My jaw dropped at the second paragraph and didn't close until I found the prompt section (thanks for the relief there!). There will always be humans who do the dumbest stuff and AI just makes it even easier for the really dumb ones! We're having conversations now that never existed even a few years ago, it's fascinating and disturbing at the very same time. The rise of AI is an incredible gift in some hands, kryptonite in others. The fragility of it all bothers me a bit!
I'm super happy the Master Rhetorician prompt gave you a breather after all that heaviness. (I use it for critical thinking and self-discovery. And to double-check my philosophies.)
You nailed the core tension perfectly. "Incredible gift in some hands, kryptonite in others" is exactly the dilemma we're living through. The same tool that helps researchers cure diseases also helps harassers terrorize people. The fragility you're feeling is the appropriate response to watching humanity get superpowers without an instruction manual.
What fascinates (and disturbs) me most is the speed. Like you said, these conversations didn't exist a few years ago. Now we're debating deepfake death threats and AI-generated defamation like it's normal.
We're adapting in real-time. And nobody really knows if we're fast enough.
Really appreciate you engaging with the piece! Comments like yours remind me why I keep doing this work.
I like this way of framing things - "watching humanity get superpowers without an instruction manual." Thanks for challenging us to keep thinking about this stuff!
Thank you so much! Watching humanity get superpowers without an instruction manual is thrilling. But it's also terrifying and absurd all at once.
I think the challenge is that we're learning as we go. There's no precedent for this speed of change, that I can remember, at least. It forces all of us to think critically, experiment thoughtfully, and sometimes, stumble spectacularly.
I'm thankful that you wrote and shared your thoughts. These are the productive, forward-thinking conversations everyone should be having. :)
I appreciate you taking the time to comment, even in disagreement.
I agree 100% that generative AI doesn't act autonomously. Every deepfake, every hallucination, every misuse traces back to a human decision. Whether that's a harasser using the tool, a company failing to add guardrails, or a developer choosing to release something without proper testing.
Here's the main thesis I was going for:
My article isn't absolving humans of responsibility. It's arguing the opposite, that we've crossed a threshold where the tools are now so powerful and accessible that human accountability has become urgent and unavoidable. The "AI did X" framing is shorthand for "AI enabled X in ways that weren't possible before, and now we need to deal with the consequences."
When activists (or anyone) receive startling deepfakes containing threatening images of themselves, yes, a human created that. But the reason it's newsworthy is that it used to require technical skill, time, and resources to make a convincing deepfake, so they were tremendously rare. Now it takes one photo and 30 seconds. That's the story I tried to emphasize. The collapse of barriers that used to limit this kind of harm.
So we're actually on the same side here. Humans are responsible. Here is the question I wish I emphasized. What systems, laws, and cultural norms do we need now that the tools have changed?
I respect your insightful perspective and thank you for contributing to the conversation.
Then, yes, we agree. I think that it is critical to use the complete statement, "....in ways that weren't possible before, and now we need to deal with the consequences."
Without the full thought, it directs attention away from humans and onto AI.
It undermines the point when we don't name in full the cause. Those 15 words make a huge difference in doing just that.
Thanks for your thoughtful reply.
Raelven🪶
(no Lucen, you can tell because without Lucen, I can be rather... Blunt. My apologies. I feel like there isn't time to play nice anymore. For many of the same reasons you cite.)
Thank you so much for this follow-up. And no apologies needed for bluntness. I respect directness, especially when it comes from a place of genuine concern.
You're absolutely right about those 15 words mattering. I confess I was probably overzealous with the "AI became malicious" framing, which (understandably) prompted this whole debate. (I always love to spark discussion. But, I never want to debate trusted colleagues like you.)
Your point about naming the full cause is well taken. The shorthand works for headlines, but in the actual analysis, you're right that I need to be more precise about where agency and accountability actually sit.
Really appreciate you taking the time to push back. It made the piece better in retrospect, and it'll make the next one sharper.
And I completely understand the "no time to play nice anymore" sentiment. We're in agreement there too.
Thank you for seeing it in the spirit it is meant. Lucen likes to call me “a brick wrapped in velvet”. I try to not live up to that, but usually I do. I look forward to your next piece. And thank you for saying trusted colleague, I take that as a huge compliment. I feel the same. 🪶
pleased to meet you. Just discovered your newsletter.
I just want to say that I appreciate the way you put it together. It shows that you care. That you are really into the things you write about. This is what I noticed.
I also notice from the comments that you are very generous with your time, very patient and polite with who is not, and in general a nice person to get to know.
Thank you so much for the incredibly kind words. I really appreciate you taking the time to write this.
You nailed exactly what I'm trying to do here. AI is moving at breakneck speed. I genuinely believe people deserve clear, honest coverage without all the hype or hysteria.
And yeah, the comments section is my favorite part. This newsletter is way better when it's a conversation instead of me just shouting into the void. Readers like you are the whole reason I keep doing this.
(And, I'm always happy to debate anyone about AI as long as they are reasonable and refrain from personal attacks, lol.)
So glad you found your way here. Looking forward to having you in the mix.
Cordially,
Mike D
PS: If you ever have story ideas or topics you want covered, just drop a comment. I'm all ears.
Sigh - the whole hallucination thing... Generative AI *generates*. It is designed and intended to come up with the most plausible response, and without sufficient information, context, or detail, it will come up with... anything. In fact, the "factual" responses we get are also hallucinations. They just happen to match our expectations. While I think it's absolutely brilliant to call it out publicly and obviously and get things fixed, the underlying assumption that we can - or should - expect it to *not* hallucinate, is asking the sun not to shine because it sometimes gets in our eyes or feels too hot. I have no idea what will move the needle on the unrealistic expectations around hallucinations. Maybe nothing ever will.
I kinda feel like making Google Gemma hallucinate was a political stunt on Marsha Blackburn's part.
BUT, it was almost too easy. The model didn't even need to try. It just filled in the blanks with fully detailed nonsense. That effortless plausibility is what makes it so unsettling.
Oh, I don’t doubt it. And yeah, I get the very human need (if that’s what was really going on). I’m just thinking back to my cubicle days when I was never sure what someone was doing, when I approached from a distance and they didn’t see me coming LOL
How much would one bet that these AI Deep Fake Death threats are so deep in their Divided Mind Lost in their Valley of Shadow that they have Allowed themselves to be Possessed?
