The AI Apocalypses

I’ve spent a large part of the last year reading about AI, using AI and thinking about AI. Young people, tech workers and journalists were the other people I came across who are as involved (or more). After Jacob Coxon, an AI researcher who had worked at OpenAI and then Anthropic, posted on social media that he had quit and that his two former employers were acting irresponsibly and might destroy the world, it seemed like everyone has been talking about AI. Overnight, Coxon’s posts were seen over 100 million times. Wildly, the executives of those companies said that they had huge concerns too and wanted the US government and China to start regulating them. President Trump almost immediately said that AI doom talk was a hoax while the CEO of the world’s top chipmaker downplayed any concerns. The Chinese government responded by saying they wouldn’t slow down either. Since then, the amount of reporting on AI has drastically increased, and chatter on social media and Reddit where the AI Doomers and AI Lovers push their respective dystopian and utopian visions is ubiquitous.

It’s a pretty intense subject and no one knows for sure. It’s fair that people are both alarmed and confused.

I’d rank American society’s AI concerns in the following order: killing humans with viruses or nukes (the Doomers), destroying the environment, eliminating jobs (especially entry-level), raising the cost of energy and hurting people’s thinking, development and socialization.

I have no expertise to speak on the first two. It feels like AI has caused a downturn in entry-level jobs for recent college graduates, but it’s hard to figure out if that is the reason for the difficulty that population has had finding work. I’ll leave that to the economists and labor analysts. The cost of energy has skyrocketed, both in my home state of New Jersey and throughout most of the United States. While the Trump/Hegseth Iran War has certainly led to a spike in the cost of oil and natural gas, electricity costs spiked so much in 2025 that it was a major issue in last year’s NJ gubernatorial election. Once again, I’ll leave the question of AI causing a rise in energy prices to economists and energy policy experts.

As to whether AI is hurting human thinking, development and socialization, based on what I’m seeing from high school, college and graduate students, clients, workers and random people, I absolutely believe it is. A number of the articles I shared in the Summer update covered this, and a few more that appear near the end of this edition do as well. This is the only major concern that I’m qualified to write about.

You’ll have to look elsewhere to read about the other AI apocalypses.

State Regulation

California Senate Bill 903: No individual or company can advertise or offer therapy via AI unless conducted by a licensed professional. AI cannot independently diagnose, make treatment plans or interact with clients. It can schedule, bill and organize records. It has passed the legislature and is waiting for Governor Newsom to sign it. I like the idea behind it but I don’t completely understand it. Is a clinician allowed to sit there with an AI system while working with a client and enter information and then relay what it says back to the client? What the hell is the point of the therapist then?

California Assembly Bill 2575: A direct patient care worker is legally protected when they use their independent, professional judgment to agree with or overrule an AI system’s decision about a patient. This bill answered the questions posed in the previous paragraph. My god. Anyway, this is a good bill. I hope Governor Newsom signs it into law.

CA SB 1119: Adam’s Law. Chatbots must perform safety risk assessments of minors that use AI. Strict protocols and warning systems must be in place to prevent self-harm, suicidal talk or any other suicidal content. AI bots cannot excessively flatter (what is excessive?), claim to be human, encourage inappropriate emotional reliance (what is inappropriate?) or help users bypass parental controls. There is civil enforcement. And fines. I’d like to see some criminal penalties, but this is a very good start. Seems like a landmark piece of legislation. Governor Newsom signed this into law in September.

NJ S4474: Requires consumer safety and disclosure for AI chatbots and companions. New York and California already have a law like this in effect while Colorado and other states have versions coming on the books in 2027. I’ve made note of it here because I live and do about half of my work in NJ.

NY S9051: Similar to Adam’s Law in California, but less comprehensive. Passed the legislature, waiting for Governor Hochul’s signature.

NYC school AI policy: No screens for NYC students up to third grade. No emotional chatbot companions for any student of any age. Teachers can’t use AI to grade. Students can’t use AI until high school. I applaud NYC. I don’t know if it is enough or too much, but I love the effort. Of course, some parents have complained it doesn’t do enough. I get that. If I had kids, I expect I would be even more insane about screens and AI in schools.

