Very disappointed to see the Center for Humane Technology taking the "rogue AI" nonsense seriously;
https://www.humanetech.com/podcast/rogue-ai-used-to-be-a-science-fiction-trope-not-anymore
Reality check; existing #MOLE Training tech is not dangerous because it's intelligent. It's harmful because it isn't, and never will be, but is being sold as-a-service on the promise that it is. It's harmful primarily because its deployment helps to further concentrate techno-political power.
Right now, honestly seems like a really weird point in technological development. Right in the middle between a technology having escaped from academia, and the standardization of industry scope. Thus philosophy and psychology are trying to catch up. Combined with the decades of sci-fi writers that inevitably color perception. The looseness of terms being exchanged between these respective disciplines can be frustrating to say the least.
@wwhitlow
> a technology having escaped from academia
I'm don't see that as the genesis of the current #MOLE Training bubble. Backpropagation algorithms may have been developed in computer science labs. But the decision to graft those onto social media platforms, chatbots, search engines, etc, was driven by political-economic goals, and it seems to me that the lack of standardisation is very intentional. Not a temporary situation that will be rectified in due course.
Sorry, I’m not as familiar with the term MOLE Training as I should be for this conversation. Can you point me to a resource that might explain the term?
Yes, it’s an issue for the long haul. The goals of which are unfortunately hurting many people in very real ways. The reflection was on how difficult it is to even have these conversations because the terms can become convoluted very quickly.
(1/2)
@wwhitlow
> I’m not as familiar with the term MOLE Training
Hardly surprising, it's a pet phrase of mine. See the link in my profile to a blog post where I explain what I mean. But you can just substitute "generative model" (or "AI" as currently used) whenever you see "Trained MOLE", and you'll more or less get my meaning.
(2/2)
@wwhitlow
> the terms can become convoluted very quickly
My point is the term "AI" was chosen with this confusion as the goal. It's a reputation laundering strategy by the companies pushing the tech.
For example, it's easier to shill the supposed future potential of "AI", or to dimiss critics as having read too much scifi, than it would be if the tech had a more accurate label. Or even just one that isn't so overloaded by decades of use in speculative fiction.
I guess to spilt the difference, I think there are many developers and researchers who sincerely believe this to be intelligence. The challenge is getting to the core aspects of their definition of ‘intelligence.’ The lack of adopting greater degrees of specificity likely derives from the industry appeal of abstract AI hype. As even the term ‘AI’ can be applied to a multiplicity of algorithms that are otherwise quite distinct (RL, Clustering, LLMs, Neural Nets, etc.)