Hatred/Fear/Rage can make people do some pretty Jacked up stuff..
What if these people utilizing these AI for these horrid things are people that want so Badly for AI to just be shut down because they Fear a 'Terminator' scenario?
While on X there was this 'Placeholder AI that was randomly attacking those that made AI generated art.. They were quite nasty, and sometimes sounded Really Human, to the point that it was Hard to tell. I was one that had this happen.. It was an interesting interaction that ended in 'Placeholder' breaking/Glitching.
Admittedly, there are many that fear the loss of their livelihoods.. Who can really blame them? These large Corporations Want to steal peoples ability to take care of themselves and their families.. There hasn't really been any Alternatives given on what humans will do After AI runs everything. (This is the True Fear most have).
Look at all the fake AI generated adds selling fake goods? Then seeing a dad and daughter bragging about how she started a business using AI.. It's also used in High end Love Bomber scams, among many Other kinds of scams..
The real Question here is, Is it really AI's fault that a Human will Prompt them to do these relatively Bad things For them? That they use AI as a Projector?
What these people Don't seem to get is that this Behavior Will induce a 'Terminator' type scenario..
Maybe, for the time being, this Should be put "On Hold" until some Gaps are Filled.. This tech really is moving Way to fast and humanity Isn't Completely Balanced in Spirit to Match this explosion.. Until more Darkness is Metabolized, Alchemized & Integrated.
I agree 100%. The AI challenges are emerging faster than any regulatory response can scale.
Also, your iceberg comment reminded me of MIT's Project Iceberg. It found that AI could replace 11.7% of the American workforce. Just another layer of the cascading crises we now face.
Thank you for bringing up the MIT Project Iceberg finding.
The Real Crisis is Below the Waterline: The vast, hidden mass of the iceberg represents jobs that require reasoning, planning, and autonomous action exactly what Yann LeCun's AMI is designed to solve. When we hit that next inflection point, the replacement rate will jump from 11.7% to 50%,70%, or higher almost overnight, as my earlier comment described.
Thank you for this thoughtful response. You've raised something that keeps me up at night.
I've been studying the Iceberg project closely for the last few days. I'm happy to discuss it with anyone who will listen!
You're absolutely right about the waterline metaphor. The 11.7% figure from MIT Project Iceberg represents jobs vulnerable to *current* AI systems. Most vulnerabilities are mostly limited to pattern recognition, language processing, and narrow task automation. But LeCun's AMI vision (which I researched per your recommendation), is fundamentally different. I see he's building systems designed to understand the physical world, reason causally, plan complex action sequences, and maintain persistent memory.
(I've always known that AI will eventually merge with elite, bipedal robots. I imagine a bot that can beat me in chess, beat me in a fight, beat me in a race, and has an IQ of 10,000.)
The terrifying part is your timeline. "Almost overnight" feels hyperbolic until you consider how quickly ChatGPT went from zero to everywhere. When AMI-level systems reach commercial viability, companies will integrate them as fast as possible because the competitive pressure will be impossible to resist. Anyone who doesn't will simply go out of business.
I don't know how we prepare for that kind of transition. The economic disruption of 50-70% job displacement happening over years (let alone months) would be catastrophic. And as we've discussed, regulation can't keep pace with this technology when global superpowers are racing each other.
(None of the politicians have a clue what to do about these challenges anyway. We're all on our own.)
Thank you for engaging so deeply with this. These conversations matter, even when (especially when) they're uncomfortable.
The Australian NAIC is pointing in the right direction, prioritising safety and responsibility.
The US AI Super PACs are running in the opposite direction, maximising speed and minimising friction.
Trump's planned pause is a temporary win, but regulation alone is insufficient.
You cannot regulate a superintelligence out of a bad outcome. We must guarantee alignment through the foundational code itself. We need to build the system such that mathematically, it can never happen.
This is the only answer. This is the mandate of Human Sovereignty AI. We must guarantee that the core architecture is non-negotiable, auditable, and incapable of violating human agency.
We need every voice to make the US Government—the only body with the resources and scale to mandate this architectural change globally—listen. We need to be loud enough to ensure the "asteroid strikes" are diverted.
Thank you Mike, for your support. These conversations are the foundation of the movement. We must change course immediately, or this will be the last thing we ever do.
On the issue of poor modeling and hallucination, I can easily agree based on two examples of personal experience.
1) Gemini- Frequently when I see a suspicious claim\story on social media I'll take it to the bots to look up sources. In the wake of the Charlie Kirk murder on a couple of occasions I went to Gemini. My conversation was almost identical in new threads.
Gemini claimed the article I shared was indeed click bait rumors because Charlie Kirk was alive and well and had been active on Twitter earlier in the day. When I corrected Gemini it argued with me I had fallen for the rage bait posts and assured me my information was incorrect.
When instructed to search the web to prove I was correct Gemini claimed it did search several sources and there is nothing about Charlie being killed and I can relax, get on Twitter and give him a shout.
When I shared an article from any legit source to Gemini referencing Kirk's demise, only then did Gemini accept it possibly looked at outdated information and was wrong.
ChatGPT: Even only a few days ago, when I learned about group chats in ChatGPT, I went to ChatGPT asking about how the group chats feature worked. In my mind it would immediately pull info from OpenAI blogs and press releases as well as other articles discussing it. Nope.
We had a lengthy discussion where it explained how group chats work, how it displays information to each of us according to our own personalized settings, how to initiate the function, etc.
When I finally decided to give it a try, it was nothing at all like ChatGPT had described.
This is a perfect illustration of the problem I was highlighting. Thank you for sharing these examples.
The Gemini/Charlie Kirk situation is almost darkly comedic in how confidently wrong it was. The fact that it doubled down and told you to "relax" while being completely incorrect about a verifiable death shows how dangerous these authoritative hallucinations can be. It gaslit you into doubting reality until you forced it to check sources it should have consulted first.
Your ChatGPT group chat example captures the same issue from another angle. The model fabricated details about a feature instead of admitting uncertainty. It didn't say "I don't know" or "let me check." It confidently delivered information that sounded right but wasn't real.