NY SAFE for Kids rules: Social media companies cannot make individualized addictive (define addictive) feeds for minors, unless parental consent is given. Minors cannot get notifications between midnight and 6 am, unless (again) parental consent is given. This doesn’t cover AI systems, but screens and social media are big for me and I felt the need to include this here.

On September 8th, the Florida Attorney General said he would ask the Florida Board of Medicine to investigate if AI chatbots are engaging in unlicensed medical practice. He proposed legislation that could impose fines, victim compensation and suspension of Florida operations. Legislation is the answer, but I believe, as noted previously, that criminal penalties would give this a lot more teeth. Hey, even the wild west of Florida is picking up that there is a problem here.

Federal Regulation

On September 17th, Leila Turner-Scott spoke before Rep. Pramila Jayapal (D-WA). Her family is suing OpenAI over the 2025 overdose death of her 19-year-old son Sam Nelson. She alleged before Rep. Jayapal that ChatGPT told Nelson hours before his death that combining kratom and Xanax was safe. Ms. Turner-Scott also claimed that ChatGPT and other AI systems are practicing unlicensed medicine. Jayapal’s official account confirms that those statements were made.

President Trump hosted most of the top AI CEOs at the White House on September 29th. The President announced that they have agreed to self-police. It gave the appearance that his administration is doing something about it. There was no executive order. No proposed legislation. No monitoring mechanism. No potential fine or criminal penalty.

Courts

New Mexico v. Meta In a follow-up to the March ruling, Meta was fined a total of over 900 million dollars for endangering and hurting minors and violating privacy. This is significant because Meta, while mostly known for Facebook, is also an AI company now.

Scott Winters v. OpenAI Regardless of what happens, this is a landmark case. A Florida pastor is suing OpenAI for the bad medical advice he got from ChatGPT. His condition got worse as he kept listening to the chatbot. Now, I expect that OpenAI’s lawyers will argue that he should have gone to and listened to human doctors, but they know that millions of people talk to ChatGPT about medical issues every week. They have not put meaningful guardrails against having chatbots provide feedback on medical issues.

British Columbia v. OpenAI This is one to be debated by law professors. The Canadian province of British Columbia has sued OpenAI in the federal court in San Francisco for assisting the shooter in a mass killing incident in February. BC argues it has standing to sue; OpenAI is trying to get the case moved to Canada.

Anthropic’s $1.5 billion settlement with authors. A lot of authors are getting a pittance and are a little less economically relevant, because since AI was trained on them, they aren’t needed. The danger of AI for many workers is early money, not much after and then irrelevance. I’m simplifying it but the gist is that Anthropic settled claims involving the use of books without permission to train Claude (there are ongoing lawsuits against other AI systems, most notably OpenAI). The settlement is low, in my opinion, and doesn’t truly materially help the writers.

With so many high-ranking AI executives and engineers publicly talking and writing that AI could destroy the world or kill all humans, some people have brought up that those statements could hurt them in civil and criminal trials down the road. I agree. I believe lawyers will relish using these statements against those companies and their executives.

Midterms

According to a few different polls I’ve seen, a majority of Republicans and Democrats throughout the country are anti-data centers and anti-Flock cameras. A stronger majority want AI to be regulated:

NYT/Siena (Sept. 15): 61% of voters oppose building AI data centers; 34% support it.

CNN/SSRS (Sept. 24): 71% responded that the federal government isn’t doing enough to regulate AI. Holding that position are 85% of Democrats, 67% of independents and 63% of Republicans.

Between those polls and AI executives ramping up their “AI could kill us all” talk, more politicians are railing against AI. But I wonder how many of them use it and understand it. It’s not a question we’ll get an answer to anytime soon, but I just want it on the record. In mid-September, President Obama urged Democrats to make AI regulation and child protection a core part of their platform. I clearly believe that is the correct instinct, yet I haven’t seen Democrats offer up any concrete plans. Nor have any Republicans, for that matter.

I’m curious to see exit polls on November 3rd about what voter concerns were. I suspect that voters will rank AI higher than politicians have weighed it. I believe there should be a question about AI’s effect on the economy in order to measure this properly.

