system would have to do to replace, rather than augment, the core value proposition of a single respected colleague tomorrow, I couldn’t deliver, even with a perfect training set for yesterday on hand. Feed your neural net Christensen and Susskind and you will get … more Christensen and Susskind, or uncanny simulacra. As for what large networked systems can and will be trained to do: I look forward to declaring more of what I do today “drudgework”, and delegating it to robots. The common denominator will rise, and my responsibility will rise with it. The “lawyering” I want to do is at the margin.

Cynicism about the legal profession’s system-wide efficacy, on the other hand, is mandatory. The facts compel it. And a great deal of pain.

Fortunately, all the fundamental technology required for an efficient, accessible, ethically laudable American deals law profession—in law and in software—has existed for decades. The system is held back by the technology lawyers have adopted, often enthusiastically, early, and at great expense, and not by the tardiness of prophecized salvation, either for or from the profession.

There are many yokes to cast off. I am particularly irked by one.

Cut, Copy, and Paste. Their wands of their invocation—Ctrl, C, X, and V—shine, pearlescent from wear, on the keyboards of young and old. In their glinting reflection of overhead light, you can nearly make out the tears of the redundant scrivener. The very symbol of law office dissent, Bartleby, vanquished!

But Cut, Copy, and Paste were no revelations. Their graphical manifestations—techno-pagan “icons” of scissors, clipboards, stacked pages—allude to their tangible predecessors. The photocopier. The spirit duplicator. Mimeo. Carbon paper. The printing press. The polygraph. All spewing fodder for diligent slicer-dicer-pasters, plus a few less diligent casualties.

Cheap copying precedes cheap networking in the history of the professional computer. The improvement was incremental. The typewriter went electric. The electric typewriter begat the word processor. The word processor reduced to software … for the mini, for the PC, for the rectangles in our pockets.

Alright, if you insist: Software has “eaten” the office. But it has not been thoroughly digested. Skeuomorphs abound. What member of the papercraft pantheon has not been “implemented” for mainframes, “ported” to personal computers, then “cloudified” again? The filing cabinet. The bookmark. The highlighter. The mailbox. A new “industry” has rendered them all, slavishly, in one less dimension. Thus has the “twenty-first-century law office” become a cartoon of the twentieth’s. And we are become cartoon characters, dashing from punchline to punchline at attention-deficit speed.

Bravo! It’s what we asked for.

Bravo! It’s what we paid for.

Bravo! It’s what we got.

Copying’s a wicked high. It pays quick and cheap, collects slow and dear. Tolerance, dependence, and congenital copiers follow in its wake. Ignore the easy cases: Copying a bad precedent. Forgetting what was changed. Notice by undersea telegraph cable in a mobile phone startup’s bylaws. Even the right form, or clause, or snippet, pasted in the correct place, costs more than it should.

Preserving the provenance of any such copying entails … more copies. In fact, among the copy-dependent, the answer is always “more copies”. File that Word doc at court? PDF copy. Start a new doc based on that filing? Messed-up Word copy. Sending those changes to opposing? Clean copy, redline copy. Comparing to standard form? In form copy. In new copy. Out messy comparison copy. Switching firms? Copy, copy, copy, and run.

Five copies are manageable. Fifty copies are a nightmare. Five hundred copies is hoarding, in a banker box or a hard drive. How will you find what it is you think you’ve saved from the memory hole?

Copies do not express identity. They can be confirmed as identical only by comparison to other copies. Selection of copies to compare is always, to some extent, manual. The process of comparison, however automated, initiates manually. If copies differ in format—Word versus PDF, fuzzy PDF versus clean PDF—comparison is manual, too. Buy a light table. Use a window, hope the sun still shines. Hire a paralegal.

Even best case, the act of comparing does not reduce the number of copies. If copies prove identical, trashing one means trashing our record that the same content appeared in that context. So best case, to preserve the same information, the number of copies stays the same. There is no new copy showing nonexistent differences. A ratchet. Copy accretion.

The alternative to copying is reference. Reference to a copy—the terms in the file located here—is just a copy with indirection. But reference to identified content itself is no copy at all. “We agree to this version of these well-known terms. I don’t care where you get your copy.” Net savings: At least one copy. Copy attrition.