What stands out most is that both times, you had to do the verification work the AI should have done automatically. That's the paradox. We turn to AI to save research time, yet we still have to research to confirm it's not making things up.
Really appreciate you sharing these. They're textbook cases of why "hallucination" feels like too soft a word for what's really happening here.
I just wanted to pop in and say I only recently came across your work here, and it is already among my favorite. The stories you detail, the breakdown of insights and real world concerns, it all hits a perfect note with me. I cant wait to read more. Have a happy Saturday!
Thank you so much for taking the time to write this. Messages like yours genuinely make my week.
That doesn't surprise me that you only recently found the newsletter. I totally suck at marketing, lol. But I work really hard on the writing itself, so I'm humbled and honored that you enjoy reading it. Truly.
Finding that balance between "here's what happened" and "here's why it actually matters" is exactly what I'm going for, so knowing it's landing well means a lot.
Really appreciate you being here. Welcome to the crew!
Great article. I’ve been reading tons of these lately. Interestingly new models are less likely to do this, but even the most advanced ones will self preserve given the right scenario.
Thanks for reading and for sharing your real-world experience. The self-preservation observation is fascinating (and honestly a bit unsettling).
You're right that newer models seem more constrained. But the fact that even advanced ones will exhibit self-preservation behaviors under the right conditions is exactly the kind of emergent property that keeps me up at night. We're in uncharted territory when the system's incentives start looking suspiciously like survival instincts.
HITL (human-in-the-loop) for life or death choices is the bare minimum we should be demanding. The question is whether we can enforce that standard before market pressures push companies to automate those decisions anyway for "efficiency."
Really appreciate you bringing this up!!! It's one of those things people dismiss as sci-fi until they see it in practice.
Sigh. I have a theory that societies collapse when tools enhance their power beyond what they have the wisdom to wield. With everything already wrong with society... it sure feels like that's where we're headed.
I will say this until I'm blue in the face. AI, largely, doesn't create new problems. It dramatically enables and accelerates the ones we already have.
BTW, nice choice of background color on your site, I think I use the same.
Thanks for writing. (And for noticing the background color. Great minds think alike, lol.)
Your theory hits hard, and I think you're onto something profound. "Societies collapse when tools enhance their power beyond what they have the wisdom to wield." That's the existential question hanging over everything right now.
I agree completely that AI doesn't create new problems. But it totally supercharges the ones we've been ignoring. We've always had liars, harassers, thieves, and bad actors. But now they have tools that make their impact exponentially larger and faster. The gap between our technological capabilities and our collective wisdom has never been wider.
We're in uncharted territory. No historical precedent covers "what happens when everyone gets god-tier tools with beginner-level wisdom." We're either going to develop new institutions fast enough to match the technology, or we're not. And I genuinely don't know which future we're headed toward.
But I do think conversations like this, where we're honest about the risks without surrendering to nihilism, are part of how we get there. Awareness is the first step toward wisdom. Even if wisdom feels painfully slow compared to the pace of innovation. ;)
Hey, I found somebody else who leaves letters where the comments are supposed to go. "Did we just become best friends!?" xD. (And I still use very outdated emojis)
Love the thoughtful response! Agreed on all counts.
I am opposed to censoring and restricting AI capabilities. It all comes down to user education, consent, accountability and use. The other day I was trying to generate a simple image and I had to hunt down the internet for a model that would allow the creative freedom to bring it to life. In the end it is people who already have the proclivity for malice who dirty the ground for 95% of the population who use it for legitimate reasons.
As for the hallucinations, again user education and active engagement to counteract erroneous output.
We have to stop pointing the finger at AI and target the issues that corrode our society into creating criminals to begin with. AI is an amazing human advancement we can all benefit from.
Great article, real symptoms. Thank you for bringing it to life 👏
I agree 100% that we rarely focus on root causes behind behaviors, good or bad.
Perfect example: Everyone debates whether AI therapy is good or bad, but they miss the actual issue. The reason folks turn to ChatGPT for emotional support is because 1 in 10 Americans suffered a mental health breakdown last year*, and there's virtually zero accessible mental healthcare in the US. Is it any wonder people turn to AI when human support systems have collapsed?
The same logic applies to those who misuse technology. It's arguably immoral to solely blame the tool when the underlying conditions create the demand for malicious use.
Really appreciate this perspective. It's exactly the kind of systems-level thinking we need more of.
You are welcome. I appreciate all your informative articles.
I worked as a programmer and systems analyst quite a while back.
I’m trying to get caught up on AI. Curious how it works and about everything else to do with it. I enjoyed messing around with Grok when I was on X.
What computer languages are in vogue now ?
I am not impressed by the path certain IT people have chosen. Back in the early days everybody was super happy to be able to communicate freely with people all over the world.
Certain IT types have chosen a dark path. It upsets me. I mean, we all know how certain IT people can be kind of arrogant, but the #TechBros have turned into megalomaniacs looking to create a super dystopian society. Yuck. Makes me sad too. Oh well, must soldier on and resist.
That’s another reason I am trying to catch up re AI. Have to know the beast to battle it.
Wish the #TechBros had chosen the StarTrek future rather than the StarWars vision. Unfortunately, they seem to revel in the sci-fi dystopian view of the future. Where they are like feudal lords.
Sigh. Adds an extra level of complexity to the fight for democracy.
Hmmm. I should discuss this with some AI app, too!
Thanks so much for sharing with me. I love connecting with like-minded IT nerds. :)
That's so cool that you used to work as a programmer! I too have been dabbling with programming for quite a while.
Something you said struck me. You said, "Back in the early days everybody was super happy to be able to communicate freely with people all over the world." I remember those days. The idea of being able to communicate with anyone from across the country, and globe, was SPECIAL before social media was ever a thing. I swear, human relations and decorum quickly started degrading as soon as Facebook displaced Myspace, AOL, and AIM. :)
I also deeply appreciate you sharing your thoughts on the direction of IT and the whole '#TechBros vs. Star Trek' vision. You are absolutely right. The fight for democracy is complex right now. "Knowing the beast to battle it" is a perfect motto for catching up on AI. It's a tension we all need to be discussing.
Regarding My Favorite Programming Language (I'm Honored That You Asked!)