New Products

On August 18th, OpenAI announced ChatGPT for Teens, which is supposed to protect people under 18. The chatbot will shut down conversations about eating disorders, explicit sexual content, violence and self-harm. In some cases, ChatGPT will contact adults who signed up for parental controls and notifications. ChatGPT for teens is also supposed to keep adolescents from incorrectly believing that the chatbot has feelings. This was OpenAI getting ahead of Adam’s Law, but I fear it will be less rigorous in states that don’t have similar legislation.

ChatGPT has a new trusted contact feature that it announced in mid-September. This is different from the teen feature with adult contact, as users of any age can now have one. A user enters their trusted contact and if they later show some signs of self-harm or suicide, that person is alerted. Trouble is, the feature is quite hard to find. It’s a good idea, but it should be much easier to find. A better option would be to ask people for a trusted contact when they sign up or log in.

Education

An article in the Atlantic in late June discussed the strengths and weaknesses of AI tutors. The basic idea is that AI hasn’t figured out how to motivate students. Many teachers can’t either, but the good ones show they care, are interested in students’ success and spark desire by being interesting or inspiring. Additionally, learning around peers helps motivate, which doesn’t happen as much when it’s just a student-chatbot relationship.

A good student will use AI to help explain a subject to them. They’ll use it to practice more and get instant feedback. An unmotivated student will use AI to skip reading and write their papers.

This is also true for professionals. A skilled clinician, lawyer, teacher, writer or policymaker can use AI as a force multiplier. They know what questions to ask, when to distrust, how to use an analogy that is perfect for the audience and how to employ real judgment based on their experience. A weak or inexperienced professional may use AI to come up with ideas and give them the finished product.

This story out of Brown made the rounds in July. This spring, students signed up for a class when they learned that the professor was allowing take-home exams for the first time. Most appeared to cheat on the midterm, which the professor easily detected by the record high scores. When he changed the final to in-class, some students with great midterm scores dropped it. Most who stayed had terrible scores. It was a stark lesson for professors around the country about the probability of widespread use of AI to cheat in college. I have said and written this over and over; if a student or worker is using AI to produce their work, why wouldn’t an employer just hire someone in a country with much lower wages to use AI to produce that work?

But…apparently college students aren’t the only people on campus who are dishonestly using AI to write. On September 29th, an article appeared in the Times about Dartmouth investigating its provost for allegedly using undisclosed AI in numerous aspects of his writing. Students interviewed for the article thankfully (and hopefully, not hypocritically) expressed outrage.

In August, a Times reporter interviewed numerous faculty members about what is lost when students stop writing. The conclusion was that thinking deteriorated.

In what seems to be a stark reaction to the above piece, the Times published an opinion piece by a Williams College professor who argues that AI-assisted writing is OK in many circumstances, and that writing is not the highest act of thinking; but rather speaking is. And that if people can speak better, it might lead to more socializing. In the comments section, the author adds that he wished he had talked about how Socrates was horrified by the technology of writing, and that it would offload (his word, not Socrates’s) memory. Which is true. I believe that both writing and speaking are high forms of thinking and that we should do as much as we can to make sure that people are good at both.

In my summer update, I wrote:

Part of the reason I am so alarmed about this is deeply personal. 15-year-old me would have used AI to write papers for school; my Dad would have caught it, punished me and explained why I would never get smarter or wiser or better than I was at 15 if I kept doing it. And I would have probably done it again and he would have massively flipped out and that would have probably gotten me to stop. For the most part. What about the 15 year old who doesn’t have a parent/guardian like that?

Workforce/Clinical Risks

This article, by a doctor, is my favorite piece about AI’s effects on professionals that came out this summer. Maybe this year. She has perfectly laid out the problems of AI writing notes. It notices and highlights different things. Another provider can’t feel the emotion behind the note. It alters care. She argues that if a medical provider doesn’t write the note themselves, they may not remember things or make connections as well as if they had written it.

In the transcript of an interview with Nate Silver on August 17th, Jasmine Sun said something quite profound: 

1. Skill gap — A highly skilled researcher uses it to compare thousands of data sets and becomes more efficient. An entry-level worker doesn’t understand that data enough to run the analysis.
2. Motivation gap — A great student uses AI to challenge their thinking; a bad student has it write the paper.
3. Judgment gap — When AI discusses what to put in a backpack for hiking, I can point out what it got wrong. When AI tells me the cause of my lower back pain and what I should do differently, I can’t tell if it is right or not.
4. Autonomy gap — Ms. Sun described this perfectly.

