Thread by @jess_miers

That's one of the major issues with the current discourse around Gen AI and 230. We have to understand the Gen AI stack before we can even consider liability.

IMO 230 could apply to Gen AI for some use cases. techdirt.com/2023/03/17/yes…

https://twitter.com/WillOremus/status/1778502060494798869

In assessing liability we have the platforms that provide the Gen AI services, the developers who create and fine tune the models. We have the folks who create the datasets and the folks who implement the datasets to train their models. We have users who supply inputs.

And we also have the platforms (again) that provide the "guidelines" and guardrails to determine what kinds of AI outputs are acceptable and aligned with the platform's overall editorial position.

Each of these aspects can involve different parties.

The same folks who develop the models may not be the same folks who implement the model as part of their Gen AI service. The folks who created the training data sets may not have anything to do with the model developers or the platforms.

This matters for assigning liability.

(This is also just a high level summary of the Gen AI stack. There is a lot more to it)

So how could 230 come into play:

--platforms that offer plug and play chatbots that use models and data sets developed independently of the platform. (this was the case for DeviantArt's chatbot that relied entirely on Stability AI's model);

--developers who supply models that are misused by the platforms that implement them or the users that jailbreak them;

--users who engage with or repost tortious AI-generated content;

--platforms with Gen AI services that "publish" hallucinations / illegal outputs at the behest of abusive users;

--platforms that implement guardrails that fail against user inputs;

--platforms that feed user input back into the models to improve them;

And then of course you have platforms and developers that purposefully design their models, datasets, and services to produce tortious and illegal content. Or platforms that simply encourage illegal content (like those Gen AI services that create NCEI of real people).

And for those situations we might look to Roommates or Lemmon v. Snap for further guidance.

The point is, the question: does 230 apply to Generative AI is inherently complicated and involves a thorough evaluation the Gen AI stack, the parties involved in the development of Gen AI services (including the users), and derivative liability.

It's not a yes / no Q.

And I'll end with this:

if 230 doesn't protect Generative AI either because the Courts refuse to apply it or Congress amends 230 to carve out AI, we better think of something else because this technology will not survive the legal deluge that is sure to follow.

@threadreaderapp unroll