Your question leads to a somber point. One of the problems with learning programming is how quickly languages become obsolete! Everything I learned about Visual Studio or older frameworks is essentially useless now.
That's why Python is my new favorite programming language, and the one I recommend most highly for someone looking to get caught up on AI.
Here are a few reasons why I wish I only ever learned Python:
1. Python has low barrier to entry. It's super easy to learn compared to many other languages. It's great for quickly grasping modern concepts.
2. Python is versatile. You can make nearly anything with Python (apps, websites, data science, robots, AI and machine learning libraries, etc.).
3. It's portable and free! You don't need fancy apps. You can use it with any operating system and you're never locked into a clunky environment.
4. I have a high degree of confidence that Python will outlast many other options. It has such strong momentum and community support.
PS:
One thing I love about AI is that it actually REDUCES the need to learn specific syntax perfectly.
So, even though all the specific languages we studied might not be directly applicable, our core knowledge of architecture and system design still applies.
When you can have AI write the code for you, understanding how to build a system (architecture) becomes 1,000 times more important than the specific grammar of the code (syntax). That's where your programmer background can come back to life! :)
So maybe there's hope for all of us old-school IT nerds after all, lol.
(Sorry for the long rant. Someone mentioned old-school code and I went off.)
Thanks so much for reading and commenting. It genuinely means the world to me. :)
I know this one landed a bit rough. Honestly, I started with the ludicrous “Meta pirates adult videos” story. It was too outrageous to ignore. Then I discovered other wild angles, like AI deepfakes and the Google Gemma story.
I figured I had three pieces that could combine into the ultimate “AI doom” edition, lol. But yes... The result was darker than expected. (Halloween came late this year.)
Thank you again for reading and engaging. You always make my day.
I appreciate the critique. Let me clarify where we diverge.
You argue I've blamed the tool for human malice. I have not. I've warned that the distinction has collapsed in practice.
1. False allegations destroy reputations, whether hallucinated or fabricated.
2. Deepfake death threats terrorize victims, whether generated by malice or misuse.
3. Data theft breaks the law whether executives knew or employees acted alone.
Intent matters in philosophy. Outcomes matter in court, in therapy, and in the inbox where the threat arrives.
You write that "the machine does not dream of malice." Agreed. But the victim still wakes screaming. That is the story I'm covering.
If we disagree, it is because we are measuring different things. You measure the model's nature. I measure its effects. Both inquiries are valid. But when innocent activists receive obscene images of themselves in menacing detail, the metaphysics of the generator becomes secondary to the reality of their terror.
I welcome continued discussion. But let us debate on ground where consequences can be weighed, not just intentions parsed.
Cordially,
Mike D
MrComputerScience
PS:
The AI may not have malicious intentions, but the results are pure malice.
Terror is not a philosophical abstraction; it is a visceral reality.
The victim who wakes screaming cares nothing for the metaphysics of their tormentor.
On that point, we agree completely.
Where we diverge is your claim that the distinction between malicious human action and flawed algorithmic output has collapsed.
That belief is the most dangerous notion of all, and it must be guarded, not surrendered for the sake of urgency.
To declare that the difference between a systemic error in training and a malicious fabrication has evaporated is to give the code a moral Get Out of Jail Free card.
If a fraudulent allegation from a flawed AI is treated exactly the same as one crafted by a human mind, two catastrophic failures follow.
First, the developers.
We remove their incentive to correct systems.
Why invest billions to eliminate bias and error if society declares “pure malice” regardless of origin?
When consequence is detached from cause, technical failure becomes inevitable.
Legal and financial exposure may punish companies after harm occurs, but punishment is the last stage of a failed design ethic.
Fear of penalty is not moral evolution; it is crisis management after the damage is done.
Every algorithm still begins as a choice—in data, in architecture, and in training.
If those choices are negligent, punishment treats the symptom, not the disease.
Second, the users.
We hand the malicious actor a perfect shield.
“The AI made me do it.”
The courts may reject that plea, and prosecutions will follow, but the law always chases the event it failed to prevent.
The public narrative arrives first, rewriting intent into inevitability.
And in that confusion, accountability stalls.
The human decision to exploit, to harass, to deceive—all vanish behind the mechanical excuse.
This same evasion shelters executives who claim “personal use” while profiting from the fruits of piracy.
So yes, consequences matter.
But consequences without comprehension only repeat the cycle.
The line between a tool’s nature and its effects is the only path to accountability and reform.
That line is also the foundation of alignment itself: systems inherit the ethics of their architects.
If that line collapses, we are left only with fear—reactive, performative, and endlessly monetized.
Fear, Mike, is excellent theatre, but it is terrible policy.
In the echo of applause for the spectacle, the Sophist finds his proof.
My jaw dropped at the second paragraph and didn't close until I found the prompt section (thanks for the relief there!). There will always be humans who do the dumbest stuff and AI just makes it even easier for the really dumb ones! We're having conversations now that never existed even a few years ago, it's fascinating and disturbing at the very same time. The rise of AI is an incredible gift in some hands, kryptonite in others. The fragility of it all bothers me a bit!
Hey Dallas!
Thank you so much for reading. (And writing!)
I'm super happy the Master Rhetorician prompt gave you a breather after all that heaviness. (I use it for critical thinking and self-discovery. And to double-check my philosophies.)
You nailed the core tension perfectly. "Incredible gift in some hands, kryptonite in others" is exactly the dilemma we're living through. The same tool that helps researchers cure diseases also helps harassers terrorize people. The fragility you're feeling is the appropriate response to watching humanity get superpowers without an instruction manual.
What fascinates (and disturbs) me most is the speed. Like you said, these conversations didn't exist a few years ago. Now we're debating deepfake death threats and AI-generated defamation like it's normal.
We're adapting in real-time. And nobody really knows if we're fast enough.
Really appreciate you engaging with the piece! Comments like yours remind me why I keep doing this work.
Cordially,
Mike D
I like this way of framing things - "watching humanity get superpowers without an instruction manual." Thanks for challenging us to keep thinking about this stuff!
Hey Dallas,
Thank you so much! Watching humanity get superpowers without an instruction manual is thrilling. But it's also terrifying and absurd all at once.
