"Open" doesn't mean what it used to

In the future I will storytell to my childern about how the word open in computing, someday, did actually mean things, some of them you could be certain about: source was there, or the protocol was something anyone could implement "freely", data could be pulled out of one system and dumped into another, I might also tell them that people didn't always honor that, but still, the word was at least under tension. Back in these days, when you see the word "open", you probably could clone something and run the export and you could also ask whether the spec was implementable without buying a license from one of its authors. People could lie to you about openness, and you could catch them lying.

The word does not work like that anymore, it now means almost nothing, or rather, it means whatever the firm using it wants you to assume on a given Tuesday, in a given press release, for a given product launch. Open is the most thoroughly hollowed-out term in the contemporary vocabulary of computing that I've ever seen.

The obvious case is OpenAI. When the lab was founded in 2015, the name was principal part of its promise; "open" research, models published "openly" and weights and code the rest of us would be allowed to read and audit and build on. The press release said so in plain English: "As a non-profit, our aim is to build value for everyone rather than shareholders. Researchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code, and our patents (if any) will be shared with the world." By 2019 the company had stopped releasing weights because the next round of models was too dangerous¹1. Does that ring a bell?. By 2023 they were a closed corporation with a private training pipeline and a compute deal with Microsoft no other lab could match and by 2025 the corporate restructuring had killed off whatever was left of the non-profit framing, however 'Open' stayed on the door. Sam Altman, the CEO, kept talking like a public servant while the firm ran one of the most aggressively proprietary research operations in the history of the sector.

After what I see recently in startup launches and in new products I see every in a while, I would say that this is not unusual at all, however, it is becoming unusually visible. Take the newer phrase open-weight, it's a careful concession to the older vocabulary. The weights are released, but under licenses that forbid using them to compete with the issuing firm past some user threshold, or forbid downstream training of certain kinds, or quietly revoke if the firm changes its mind about a category of customer. Meta's Llama license is a textbook example, with monthly-active-user threshold past which you have to go negotiate with Meta directly, a model is open in the sense that you can download a file and "open" it. It isn't open in any sense that survives the OSI's definition of open source, or the FSF's definition of free software, or any older meaning that hasn't already been negotiated away, Meta or whoever phrased Open-weight, coined it to keep the goodwill of open (or whatever corporate marketing wants to call it) and drop the constraints.

The same trick is run elsewhere though, Open access in academic publishing now routinely means that the author has paid an Article Processing Charge of several thousand dollars to a publisher whose marginal costs are close to zero, while the paper itself is licensed under terms that often forbid reuse outside narrow academic settings. The reader does not pay, the author pays and the institution pays the author's grant, which paid the publisher, the system has been opened by being paywalled at a different point in the chain.

Open standards has, in much of its life, meant a document over which a single firm exercised effective control through working-group capture. The most baroque case is the OOXML standardization fight in 2007 and 2008, where ISO was flooded with newly created national bodies in time to ratify a six-thousand-page Microsoft specification that no independent implementer could fully realize.

Open banking, which I got to learn about in my current job is regulatory boilerplate for a set of mandated APIs that ordinary customers cannot meaningfully use without becoming clients of yet another fintech intermediary. Open data portals, in most of the governments that promised them after 2010, now publish either nothing or stale extracts whose schemas change without notice and whose maintainers have long since rotated off the project.

Open²2. Ironically enough, Open society was Karl Popper's specific philosophical term in 1945, it was meant to be against fascism and Stalinism, it is now most often the name of a foundation network and a piñata for several different conspiracy traditions, with very little of Popper's argument surviving into the present register of either side. has been eaten more recently, and more efficiently, because the people doing the eating were technically literate enough to imitate the surface signs. Orwell saw the structural move in Politics and the English Language, although he was writing about a different decade and a different vocabulary, when a word becomes valuable to power, power will take it, and the word will mean whatever power needs it to mean.

The defense for such words used to be philological pedantry, which is the right instinct, namely, insist on the older sense, refuse the new one, embarrass the people who use the new one lazily.

Open deserves that treatment now. When someone says open, the right reflex is to ask, roughly in this order: open how? open to whom? open under what license? with what audit rights? on what platform? with what restrictions on downstream use? against what risk of getting revoked next quarter?

I have come to hate the word, then, in the way one comes to hate any vocabulary one has watched degrade in real time. It is an embarrassment to the things it used to describe like genuinely open projects; OpenSSL, OpenCV, OpenOffice OpenStreetMap, Wikipedia, Debian, PostgreSQL, these deserve a word that has not yet been hollowed out by people who have learned how to make releases without releasing anything. Until that word arrives, the next best move is to refuse to take open at face value, and to ask, with what should now be reflexive cynicism, what exactly is being closed off this time under this cover or corporation marketing technically illiterate terminology. #Programming #Politics #The free software movement is a political movement