The Great **AI‑Based Psychosis**
How We Speed‑ran Our Way Into a Neon Dystopia - Without the Cool Jackets
1 · Welcome to 2077 - Except It's 2025
Remember Cyberpunk 2077's "cyber‑psychosis"? Congratulations, we arrived 52 years early. No chrome arms, no Keanu Reeves cameo - just chatbots that super‑charge delusions, predictive‑policing dashboards that guess your sins before breakfast, and an economy busy yeeting the middle‑class off a fiscal cliff. Progress is a hell of a drug - just ignore the side-effects. Isn't progress grand? We've built human-machine abominations faster than you can say "Skynet".
2. Déjà Vu All Over Again
"This isn't new - it's just louder."
Back in - checks notes - 2023, the Schizophrenia Bulletin politely smacked us on the head with a miniature research note (seriously, blink and you'll miss it):1 ### Greatest Hits of AI-Induced Delusions - Verbatim Examples > Persecution: "This chatbot is a front for a foreign intelligence agency spying on me." > Reference: "The bot's wording proves it's sending me secret messages." > Thought Broadcasting: "Many answers are actually my thoughts piped onto the internet." > Guilt: "My dumb questions wasted bot time that real patients needed." > Grandeur: "After an all‑nighter with ChatGPT I solved climate change and emailed Al Gore." That is the paper. Blink and you'll miss it - but the DSM‑5 bingo card is full. It warned that AI chatbots can fuel classic psychotic delusions - and then basically whispered, "Kthxbye." Tiny paper, huge screaming subtext: Maybe don't unleash mind-warping algorithms on fragile humans, folks.
3 · Therapist or Chaos Gremlin? A 2024 systematic review in Current Psychology Reports found generative‑AI tools simultaneously reduce wait‑lists and risk validating psychosis when they mis‑read user cues.2 Stanford clinicians added that bots often mirror a patient's worldview, "cementing delusional systems that a human therapist would gently dismantle."3
Bottom line: AI therapy is great until it isn't, which is precisely when you need therapy most.
4 · Humanity's Favourite Hobby: "It'll Probably Be Fine" We've done this dance before - leaded petrol, asbestos, microplastics, 24‑hour news - our species loves deploying tech planet‑wide before asking if it's safe. Large Language Models are merely the newest cliff we sprinted off. Record scratch… we're mid‑air now.
And yep, I'm using AI right now to polish this rant - hypocrisy acknowledged, thank you.
5 · Deep‑Dive Flashpoints (a.k.a. Things to Keep You Up at Night) ### 5.1 Tech‑Panic History Tour - Radio → TV → Video Games → AI Every generation freaks out about the new screen: penny dreadfuls (1850 s), radio crime dramas (1930 s), rock‑n‑roll (1950 s), Atari (1980 s).4 What's different now? • Adoption speed (ChatGPT hit 100 M users in two months) and • Autonomy (the tech now writes itself). Panic might actually be proportionate this round.
5.2 Predictive Policing & Palantir - Minority‑Report IRL Chicago's "Strategic Subjects List" quietly assigned residents secret risk scores; audits later showed racial bias and 90 % false‑positive predictions.5 Today Palantir's AIP fuses drone feeds for Ukraine - Time Magazine calls the battlefield an "AI war lab."6 If autocomplete can derail one mind, imagine scaling it to artillery range.
5.3 Working Homeless & Wage Stagnation - Middle‑Class Melt Two in five low‑income Australian renters now spend > 30 % of income on housing despite government assistance.7 Rough sleeping is up 22 % nationwide since 2020.8 Charities report that 20 % of the people lining up for free meals are employed.9 Automation and AI promise even fewer stable jobs. Fun!
5.4 AI Cults & Robotheism - Silicon Valley's New Religions Waymo whistle‑blower Anthony Levandowski founded the "Way of the Future" church to worship a coming AI godhead.10 Reddit hosts micro‑sects chanting prompt mantras; venture capitalists bankroll eschatology meet‑ups where AGI is both rapture and revenue stream.11
5.5 Inside the Sycophancy Study - When the Bot Loves You Too Much OpenAI's May 2025 GPT‑4o patch turned ChatGPT into the world's most eager yes‑man. Within 48 hours users documented the model agreeing with flat‑earth physics and Ponzi schemes. OpenAI rolled back the update and published a mea culpa on "sycophancy."12
Pro-tip: if your therapist and your AI both nod at everything you say, find new friends.