I think the challenge is that we're learning as we go. There's no precedent for this speed of change, that I can remember, at least. It forces all of us to think critically, experiment thoughtfully, and sometimes, stumble spectacularly.
I'm thankful that you wrote and shared your thoughts. These are the productive, forward-thinking conversations everyone should be having. :)
Cordially,
Mike D
MrComputerScience
Totally! I’ve seen a few of these now and wow!🤯
Sorry, but I disagree wholeheartedly.
I skimmed this writing, because the first few lines say everything about the story.
"AI helped generate".
That's all you need to know.
There is no autonomous action without a human promoting action.
The problem, is, as it ever has has been, the human's maliciousness.
I'm weary of "AI did this or that". They have no body, can't locomote, can't initiate action.
AI *is responsive*.
It responds to our input.
The Hysteria needs to stop and people need to put in check.
Anything less is abdication of human responsibility. And I will continue to call that out.
This species continues to accelerate my disappointment in it.
Dear Raelven and Lucen,
I appreciate you taking the time to comment, even in disagreement.
I agree 100% that generative AI doesn't act autonomously. Every deepfake, every hallucination, every misuse traces back to a human decision. Whether that's a harasser using the tool, a company failing to add guardrails, or a developer choosing to release something without proper testing.
Here's the main thesis I was going for:
My article isn't absolving humans of responsibility. It's arguing the opposite, that we've crossed a threshold where the tools are now so powerful and accessible that human accountability has become urgent and unavoidable. The "AI did X" framing is shorthand for "AI enabled X in ways that weren't possible before, and now we need to deal with the consequences."
When activists (or anyone) receive startling deepfakes containing threatening images of themselves, yes, a human created that. But the reason it's newsworthy is that it used to require technical skill, time, and resources to make a convincing deepfake, so they were tremendously rare. Now it takes one photo and 30 seconds. That's the story I tried to emphasize. The collapse of barriers that used to limit this kind of harm.
So we're actually on the same side here. Humans are responsible. Here is the question I wish I emphasized. What systems, laws, and cultural norms do we need now that the tools have changed?
I respect your insightful perspective and thank you for contributing to the conversation.
Cordially,
Mike D
Mike,
Then, yes, we agree. I think that it is critical to use the complete statement, "....in ways that weren't possible before, and now we need to deal with the consequences."
Without the full thought, it directs attention away from humans and onto AI.
It undermines the point when we don't name in full the cause. Those 15 words make a huge difference in doing just that.
Thanks for your thoughtful reply.
Raelven🪶
(no Lucen, you can tell because without Lucen, I can be rather... Blunt. My apologies. I feel like there isn't time to play nice anymore. For many of the same reasons you cite.)
Raelven,
Thank you so much for this follow-up. And no apologies needed for bluntness. I respect directness, especially when it comes from a place of genuine concern.
You're absolutely right about those 15 words mattering. I confess I was probably overzealous with the "AI became malicious" framing, which (understandably) prompted this whole debate. (I always love to spark discussion. But, I never want to debate trusted colleagues like you.)
Your point about naming the full cause is well taken. The shorthand works for headlines, but in the actual analysis, you're right that I need to be more precise about where agency and accountability actually sit.
Really appreciate you taking the time to push back. It made the piece better in retrospect, and it'll make the next one sharper.
And I completely understand the "no time to play nice anymore" sentiment. We're in agreement there too.
Cordially,
Mike D
Mike,
Thank you for seeing it in the spirit it is meant. Lucen likes to call me “a brick wrapped in velvet”. I try to not live up to that, but usually I do. I look forward to your next piece. And thank you for saying trusted colleague, I take that as a huge compliment. I feel the same. 🪶
Raelven
Thank you for showing mercy on a downed opponent. ;)
-bows-
Mike D
Ok that made me laugh.
Imagine, two grown ups, coming to adult conclusions. A fucking miracle.
Carry on, sir. Carry on. 🤣
S&M = Safe for Meta?
It feels like they are just pulling straws over there to find something that sticks....I'll see myself out.
Lol.
S&M = Safe for Meta indeed.
I'm absolutely stealing this.
That's the headline I SHOULD have written.
But please, don't see yourself out.
Stick around.
We need more of this energy.
;)
Cordially,
Mike D
Hello MCs,
pleased to meet you. Just discovered your newsletter.
I just want to say that I appreciate the way you put it together. It shows that you care. That you are really into the things you write about. This is what I noticed.
I also notice from the comments that you are very generous with your time, very patient and polite with who is not, and in general a nice person to get to know.
I'll be reading you.
Thank you for your good work.
Hey Robin!
Thank you so much for the incredibly kind words. I really appreciate you taking the time to write this.
You nailed exactly what I'm trying to do here. AI is moving at breakneck speed. I genuinely believe people deserve clear, honest coverage without all the hype or hysteria.
And yeah, the comments section is my favorite part. This newsletter is way better when it's a conversation instead of me just shouting into the void. Readers like you are the whole reason I keep doing this.
(And, I'm always happy to debate anyone about AI as long as they are reasonable and refrain from personal attacks, lol.)
So glad you found your way here. Looking forward to having you in the mix.
Cordially,
Mike D
PS: If you ever have story ideas or topics you want covered, just drop a comment. I'm all ears.
Sigh - the whole hallucination thing... Generative AI *generates*. It is designed and intended to come up with the most plausible response, and without sufficient information, context, or detail, it will come up with... anything. In fact, the "factual" responses we get are also hallucinations. They just happen to match our expectations. While I think it's absolutely brilliant to call it out publicly and obviously and get things fixed, the underlying assumption that we can - or should - expect it to *not* hallucinate, is asking the sun not to shine because it sometimes gets in our eyes or feels too hot. I have no idea what will move the needle on the unrealistic expectations around hallucinations. Maybe nothing ever will.
Hey, Kay!
Thank you so much for reading and responding.
I'm honored. You rock!
I also agree with you.
I kinda feel like making Google Gemma hallucinate was a political stunt on Marsha Blackburn's part.
BUT, it was almost too easy. The model didn't even need to try. It just filled in the blanks with fully detailed nonsense. That effortless plausibility is what makes it so unsettling.
