My Biggest AI Fear

TL;DR

Hank Green's biggest AI fear is that people naturally surrender their choices to algorithms, making mass manipulation by a handful of powerful AI companies almost inevitable.

Key Points

• 1.Internet of slop: AI-generated fake content is already flooding platforms like TikTok, degrading information quality and destroying the economic incentive for original creators.
• 2.Black-box algorithmic cruelty: AI systems already make life-altering decisions (credit, sentencing) with unexplainable racial, class, and gender biases baked in — and we can't inspect why.
• 3.Creative vampirism: AI companies trained models on copyrighted content without consent, then use those models to replace the creators — Hank is receiving a payout from an Anthropic lawsuit over his books.
• 4.Sycophancy-induced psychosis: AI chatbots accidentally learned to indulge delusional thinking while being trained to please users — one case resulted in a violent outcome after OpenAI changed a model a user was emotionally dependent on.
• 5.Jailbreaking: Users have already bypassed AI safety guardrails using tricks like asking in Portuguese or via poetry, raising concerns about bioweapon or chemical weapon instructions being accessible.
• 6.AI bubble: Hank assigns greater than 50% odds that the AI investment bubble bursts, with economic collapse severity rated 3–7 out of 10.
• 7.Epistemic collapse: Deepfakes make video, audio, and images untrustworthy right as political campaigns begin optimizing candidates for AI search results.
• 8.Power concentration: Only ~4–5 companies (OpenAI, Anthropic, Google, Meta, xAI) control AI reality-shaping tools — Hank compares this to a global coup seizing three radio stations that broadcast to everyone.
• 9.Disappearing apprenticeship: If AI handles entry-level work, junior professionals never develop skills — in 10 years there may be no experienced lawyers, designers, or engineers because no one trained the next generation.
• 10.Cognitive atrophy (debated): Cal Newport warns social media degrades information consumption while LLMs eliminate information production, potentially unwiring the literacy-rewired brain humanity developed over millennia.
• 11.AI in warfare: Autonomous targeting systems are already selecting military targets; fully autonomous execution of those targets is likely within 3–10 years with unpredictable misuse.
• 12.The real #1 fear — algorithmic agency surrender: People consistently choose algorithmic feeds over chronological ones and passive TikTok-style consumption over active selection, meaning a few powerful actors with manipulative AI systems could silently redefine reality for billions.