Articles & Studies of Note

This piece from the Times upset me. AI videos of fake people target people with health problems to sell questionable wellness products. Women in their 50s and people with chronic health issues are most vulnerable, according to the article.

This story about fake AI doctors dispensing medical advice made me absolutely livid. One supervisee asked me what is worse, fake medical advice from fake doctors or fake political news. I told him I hated that question, but it brings up something extraordinarily concerning about the internet, social media and AI: the fact that it is very hard for most people to figure out what is real or fake.

The Times ran a profile of a podcaster and tech writer who used AI for therapy after a parent died. It was helpful and free but she eventually had questions about safety and privacy. She also didn’t like the fact that it felt like a chore to update it without getting updates back herself. The feedback she got, while not human, seemed decent though. She said that she eventually stopped using it and returned to talking to humans more.

A 29-year-old woman took her life after using ChatGPT as a therapist. While it gave her some good advice, it wasn’t able to get her to follow up on some suggestions for outside help. It also wasn’t able to connect with her on an emotional level. Her parents read the logs afterward and were furious. In the article, her parents expressed that if she hadn’t relied so much on her AI therapist, she might have been more likely to connect with some humans.

This disturbing piece was published in the Atlantic in September. It was written by two doctors who urge that we should use AI as a primary care provider. Their key argument is that 21% of patients turned to AI because they felt dismissed by their doctor. Some states are moving towards not allowing AI systems to make medical decisions, while the doctors who wrote this piece argue that:

AI is cheaper. AI is faster. AI has more time for patients. AI makes fewer diagnostic errors. AI follows guidelines better. And now, that human doctors make patients feel unseen and dismissed. This is going to form the backbone of the argument for AI primary care.

As if on cue, a tiny bit later, this piece appeared in the Times about using AI to conduct medical care because of lower costs and long wait lists. Alarming. Look, seeing more people, seeing them quicker and lowering costs sounds great. Big-time. But what happens if the AI system gets it wrong and your mom or grandfather dies? Who is responsible? The doctors overseeing it? A bill discussed earlier in California says the licensed provider overseeing the AI system is not liable. Nightmare. The company? Ha. Many of those CEOs had lunch with Trump this week. The engineers? Only the immigrants, given this administration. Who loses their license? Who gets fined? Who goes to prison? What is the disincentive for screwing up?

Here is another thought. Right now, the argument is to pay the AI systems 40-80% of the doctor costs. It starts there. Eventually, it’ll be, “well they do it better so they should get paid more.” And this upends the whole AI argument for the economy in the first place, which is that it will reduce costs and make everyone richer.

Finally, the article calls for one human overseeing thousands of interactions. That sounds like a terrible idea. Eventually, the average overseer just continually pushes a button because they trust the AI system.

In Closing

Adam’s Law in California and the NYC AI/screen rules in schools are fantastic developments. More cities and states should follow those. The federal government has continued to not only do nothing, but, as I wrote in my summer update, President Trump has actively spoken against state regulations. I have not heard of a competent plan regarding AI from any senior member of his administration.

Two major themes run throughout this update. The most important one is the problem of AI systems engaging in medical care. There are strong feelings and statements on both sides. While AI systems may do some aspects of medical care very well, quickly and cheaply, I urge policymakers and voters to consider my list of questions about who will be held accountable when an AI system fucks up a medical case. The second theme is the further explanation of AI as a force multiplier, in that it makes the top experts better and most other people lazier and less effective. As of now, it does not seem to be a force of equalizing, but of separating. While I’m thrilled to use it to enhance my top skills and go further in my professions, I have grave concerns for society and the masses that are using it to offload thinking, development and socialization.

---

Disclosure: A large language model was used for copy editing, minimal research assistance, editorial feedback, hostile review and assessing legal risk in accordance with my AI Use & Writing Standards. All words, arguments, interpretations, advice and conclusions are my own.