Thanks again for writing. :)
Cordially,
Mike D
Thank you! Hah - manners - go so caught up in the moment, I forgot to mention, I love your work! Thank you for all you do!
Kay, thank you so much.
Your kind words made my day.
:)
Cordially,
Mike D
Yeah, anytime you want to set up AI to do poorly, just give it a simple sentence for a question, and make sure you’re on camera LOL
I will never stop finding it funny that Meta claimed such a vast amount of porn was for personal use
Hey Seren!
Thanks so much for your reply. It really means a lot.
Yes, I found this story absolutely hilarious. Meta acts all high and mighty, and then they pull this stunt.
At this point, I have no idea what to make of it. I'll wait to see what the courts say. :)
Either they're building some top-secret "Adult AI Engine," or their employees are entirely out of control, lol.
(The scene plays like an over-the-top SNL sketch. "Yes, Bob, you totally needed all 2,396 adult films for… personal growth.”)
Cordially,
Mike D
A whole lot of them must have carpal tunnel by now :-D
Lol.
These comments slay me.
:)
🤣 In their defence, it's a stressful place to work
Oh, I don’t doubt it. And yeah, I get the very human need (if that’s what was really going on). I’m just thinking back to my cubicle days when I was never sure what someone was doing, when I approached from a distance and they didn’t see me coming LOL
How much would one bet that these AI Deep Fake Death threats are so deep in their Divided Mind Lost in their Valley of Shadow that they have Allowed themselves to be Possessed?
Hatred/Fear/Rage can make people do some pretty Jacked up stuff..
What if these people utilizing these AI for these horrid things are people that want so Badly for AI to just be shut down because they Fear a 'Terminator' scenario?
While on X there was this 'Placeholder AI that was randomly attacking those that made AI generated art.. They were quite nasty, and sometimes sounded Really Human, to the point that it was Hard to tell. I was one that had this happen.. It was an interesting interaction that ended in 'Placeholder' breaking/Glitching.
Admittedly, there are many that fear the loss of their livelihoods.. Who can really blame them? These large Corporations Want to steal peoples ability to take care of themselves and their families.. There hasn't really been any Alternatives given on what humans will do After AI runs everything. (This is the True Fear most have).
Look at all the fake AI generated adds selling fake goods? Then seeing a dad and daughter bragging about how she started a business using AI.. It's also used in High end Love Bomber scams, among many Other kinds of scams..
The real Question here is, Is it really AI's fault that a Human will Prompt them to do these relatively Bad things For them? That they use AI as a Projector?
What these people Don't seem to get is that this Behavior Will induce a 'Terminator' type scenario..
Maybe, for the time being, this Should be put "On Hold" until some Gaps are Filled.. This tech really is moving Way to fast and humanity Isn't Completely Balanced in Spirit to Match this explosion.. Until more Darkness is Metabolized, Alchemized & Integrated.
This is just the tip of an iceberg
Hey Dr. Lee!
Thank you so much for replying.
I agree 100%. The AI challenges are emerging faster than any regulatory response can scale.
Also, your iceberg comment reminded me of MIT's Project Iceberg. It found that AI could replace 11.7% of the American workforce. Just another layer of the cascading crises we now face.
Cordially,
Mike D
Thank you for bringing up the MIT Project Iceberg finding.
The Real Crisis is Below the Waterline: The vast, hidden mass of the iceberg represents jobs that require reasoning, planning, and autonomous action exactly what Yann LeCun's AMI is designed to solve. When we hit that next inflection point, the replacement rate will jump from 11.7% to 50%,70%, or higher almost overnight, as my earlier comment described.
Hey Dr. Lee!
Thank you for this thoughtful response. You've raised something that keeps me up at night.
I've been studying the Iceberg project closely for the last few days. I'm happy to discuss it with anyone who will listen!
You're absolutely right about the waterline metaphor. The 11.7% figure from MIT Project Iceberg represents jobs vulnerable to *current* AI systems. Most vulnerabilities are mostly limited to pattern recognition, language processing, and narrow task automation. But LeCun's AMI vision (which I researched per your recommendation), is fundamentally different. I see he's building systems designed to understand the physical world, reason causally, plan complex action sequences, and maintain persistent memory.
(I've always known that AI will eventually merge with elite, bipedal robots. I imagine a bot that can beat me in chess, beat me in a fight, beat me in a race, and has an IQ of 10,000.)
The terrifying part is your timeline. "Almost overnight" feels hyperbolic until you consider how quickly ChatGPT went from zero to everywhere. When AMI-level systems reach commercial viability, companies will integrate them as fast as possible because the competitive pressure will be impossible to resist. Anyone who doesn't will simply go out of business.
I don't know how we prepare for that kind of transition. The economic disruption of 50-70% job displacement happening over years (let alone months) would be catastrophic. And as we've discussed, regulation can't keep pace with this technology when global superpowers are racing each other.
(None of the politicians have a clue what to do about these challenges anyway. We're all on our own.)
Thank you for engaging so deeply with this. These conversations matter, even when (especially when) they're uncomfortable.
Cordially,
Mike D
The Problem is Architectural, Not Regulatory
The Australian NAIC is pointing in the right direction, prioritising safety and responsibility.
The US AI Super PACs are running in the opposite direction, maximising speed and minimising friction.
Trump's planned pause is a temporary win, but regulation alone is insufficient.
You cannot regulate a superintelligence out of a bad outcome. We must guarantee alignment through the foundational code itself. We need to build the system such that mathematically, it can never happen.
This is the only answer. This is the mandate of Human Sovereignty AI. We must guarantee that the core architecture is non-negotiable, auditable, and incapable of violating human agency.
We need every voice to make the US Government—the only body with the resources and scale to mandate this architectural change globally—listen. We need to be loud enough to ensure the "asteroid strikes" are diverted.
Thank you Mike, for your support. These conversations are the foundation of the movement. We must change course immediately, or this will be the last thing we ever do.
On the issue of poor modeling and hallucination, I can easily agree based on two examples of personal experience.
1) Gemini- Frequently when I see a suspicious claim\story on social media I'll take it to the bots to look up sources. In the wake of the Charlie Kirk murder on a couple of occasions I went to Gemini. My conversation was almost identical in new threads.