Reference, back in the day when the profession was busy getting its Word and WordPerfect on, was kind of a dick move:

Great, thanks for referencing that “standard” industry form. I don’t have a subscription, and it takes three days and a few hundred bucks to get one. Oh, you were on the drafting committee? Why the Hell didn’t you put it on the floppy? Why the Hell didn’t you send it FedEx? Notice. Merger. Meeting of the minds. Asshole…

The copy-computers had proliferated. They did not yet converse. Copying was quick and cheap, networking still slow, limited, and expensive. Resolving references, therefore, remained slow, limited, and expensive.

But these days, the answer to “Where did you get your copy?” is always the same. You got it from the Internet, a worldwide social club of relentlessly chatty computers, the same way you got the reference. It’s in your pocket, for Chrissake. And the device in your pocket has less storage space—space for copies—than the desktop computer it replaced.

The short-term-memory machines in our pockets won out because Internet access is worth more than copy space. We can refer to a practically infinite store of knowledge. We need copy only little of that, and those copies can be ephemeral. Speed—the ability to riffle the infinite in search of the relevant—ranks far higher than space on our newfangled professional hierarchy of needs. You probably care more about 3G versus 4G than 8GB versus 16GB.

One way to identify by reference is to appeal to some stalwart, standardizing authority. “Version such-and-such of the American Institute of Architect’s such-and-such form.” “The such-and-such year revision of the ISDA Master Agreement.” “The FAR provisions at such-and-such C.F.R. such-and-such, as of current date.”

To rely on such a reference to lead to a particular set of terms—to identify the terms intended—is to rely on the AIA, ISDA, the federal government, &c. to notice and mercilessly crush any miscreants brazen enough to pass counterfeits. Realistically, there are better scams. Very rarely will anyone try. So this kind of reference already works well if you want to reference terms “off the shelf” and they are actually on a shelf somewhere. Say, the Library of Congress.

Of course you don’t. You need to mess with the standard. Your particularly perverse, customized edit is the deal, because that’s what your perverse client ordered, and we’ve broken you down low enough to give it to them. How could we possibly identify and reference the degenerate customizations and expedient, extemporized scribblings of every competent practitioner?

We have the technology. We have since the seventies. The technology is called cryptographic hash functions. Don’t click the link yet. Just think “fingerprint”.

Anything in cartoon office land—anything computers can store or display for you—can go into a hash function. Out the other end pops a “hash”. Hashes look like this:


Here is a box, where you can type anything:

Out pops its hash:

The hash function can’t hurt you. Play with it!

While you’re at it, might I suggest a couple of experiments?

  1. Replace the text in the box with a short sentence, like

    I am the walrus.

    Note the first four characters of the hash: “475b…”.

    Now add a word or two:

    I am the best walrus.

    Watch the hash change.

    Now delete the new word, to bring it back to the original sentence. Is the hash the same as before?

  2. Try a sentence with a potential ambiguity, like:

    I asked Bob, a lawyer and a gentleman.

    Make a small, meaningful change:

    I asked Bob, a lawyer, and a gentleman.

    Did the hash change? A little or a lot? Did it get longer or shorter?

Unlike actual fingerprints, “cryptographic hashes” are spot-on for unique identification. There are more possible hashes than anyone can fathom: Each hash is 64 characters long, and each character can be any of 0 through 9 and a through f. 64 slots with 16 possibilities each makes over “115 quattuorvigintillion” possibilities. Big haystack.

There aren’t any folks with mortgages and mood swings and deadline anxiety trying to match these fingerprints on the basis of “objective” criteria. Just cold, hard, Led Zeppelin-era software doing lots of rote arithmetic and counting. Anyone, and any computer, can run a hash function—the National Security Agency invented, published, and publicly licensed the one used above—and come up with the same hash for the same data.

Exciting opportunity! Find a way to come up with data that lead to a specific hash that you want, quit law, pillage the online payment system of your choice, and retire to a private island. Also: Expect a call from the National Security Agency.

In short, the world, collectively, is betting a Hell of a lot more than whatever your deal is worth—whatever your deal is worth—that these hash functions are easy and reliable to work from data to a hash, but impossibly hard to work backwards from a hash to data that produces it. If the hash functions crack, don’t stress over your bar license. Square with the higher power of your understanding.

