Do we need to mention the level of autonomy of AI systems?

In a comment to draft 0.0.4, @zack asked to remove the text of the OECD definition Different AI systems vary in their levels of autonomy and adaptiveness after deployment. because

it’s generic (and pretty obvious) information, which doesn’t bring anything to the goal of defining Open Source AI```.

This is part of the OECD definition, which is increasingly becoming the de facto definition used by regulators around the world. My gut says we shouldn’t touch it. What would be the implication to keep this text? What would happen if we remove it?

I agree with Stefano that we shouldn’t change the OECD definition

That last sentence (“Different AI systems vary…”) is not, semantically, part of the definition. It doesn’t say what an “AI system” is, that is done in the previous part of the definition. That last sentence just elaborates on the different kinds of AI systems that exist (and do so in a tautological way in my opinion, but that is beside the main point here).

I think there is value in having simpler/shorter text. Hence the proposal to leave that part out. Arguably, it would not necessarily be seen as modifying the definition, we’re just leaving out something that is entirely useless/redundant.

And I am assuming we are going to have a citation/reference to the full definition anyway, so everybody can find the full text/context at the source.

I just find the extent how autonomy an AI system is, is completely irrelevant to defining open source AI.

We are already doing a partial quote from the reference document. It is fine to only quote the most relevant texts. If not, I will start wondering why the rest part of the lengthy OECD is not quoted.

We appreciate short text because long texts are more prone to faults.

I highly recommend you to read the short explanation of why the OECD changed the definition Updates to the OECD’s definition of an AI system explained - OECD.AI

Specifically, on the concept of Adaptiveness

Adaptiveness: This reflects that some AI systems can continue to evolve after their design and deployment (for example, recommender systems that adapt to individual preferences or voice recognition systems that adapt to user’s voice) and is an additional characterisation of an important group of AI systems. Also, the previous wording, “operate with varying levels”, might be read as describing a single system whose level of autonomy and adaptiveness might change over time, which was not the intended reading.

I think that text is an integral part of the definition of AI system: removing it would mean changing the definition. We can do it but we may as well merge this conversation with this: Is the definition of "AI system" by the OECD too broad? - #12 by stefano

Thanks for the source and the explanation. I think I get your point now.
In the sense that we organize the AI definition part in a modularized way in OSAID (so that we can smoothly switch to another AI definition), I think it is fine to include the verbatim quote without any modification – modification is simply not necessary, and may even cause more issues.

In this way, we can fully focus on the “OS” part in “OSAID”. Any question and issues about AI definition can be relayed to the underlying AI definition module, instead of the OSAID itself.

Let’s keep the original quote as is. In this way, although it is not simple in terms of text, it is logically simple, so that the OSAID drafting work will involve less non-core issues.

1 Like