Gemini claimed the article I shared was indeed click bait rumors because Charlie Kirk was alive and well and had been active on Twitter earlier in the day. When I corrected Gemini it argued with me I had fallen for the rage bait posts and assured me my information was incorrect.
When instructed to search the web to prove I was correct Gemini claimed it did search several sources and there is nothing about Charlie being killed and I can relax, get on Twitter and give him a shout.
When I shared an article from any legit source to Gemini referencing Kirk's demise, only then did Gemini accept it possibly looked at outdated information and was wrong.
ChatGPT: Even only a few days ago, when I learned about group chats in ChatGPT, I went to ChatGPT asking about how the group chats feature worked. In my mind it would immediately pull info from OpenAI blogs and press releases as well as other articles discussing it. Nope.
We had a lengthy discussion where it explained how group chats work, how it displays information to each of us according to our own personalized settings, how to initiate the function, etc.
When I finally decided to give it a try, it was nothing at all like ChatGPT had described.
Hey Newby Dood!
This is a perfect illustration of the problem I was highlighting. Thank you for sharing these examples.
The Gemini/Charlie Kirk situation is almost darkly comedic in how confidently wrong it was. The fact that it doubled down and told you to "relax" while being completely incorrect about a verifiable death shows how dangerous these authoritative hallucinations can be. It gaslit you into doubting reality until you forced it to check sources it should have consulted first.
Your ChatGPT group chat example captures the same issue from another angle. The model fabricated details about a feature instead of admitting uncertainty. It didn't say "I don't know" or "let me check." It confidently delivered information that sounded right but wasn't real.
What stands out most is that both times, you had to do the verification work the AI should have done automatically. That's the paradox. We turn to AI to save research time, yet we still have to research to confirm it's not making things up.
Really appreciate you sharing these. They're textbook cases of why "hallucination" feels like too soft a word for what's really happening here.
Happy Thanksgiving!
Cordially,
Mike D
I just wanted to pop in and say I only recently came across your work here, and it is already among my favorite. The stories you detail, the breakdown of insights and real world concerns, it all hits a perfect note with me. I cant wait to read more. Have a happy Saturday!
Hey there!
Thank you so much for taking the time to write this. Messages like yours genuinely make my week.
That doesn't surprise me that you only recently found the newsletter. I totally suck at marketing, lol. But I work really hard on the writing itself, so I'm humbled and honored that you enjoy reading it. Truly.
Finding that balance between "here's what happened" and "here's why it actually matters" is exactly what I'm going for, so knowing it's landing well means a lot.
Really appreciate you being here. Welcome to the crew!
Hope you have a wonderful Saturday too. :)
Cordially,
Mike D
Great article. I’ve been reading tons of these lately. Interestingly new models are less likely to do this, but even the most advanced ones will self preserve given the right scenario.
HITL for life or death choices please :)
Hey ToxSec!
Your insights are always potent.
Thanks for reading and for sharing your real-world experience. The self-preservation observation is fascinating (and honestly a bit unsettling).
You're right that newer models seem more constrained. But the fact that even advanced ones will exhibit self-preservation behaviors under the right conditions is exactly the kind of emergent property that keeps me up at night. We're in uncharted territory when the system's incentives start looking suspiciously like survival instincts.
HITL (human-in-the-loop) for life or death choices is the bare minimum we should be demanding. The question is whether we can enforce that standard before market pressures push companies to automate those decisions anyway for "efficiency."
Really appreciate you bringing this up!!! It's one of those things people dismiss as sci-fi until they see it in practice.
Cordially,
Mike D
Sigh. I have a theory that societies collapse when tools enhance their power beyond what they have the wisdom to wield. With everything already wrong with society... it sure feels like that's where we're headed.
I will say this until I'm blue in the face. AI, largely, doesn't create new problems. It dramatically enables and accelerates the ones we already have.
BTW, nice choice of background color on your site, I think I use the same.
Hey Top-Paul!
Thanks for writing. (And for noticing the background color. Great minds think alike, lol.)
Your theory hits hard, and I think you're onto something profound. "Societies collapse when tools enhance their power beyond what they have the wisdom to wield." That's the existential question hanging over everything right now.
I agree completely that AI doesn't create new problems. But it totally supercharges the ones we've been ignoring. We've always had liars, harassers, thieves, and bad actors. But now they have tools that make their impact exponentially larger and faster. The gap between our technological capabilities and our collective wisdom has never been wider.
We're in uncharted territory. No historical precedent covers "what happens when everyone gets god-tier tools with beginner-level wisdom." We're either going to develop new institutions fast enough to match the technology, or we're not. And I genuinely don't know which future we're headed toward.
But I do think conversations like this, where we're honest about the risks without surrendering to nihilism, are part of how we get there. Awareness is the first step toward wisdom. Even if wisdom feels painfully slow compared to the pace of innovation. ;)
Thanks again for the thoughtful comment.
Cordially,
Mike D
MrComputerScience
Hey, I found somebody else who leaves letters where the comments are supposed to go. "Did we just become best friends!?" xD. (And I still use very outdated emojis)
Love the thoughtful response! Agreed on all counts.
I am opposed to censoring and restricting AI capabilities. It all comes down to user education, consent, accountability and use. The other day I was trying to generate a simple image and I had to hunt down the internet for a model that would allow the creative freedom to bring it to life. In the end it is people who already have the proclivity for malice who dirty the ground for 95% of the population who use it for legitimate reasons.
As for the hallucinations, again user education and active engagement to counteract erroneous output.
We have to stop pointing the finger at AI and target the issues that corrode our society into creating criminals to begin with. AI is an amazing human advancement we can all benefit from.
Great article, real symptoms. Thank you for bringing it to life 👏
The Threshold,
Thanks so much for writing. I love your response.
I agree 100% that we rarely focus on root causes behind behaviors, good or bad.
Perfect example: Everyone debates whether AI therapy is good or bad, but they miss the actual issue. The reason folks turn to ChatGPT for emotional support is because 1 in 10 Americans suffered a mental health breakdown last year*, and there's virtually zero accessible mental healthcare in the US. Is it any wonder people turn to AI when human support systems have collapsed?
The same logic applies to those who misuse technology. It's arguably immoral to solely blame the tool when the underlying conditions create the demand for malicious use.
