The Ouroboros of AI

In September 2010, Aaron Swartz accessed MIT’s network to download scholarly journal articles en masse. The specific internet address (IP) was banned by the journal for accessing too many items at once, and then Aaron logged in from new IP addresses to download more. In total, he allegedly vacuumed up 4.8 million articles. 

For this act, he faced a 50-year sentence and $1 million fine, plus additional fees and forms of restitution. Ultimately, Aaron Swartz would serve no time in prison, as he died by suicide in January 2013.

Plenty of people harangued government prosecutors for coming down so hard on this individual, and that years of threats, law enforcement actions, and pending doom must have impacted his mental health and well-being.

Sixteen years have passed since that original incident at MIT. Today, we exist in a world of large language models (LLMs) vacuuming up sums of data, literature, science, comedy, art, photography, videos, and more that Aaron would’ve never been able to fathom. LLMs rarely — if ever — have asked permission to scrape all of humanity’s creation. AI companies needn’t stalk the MIT server closet — they collect out in the open for us all to see. We’re becoming accustomed to all our data being used for generative AI.

Sometimes I imagine a journal article, an old personal finance column, or pictures of mine getting sucked up into the AI machines. I never consented, but there it goes. Wee! I don’t have rights to say no (do I?), and I don’t even know who I’d talk to if I didn’t want my work to be included. Heck, even this writing you read right now (and thanks for doing so!) is probably an infinitesimally small crumb of data that Claude, ChatGPT, and Gemini just hoovered up.

We can hear the vacuums of our epoch whirring away in places like Virginia. Frontier models’ motivation to be the first drives a never-ending hum and thrum of data collection, collation, and generation. 

AI needs our knowledge — without it, it is nothing. It’s sucking up our data, generating, and being used to create by us humans feeds a desperate nihilism for many. But my own fears come from the continuation of that ouroboros of AI: when we then create text and speech from AI, which gets sucked up by AI again for its recursive, always-on learning ability. This creates the perfect circle of human disintegration from creation. AI learning from AI to generate more AI. By then, we will be using AI to save human hours at the cost of our humanity.

The CEO of Nvidia, Jensen Huang, recently spoke with Ezra Klein of The New York Times. He analogized AI to a calculator, which has helped us all solve complicated problems with greater ease. A calculator saves us time, makes us more efficient, and more likely to be right. Jensen’s thinking seems to be that knowing how to operate a calculator allows us to spend time elsewhere or learn more about math in other ways. 

If the calculator solves a math problem for us, move on and learn/perform something else. Divert your attention elsewhere. Problem solved.

When he shares this — probably highly workshopped analogy and take-home message within Nvidia’s marketing and PR teams — it sounds plausible and reasonable. I’m certainly no luddite or anti-technology advocate. Calculators are fine pieces of technology. But a calculator can’t be asked to write a 10-page essay on the philosophical principles of Kant or to reflect on the meaning of life or come up with an ethical decision-making model. 

I’ve read AI-generated paragraphs and essays; accidentally, I find them on forums like Reddit and websites providing summaries of ideas. My colleagues in academia face a deluge of AI-generated essays and contributions from students. Having been a teacher and professor for years, I cannot imagine the utter despair I would feel receiving such work.

AI’s generative ability doesn’t make poorly or badly written text; frankly, it can be a lot better than the average person. Spelling and grammar tend to be tight and precise. The text is understandable. But it lacks something hard to fully explain. The generated words feel lifeless, meaningless.

In contrast, humans err. Humans write stylistically. Humans don’t always follow the collective convention. When you see those quirks in writing, I see someone. There’s a chance to find something new and exciting in the words.

Writing serves two key purposes for me: 1) to sharpen my ideas and help me learn more deeply and 2) to share concepts with a broader audience for shared thinking. AI seems to strip us of both these potentials, resulting in poorer educational outcomes and a pointlessness in our AI-generated published word. Every word AI generates for us seems to circumvent our own brainpower — our cognitive energy is replaced with fossil fuels in data centers. We simply cannot learn as much by generating by prompt and providing that language as our own. And more than likely, your audiences will glaze over from the content LLMs create.

Pop culture, media depictions of relationships, love, and sexuality have always affected us humans. It’s taken something, generalized a concept, and published the message to a wide audience. Watchers and listeners and readers have historically been informed by this process — sometimes to the detriment of our relationships, love, and sexuality. The generalizations can become expectations for humanity — whether fair or right or just. 

I am deeply concerned about the conscious and unconscious effects of AI speech, writing, and content creation. Every time I interact with a chatbot, it responds in a style of its own. Similar to the mid-2010s listicles on BuzzFeed, AI aims to be pithy, catchy, and organized in a way that is familiar but, again, lifeless. But how will we all be informed and changed by this style of writing repeatedly, endlessly by AI? How will we start to adapt its writing conventions and styles? How will we start to expect more from our relationships, loves, and sex by its own data analyses?

Our humanity seems to be endangered; not in the way that AI doomers suggest an apocalypse will occur and humans will be wiped out, but in our speech. If we interact with this technology long enough, where do our voices go?

Aaron Swartz wanted information to be freely accessible and understood — in service of science and humanity. The millions of journal articles he stole pale in comparison to the millions of journal articles that will be created by AI in the future to come.

AI generating more AI, generating more AI, generating more AI.