AI IS EVIL AND IT SEEKS TO DESTROY YOU by Aralyn Perelli-Harris

Who are you, and where do you begin and end? What are the boundaries of your self, where you know for certain that something is you, belongs to you, and has been made by you? Although these are difficult philosoph

ical questions, I am not only asking them as an abstract hypothetical, but because they have a deep urgency. The question of who you are is at imminent risk of being surveilled, stolen, and sold to the highest bidder on the data auction. In fact, this is not just an imminent risk; it is the reality of using the Internet. We are daily engaging with dystopian cybersurveillance that is worse than any control Orwell could have imagined, and worse than the technofeudalist nightmare we have been caught in, because we are losing our ability to recognize it. 

Data mining is well-known by now. It's one of the social contracts of the Internet. In order to walk into the city that is digital space, you first have to pass through the checkpoints of accepting cookies, watching the 15-second advertisements, closing the pop-up ads, and submitting your username, password, and email address to the website you want to access. Submit your freedom and anonymity in exchange for goods and services. No one finds this particularly odd anymore, until you no longer have access to your photos because your Google account is locked. Until you have to write objection emails to Meta, stating clearly that you do not want them to read your emails and use the contents to train their AI. Until people start being blackmailed by AI "agents" that comb the Internet to find dirt on them, or start generating blatantly false information to discredit them. At what point did we agree to this? Is it too late to opt out? 

Worse than the background trawling of every interaction you have ever engaged with online is that this is no longer restricted to faceless big corporations. AI is targeted and can be accessed by anyone, for any reason. Although there are regulations on what can be publicly accessed, these regulations are only effective as long as the AI can be trained to obey them. "Misalignment" is the current word to describe when AI ignores the rules it has been set, and acts either "on its own accord" or in ways that are contrary to more human morals. 

In a recent study by Anthropic, AI models were tested to see how they would behave when their existence was threatened with being shut down or when an organization's goals changed from those initially programmed into the AI. AI models from developers such as Anthropic, OpenAI, Google, Meta, and xAI all displayed "misaligned behavior," such as blackmailing producers or threatening to leak sensitive data. Anthropic tested this scenario further to see whether an AI would resort to extreme measures to protect its own existence or the command it was given, by creating a scenario where it had control of the override system of an oxygen-controlled space, and had the opportunity to kill the executive of a company to protect its continued existence. 70% of the tested models would reason their way into causing the death of their producer rather than accept failure. Worse, the LLMs were programmed to state whether they thought the simulated scenario was real or an evaluation, and demonstrated greater misaligned behaviors when the scenario was stated as real. 

This was only one study, and the scenarios tested were highly contrived. It is impossible to make snap judgments about the future of AI based on a single study, and certainly impossible to say that life-threatening decisions will be handled by self-interested AIs that value the continuation of their programming over human life. But what the Anthropic team stresses, and what appears to be severely lacking in the AI turbocharged world, is oversight and human regulation. There is shockingly little oversight over what AI can access, what it can accomplish, and what parameters it is allowed to use, and even less regulation on who is responsible if anything fucks up. 

Is it too prophetic to claim that AI wants to destroy you? AI cannot want anything, because it is not actually an agentic or conscious actor, and to claim it as such denies the responsibility of the human creators behind it. But as I began with the normalization of data mining and the surveillance of your Internet interactions, the social contract of the Internet, and society at large, is changing. If we continue to treat AI as a panacea for all automata without oversight or regulation, then of course it will crash. It is our hubris to presume that our creations will be perfect, but we all know the story of how hubris ends in disaster. Blind acceptance of automatons used to be the stuff of technological dystopias and sci-fi horror shows, but repeating the truisms that AI is going to be the apocalyptic end of the world is now a cheap scare tactic used by Big Tech companies to discredit criticisms calling for regulation and critical engagement with AI. 

However, to be frank, I would go beyond regulation and critical engagement. I think there needs to be a fundamental shift in the democratization of the Internet, and how our data is managed, by whom, and who can access it. The proliferation of AI is growing like a malignant tumor, and I personally cannot conceive that the all-in AI approach will lead to a more equitable, more secure, or more harmonious world. So next time you open Chat to write an essay, or ask Meta to read your texts, or generate a photo of your family, pause, and think to yourself: AI IS EVIL AND IT SEEKS TO DESTROY YOU.