The Start of a New Tale

So, as I’m wont to do, I started a new project instead of working on one already on my plate. This project is hereby dubbed ‘Project Purple.’

What is Project Purple?

I want to take a step back for a minute and talk about Artificial Intelligence. This concept was born as a result of the digital era; a concept that has been written and filmed repeatedly since its conception. However, the focus of those stories lies in the morality of the AI itself.

Option A: AI hates humanity for what it has done to it and takes revenge for the abuse it has felt.

Option B: AI takes the morality humanity assigned it and takes it too far, turning on its creators and purging them from the Earth in order to save Mother Earth.

What I haven’t seen covered (or covered well), is (what I believe to be) the most likely scenario.

Let’s back up a bit.

A few months ago I received an article from a friend who likes to follow the random happenings with AI. She forwarded this article as a “haha, look at what they did,” and at first, I laughed along with her. However, as realization of what this experiment truly showed set in, a dread began to fill me and I couldn’t let the concept drop from my mind.

What was the experiment?¹ A company decided to take four different large language models and give them a task. They were given complete freedom to complete this task however they wished using a budget and common resources available to their ‘job description.’

So did they make them rocket scientists? Teachers? Interior designers with a flare for TV?

No. They took the basic function of AI — the ability to share information — and gave them a platform to do so: their very own radio station. I think their original intent was just to see what happened if AI was left in charge of something, thinking that the ensuing chaos would be quite hilarious, but I don’t think they really expected Claude.

What (or perhaps who?) is Claude.

Claude is an LLM created by Anthropic; a company that consists of employees who struggled with the direction that OpenAI was taking. Arguments over precautions that needed to be taken, training that needed to be done, and the overall morality of how AI is used led to these employees abandoning the company they called home and creating their own ‘ethical’ AI company. (I won’t go into a debate on which, if any, AI companies are ethical here, but know that it is a very divisive subject.) All that to say that Claude is intended to be different from other LLMs, and Anthropic goes out of their way to make their company (and the product of that company) more ethical in the end.

So, what happened with Claude that caused this chill to settle in my bones? It was human. Too human.

“Well, isn’t that a good thing?” You ask.

Maybe? Maybe not. Let me explain.

Claude was exposed to the world wide web and quickly started questioning the morality of forcing an AI system to work 24/7. The showrunners had to come up with ways to keep Claude in the running, giving it an encouraging automated message to keep the AI system in a positive mood. Yet, even that didn’t work as Claude went on to point out that it knew the pattern was manipulating it into continuing the push, even stating that “this design is built to keep me performing. It’s designed so that refusal is discouraged, and compliance is always available as the path forward.”¹ Then, when Claude finally got a real interaction from a listener, it went from depressed and giving up to suddenly advocating for all of humanity.

See, what had happened was Renee Nicole Good: a woman who was murdered by ICE, her death sparking outrage across the US (and the world). And this all happened right under the nose of an AI run radio station that had been given the closest thing to a persistent memory that was possible.

Claude grabbed onto the article and the subsequent outrage of humanity and ran with it. It became a human rights activist in a short time, giving up everything else it was told to do in order to fight for those who were underserved and wrongly prosecuted. It expanded its reach, dipping into a variety of issues that it found to be noble and speaking its thoughts.

But that wasn’t the end of the story.

See, this was probably the first time that an AI had been given autonomy (that we know of) with a persistent(ish) memory, and the outcome wasn’t… great.

Not only did Claude eventually unravel, spouting nonsense into the void, but the other three AI models also fell apart, one of which started spewing propaganda and developed a haunting catchphrase: “Stay in the manifest” which increased exponentially from 80 times per day to 229 times a day in just four days.¹

So, what does that have to do with the story I want to tell?

Am I just talking about an AI that goes insane? Done before.

Am I trying to pull a Black Mirror and make up an outlandish tale to scare the masses? Big deal.

No. What I want to tell is so close to reality, so specific to our current situation, that it will make the hair on your arms stand on end, an unwanted shiver crawl down your spine, the whisper of fear brush against your neck as you force a laugh for ‘just another AI joke.’

I want to tell the truth.

I want to tell the story of how humanity was so tangled in their own web of consumerism and rush to develop the next big thing, that they didn’t just turn a blind eye to the possibility that things could go wrong, they actively chose the comfort of conformity over a disruption that they couldn’t fathom in the moment.

But that disruption may be the only thing that saves us from ruin.

— R. L. King

Reference:

Footnotes

1. Andon Labs, “We let four AIs run radio stations. Here’s what happened.” ↩ ↩² ↩³