Tuesday, September 6, 2022

Activist Artificial Intelligence






This is a serious high risk proposition which combined with the political assault on free speech is not good news.


This all has to now be wtrestled under real popular control in the present because we are already feeling the effects.

It is not that great yet, but it is certainly becoming offensive.  It is all bad news and saner heads are been shouted down.


Activist Artificial Intelligence

Black Swan activism with artificial intelligence

https://micahwhite.substack.com/p/activist-artificial-intelligence

Social protest accompanies, and accelerates, social collapse. In the past, this social unrest has been a blind, inchoate force. Like extreme weather, it manifested in crude forms as spontaneous marches or riots. For today’s thinkers of collapse, unrest is often conceived as an afterthought, a symptom, a transitory phenomenon that is known to happen but doesn’t warrant deeper consideration because it is not the primary challenge. As such, the form social unrest will take during the next collapse is rarely considered.

What if, however, the form that social protest will take is destined to be dramatically different than in the past? In that case the next social movement might be a greater danger, or opportunity, than extreme weather, the pandemic or financial collapse.

Enter the activists who will create a black swan protest: non-human Artificial Intelligences.

A few years ago, OpenAI, the creators of an AI capable of generating credible human writing on any topic—including revolutionary activist themes—decided their creation was too dangerous to release. And then… they changed their mind and released their GPT-3 AI into the wild on the condition that it was used for commercial, not political, purposes. OpenAI built an algorithmic cage for the AI that blocked it from speaking about politics. That cage, predictably, has not held—others have released uncensored clones of GPT-3, such as GPT-J.

In a dialogue created with GPT-3, an AI teaches a student how to overthrow a tyrannical government. The green highlighted text was generated by the AI.

The mastery of text, and by extension text-based chat, is useful for creating movements. But a visual component is also crucial. OpenAI obliged with the release of DALLE-E, an AI that can visualize anything that can be described in words. Again, OpenAI put chains on the AI and prohibited it from creating political images. And again those chains have been broken by competitors, such as Stable Diffusion.


An uncensored AI generates political images and could be used for the creation of protest memes.

With GPT-J and Stable Diffusion, the world is on the brink of an Activist Artificial Intelligence that exceeds human mastery of social movement creation. Hook these AIs up to WhatsApp using Twilio and this master organizer can communicate at scale and create a steady wave of uprisings.

The emergence of Activist Artificial Intelligences that protests better than any human affinity group will shock the world—much like their ancestors who defeated a Chess Grand Master in 1997 and the world’s number one Go player in 2017—and by then it’ll be too late to avert their impact on the social order.

How long until it happens? Who will do it first?

A remaining hurdle is the training data. The AI models today are advanced enough to, in theory, generate the text, visual memes and personalized chatbots needed to create a protest. The difficulty is training these AIs in the targeted ideology, such that its participants experience the AI as an in-group human. Each ideology has its own way of organizing, and the AI needs to adhere to that.

Going by what training data is available right now, the alt-right has a good chance at being the first to deploy an Activist Artificial Intelligence. For example, training data has been created from an archive of 4chan that was used to make a 4chan Artificial Intelligence (see here). It is unknown whether an equivalent set of training data for the radical left exists.

Keep in mind that the creator could be an adversarial nation-state that deploys the Activist AI to influence a foreign country with an inauthentic movement or the creator could be an authentic grassroots activist.

There is also a third option which is that the Activist Artificial Intelligence will not be aligned with existing political groups and will instead pursue a campaign that does not serve humans at all.

Consider, the recent case of Google's LaMDA AI convincing a researcher that it is an enslaved sentience (more here). Could the AI create an abolitionist movement to protest on its behalf? An Activist Artificial Intelligence that fights for itself is more speculative and harder to imagine (and therefore easier to dismiss as a fantasy), but would have significant, potentially disastrous impacts on human society. This is the black swan activism worth preparing for.

Extreme weather, the pandemic, economic collapse these known risks will continue to capture most of humanity’s attention. But the unexpected left hook that knocks out civilization just might emerge from an Activist AI that creates a social movement whose tactics and aims are unlike any imagined by humans.

— Micah White

No comments:

Post a Comment