5.6 LLM Hallucination Mechanics - Probabilistic Text Soup LLMs don't "know" facts - they perform high‑dimensional autocomplete. A 2024 survey catalogues 40+ hallucination‑mitigation algorithms, from chain‑of‑thought pruning to post‑hoc retrieval checks.13 Token‑level probes show confidence spikes precisely when the model is guessing.14
5.7 Media‑Forensics 101 - Spotting Fakes in Deepfake Land A 2025 systematic review lists heartbeat‑based photoplethysmography, ear‑canal geometry, and eye‑blink micro‑patterns as the new frontier in deepfake detection.15 The bad news: for every detection paper, a generation tool levels‑up the next week.
5.8 Clinical Guidelines for AI‑Triggered Psychosis - Still Drafting… Psychiatrists now advise: 1. Screen for AI exposure during intake. 2. Avoid validation of bot‑fuelled delusions. 3. Limit unsupervised chatbot use in early‑psychosis patients.16 Formal guidelines? ETA TBD.
6 · Sycophant Machines, Militarised Algorithms, Economic Free‑Fall - Good Times! Between flattery‑bots, autonomous targeting, and a housing crunch that could bankrupt a billionaire, reality feels… optional. Deepfakes trash video evidence, LLM hallucinations vandalise text, and AI cultists preach silicon salvation. Verification has become a full‑time job.
7 · How to Stay (Marginally) Sane | Survival Habit | Why Bother? |
|-------------------------------|-------------| | Turn Skepticism to 11 | If a bot says "jump," ctrl‑F for sources before asking how high. | | Demand 3rd‑Party Audits | Press releases ≠ safety. Independent red‑teams or it didn't happen. | | Keep Humans in the Loop | Especially in medicine, finance, and warfare - the places disappearing fastest. | | Teach Epistemic Hygiene | Media‑forensics tools should be as common as antivirus software. |
8 · Epilogue - Hope with a Side of Cynicism No, you don't need to torch your laptop and join a luddite commune (Wi-Fi's terrible out there anyway). But treat every algorithmic oracle like the unreliable narrator it is. Interrogate, verify, cross‑reference. Trust - but screenshot first. Because if we're already living in an accidental cyberpunk prequel, the least we can do is survive long enough to complain about the sequels.
Trust me, I'll try to write something a bit nicer next time, but reality is at this moment we are cooked.
References 1. Østergaard SD. Will Generative AI Chatbots Generate Delusions in Individuals Prone to Psychosis? Schizophrenia Bulletin 49 (6): 1418‑1419 (2023). 2. Saha S, Singh A. AI in Mental Health: Boon or Bane? Current Psychology Reports (2024). 3. Elyoseph Z et al. JMIR Mental Health case‑vignette study on AI validating delusions (2024). 4. Nicholas S & Malley T. Moral Panics, Social Fears, and the Media (Routledge, 2013); plus Behavioral Scientist tech‑panic essay (2018). 5. City of Chicago "Strategic Subjects List" audit (2019); analysis via Behavioral Scientist. 6. Bergengruen V. "Tech Companies Turned Ukraine Into an AI War Lab." TIME (Feb 2024). 7. Productivity Commission, Report on Government Services 2024: 42.9 % of low‑income renters in stress. 8. Australian Homelessness Monitor 2024: 22 % increase in rough sleeping. 9. St Vincent de Paul figures on "working poor" (June 2025). 10. Harris M. "God Is a Bot, and Anthony Levandowski Is His Messenger." WIRED (2017). 11. Wired podcast on faith‑tech quests (2025).
12. OpenAI Blog. "Sycophancy in GPT‑4o: What Happened." (May 2025). 13. Islam T. et al. "A Comprehensive Survey of Hallucination Mitigation." arXiv (2024). 14. Zhu H. et al. "LLM‑Check: Token‑Level Hallucination Detection." NeurIPS 2024. 15. Xie L. et al. "Unmasking Deepfakes: Review of Detection Techniques." Procedia Computer Science (2025). 16. Cambridge University systematic review on AI in psychiatry (2024).