You may be thinking: These hashes, they look like nonsense. I can’t tell them apart. I can’t keep even a one of them in my head.

You’re right, they’re not exactly three-minute pop tunes. The computers have no trouble with them—they’re an incredibly cheap way to store references—but a computer that dutifully summons a California separation agreement where you needed a Japanese ForEx master agreement because you can’t type 64 characters with flawless accuracy isn’t much help. Time to learn the two-step.

First, consult a reliable source for a list of useful references. If you need a venture capital financing document, that might mean the website of the National Venture Capital Association. For a standard corporate form, perhaps the site of your local Secretary of State’s office. You might even check a book.

What you expect is a kind of directory of forms and hashes. Running with the NVCA hypothetical, you might find:

There is an organization called “National Venture Capital Organization”.

That organization has published two versions of a form called “Model Preferred Stock Purchase Agreement”.

There is a first edition of that form. Its hash is:


There is also a second edition of that form. Its hash is:


If you and I refer to:

The first edition of the National Venture Capital Association’s Model Preferred Stock Purchase Agreement

we can use this directory to find that’s just the same as:

The form with the hash:


We can then ask around for the text of a form with the hash bb1d..., and check for ourselves that the hash of the text we’re provided does in fact hash to bb1d.... We know past any shadow of a doubt that we have the same terms in mind. We could even make ourselves a tidy little contract:

We agree to the form with the hash:



Bob Adviser /s/ Bob Adviser August 1, 2016

SomeCo, Inc. /s/ Joe Founder Chief Executive Officer August 1, 2016

If Bob decides he wants to make on we agreed to different terms later, he has to come up with terms that match the hash we agreed to. Once he manages that, he might as well forget about our deal, pillage the payment system of his choice, and retire to a private island instead.

The infrastructure for all of this has already been built.

Here is a directory of forms I have published under my own name:

Here is the first edition of a debt finance instrument I’ve prepared:

That form as a whole, as well as every part of hit, can be requested by hash. The notice provision, for example, is 7af48f3af1997f7ad8c11d386727fb232ed430f34ef860a99ec7eb6d1fa2e59f

There are only three tricks to this.

First, the software standardizes a way to type out contracts and bits of contracts—from sections and headings to definitions and fill-in-the-blanks—before they go into the hash function. This makes it possible to break a form into pieces, hash each one, and make sure they fit together nicely when combined.

Second, I am not an illustrious, form-promulgating industry group. But I gave myself an account that I can use to publish forms under a user name—mine is “kemitchell”—anyway. If you’d like an account for you, your own illustrious, form-promulgating industry group, or even just a project that deserves its own name, send me an e-mail. Here’s a project of mine, as an extra example.

Third and finally, notice the green lock next to the URL of the webpage in your web browser. The server——is using encryption to keep Internet miscreants from snooping on which directories and forms you request and messing with the responses you get back.

None of these systems will tell you what to write, how to write it, or whether it’s what your client needs. But they can help make the process of pulling together those terms a lot more like shopping than retreating to a mountain cabin to write a masterpiece. Hashes and a standard way for computers to represent contracts make it easy, safe, and reliable to references and reuse others’ work and share your own, without accumulating copies.

As a side benefit, cluing computers in to contracts structure lets us offload the mechanical tasks of checking definitions, validating references, formatting, and, most crucially, filling the details of specific transactions into generic forms. Rather than wait for lulls in our practices to prep new “model” forms or cobble together a CLE packet, we can work on generic forms as we go, sharing the generic bits along the way. We can all be legal publishers.

I’m giving all my work on this away, free of charge and intellectual property restriction. Partly because I think it’s too important to get it right to impose all the additional challenges and constraints of a business venture. Partly because it takes the form of software, and the best software is built in this open way. Partly because, as a practitioner, I wouldn’t trust the project long enough to develop understanding if I had to wonder at the values and motivations of those involved.

I call the project Common Form, and it’s already yours.

As always, feedback is most welcome, by e-mail or Twitter.

Your thoughts and feedback are always welcome by e-mail.

back to topedit on GitHubrevision history