Author: kemitchell

_posts/2024-11-14-OSI-AI.md

@@ -40,17 +40,17 @@ More directly relevant to what's happening in "AI" these days, the difference be
 
 In the past, source code under open licenses that OSI hadn't approved wasn't somehow "proprietary" until OSI got around to blessing it.  It was "open source software", just not under an OSI-approved license.  Or not yet.
 
-Under the new OSI AI definition, a complete package of data, code, and parameters under a new license---perhaps one written specifically for artificial neural networks, to make clear that it covers software, weights, data, and so on---_doesn't_ meet OSI's definition, even if that license lets anyone do whatever they want.  The new license might be an open source license, but it is not yet an _OSI-approved_ one, and "OSI-approved" is what it takes to be "open source AI" under the new doc.
+Under the new OSI AI definition, a complete package of data, code, and parameters under a new license---perhaps one written specifically for artificial neural networks, to make clear that it covers software, weights, data, and so on---_doesn't_ meet OSI's definition, even if that license lets anyone do whatever they want.  The new license might be an open source license, but it is not yet an _OSI-approved_ one, and OSI-approved is what it takes to be "open source AI" under the OPI's new declaration.
 
 ## Why?
 
-I purposefully declined to burn any time or attention on OSI's AI definition drafting process.  As a result, I don't know where this "OSI-approved" language came from, or whether it was ever discussed in this light.  I won't be spelunking forum archives to find out.  But I imagine it's entirely possible no one involved ever thought about this, that "OSI-approved" simply rolled off the drafter's fingers---and passed under reviewers' eyes---with ease.
+I purposefully declined to burn attention on OSI's AI definition drafting process.  As a result, I don't know where this "OSI-approved" language came from, or whether it was ever discussed in this light.  I won't be spelunking forum archives to find out.  But I imagine it's entirely possible no one involved ever thought about this, that "OSI-approved" simply rolled off the drafter's fingers---and passed under reviewers' eyes---with ease.
 
 If so, that strikes me as an unfortunate, but honest, Freudian slip.  OSI is trying to establish new authority, its second favorite word after "open".  It wants industry to self-submit to its discipline.  Given the amount of controversy in blogs and news pipes leading up to the 1.0, especially around the omission of training data, "consensus"---OSI's third favorite word---is already dead on the record.  Vain but understandable claims of "we had a committee" aside, that leaves the definition itself, standing on its own.
 
-At the same time, I'm not sure there's even a claim to real definitiveness here.  OSI's own blog refers to 1.0 as "the Industry's _First_ Open Source AI Definition".  Sometimes there's a prize for being first, but the very expectation that others will follow---perhaps with "definitions", "guidelines", "principles", or "frameworks" privileging their _own_ institutions---cuts against the idea of a single meaning out there just waiting for some kind working group to put it down.
+At the same time, I'm not sure there's even a claim to real definitiveness here.  OSI's own blog refers to 1.0 as "the Industry's _First_ Open Source AI Definition".  Sometimes there's a prize for being first, but the very expectation that others will follow---perhaps with "definitions", "guidelines", "principles", or "frameworks" privileging their _own_ institutions---cuts against the idea of a single meaning out there just waiting for some faithful conclave to write it down.
 
-Moreover, writing your say-so into a definition is obviously not the strong path to presenting it as universal.  Dictionaries define words in terms of other words they also define, but nowhere do Merriam Webster, Collins, Cambridge, or Oxford define words in terms of themselves.  Something different's going on here.
+Moreover, writing your say-so into a definition is obviously not the strong path to presenting it as universal.  Dictionaries define words in terms of other words they also define, but nowhere do Merriam Webster, Collins, Cambridge, or Oxford define words in terms of their publishers.  Something different's going on here.
 
 If there's a competition among potential mediating bodies to be the one who gets to [kick projects and companies off "Open Source AI Island"](https://en.wikipedia.org/wiki/Survivor_(franchise)#Tribal_Council)---and a serious, lingering question of whether a centralized referee over a vague marketing term is even desirable or necessary---that's great.  The more done to make that transparent to more people, the better.  And let the questions come about what the difference is, anyway, between an organization pronouncing a definition it made up and an organization pronouncing a process of calling it as it sees it.