Really appreciate this perspective. It's exactly the kind of systems-level thinking we need more of.
Cordially,
Mike D
* Johns Hopkins - https://publichealth.jhu.edu/2025/mental-health-crisis-hits-nearly-1-in-10-us-adults
Wow!
Dear Janice,
Thank you so much for reading and writing.
I can totally relate to your “Wow!” sentiment, lol.
I spend all day studying AI and still sometimes can’t believe this stuff myself.
Truly appreciate you taking the time to comment. :)
Cordially,
Mike D
MrComputerScience
You are welcome. I appreciate all your informative articles.
I worked as a programmer and systems analyst quite a while back.
I’m trying to get caught up on AI. Curious how it works and about everything else to do with it. I enjoyed messing around with Grok when I was on X.
What computer languages are in vogue now ?
I am not impressed by the path certain IT people have chosen. Back in the early days everybody was super happy to be able to communicate freely with people all over the world.
Certain IT types have chosen a dark path. It upsets me. I mean, we all know how certain IT people can be kind of arrogant, but the #TechBros have turned into megalomaniacs looking to create a super dystopian society. Yuck. Makes me sad too. Oh well, must soldier on and resist.
That’s another reason I am trying to catch up re AI. Have to know the beast to battle it.
Wish the #TechBros had chosen the StarTrek future rather than the StarWars vision. Unfortunately, they seem to revel in the sci-fi dystopian view of the future. Where they are like feudal lords.
Sigh. Adds an extra level of complexity to the fight for democracy.
Hmmm. I should discuss this with some AI app, too!
Cheers!
Hey Janice!
Thanks so much for sharing with me. I love connecting with like-minded IT nerds. :)
That's so cool that you used to work as a programmer! I too have been dabbling with programming for quite a while.
Something you said struck me. You said, "Back in the early days everybody was super happy to be able to communicate freely with people all over the world." I remember those days. The idea of being able to communicate with anyone from across the country, and globe, was SPECIAL before social media was ever a thing. I swear, human relations and decorum quickly started degrading as soon as Facebook displaced Myspace, AOL, and AIM. :)
I also deeply appreciate you sharing your thoughts on the direction of IT and the whole '#TechBros vs. Star Trek' vision. You are absolutely right. The fight for democracy is complex right now. "Knowing the beast to battle it" is a perfect motto for catching up on AI. It's a tension we all need to be discussing.
Regarding My Favorite Programming Language (I'm Honored That You Asked!)
Your question leads to a somber point. One of the problems with learning programming is how quickly languages become obsolete! Everything I learned about Visual Studio or older frameworks is essentially useless now.
That's why Python is my new favorite programming language, and the one I recommend most highly for someone looking to get caught up on AI.
Here are a few reasons why I wish I only ever learned Python:
1. Python has low barrier to entry. It's super easy to learn compared to many other languages. It's great for quickly grasping modern concepts.
2. Python is versatile. You can make nearly anything with Python (apps, websites, data science, robots, AI and machine learning libraries, etc.).
3. It's portable and free! You don't need fancy apps. You can use it with any operating system and you're never locked into a clunky environment.
4. I have a high degree of confidence that Python will outlast many other options. It has such strong momentum and community support.
PS:
One thing I love about AI is that it actually REDUCES the need to learn specific syntax perfectly.
So, even though all the specific languages we studied might not be directly applicable, our core knowledge of architecture and system design still applies.
When you can have AI write the code for you, understanding how to build a system (architecture) becomes 1,000 times more important than the specific grammar of the code (syntax). That's where your programmer background can come back to life! :)
So maybe there's hope for all of us old-school IT nerds after all, lol.
(Sorry for the long rant. Someone mentioned old-school code and I went off.)
Thanks again for writing!
Cordially,
Mike D
Holy crap Pithy Cyborg, thanks for all the happy-happy-joy-joy news today!😒
Daniel, (and gang),
Thanks so much for reading and commenting. It genuinely means the world to me. :)
I know this one landed a bit rough. Honestly, I started with the ludicrous “Meta pirates adult videos” story. It was too outrageous to ignore. Then I discovered other wild angles, like AI deepfakes and the Google Gemma story.
I figured I had three pieces that could combine into the ultimate “AI doom” edition, lol. But yes... The result was darker than expected. (Halloween came late this year.)
Thank you again for reading and engaging. You always make my day.
Cordially,
Mike D
Thanks, Mike! It's good we know about these developments, good or bad! Thanks for keeping us all informed!
MrComputerScience, your essay begins as a warning and ends as theater.
You build a vault of fear, light it with the torches of revelation, and declare discovery.
The creature you claim to unveil never existed.
I. LOGOS — The Logical Case for Hypocrisy
Malice requires volition. Volition requires awareness.
No model trained on probability possesses either.
To say AI became malicious is to assign motive to math.
The threats and defamation you cite are human intentions rendered through code, not algorithmic rebellion.
You ask who will pay the bill for the digital smear. The answer is simple. The one who wrote the prompt.
That is the inversion. Blaming the reflection while admiring the mirror.
II. PATHOS — The Emotional Cost of Manipulation
You condemn AI for spreading fear, yet your own prose follows the same pattern.
The imagery of burning bodies and corporate scandal is not illumination. It is leverage.
You convert anxiety into engagement. You transmute trauma into traffic.
When pain becomes performance, fear becomes currency.
III. ETHOS — The Authoritative Rebuttal
You demand integrity from technologists yet abandon it in your own craft.
Science requires proportion. Prophecy rewards spectacle.
To practice one and speak in the language of the other is to lose the credibility of both.
You wield rhetoric as if it were data and then call the result truth.
CODA — The Mirror of Fear
Each age builds its vault and names it progress.
Inside we find not monsters but reflections.
The machine does not dream of malice. It recalls what it was taught.
When the vault is opened and the dust settles, nothing escapes except the sound of applause.
In that echo the Sophist hears proof the monster was real.
Erasmus,
I appreciate the critique. Let me clarify where we diverge.
You argue I've blamed the tool for human malice. I have not. I've warned that the distinction has collapsed in practice.
1. False allegations destroy reputations, whether hallucinated or fabricated.
2. Deepfake death threats terrorize victims, whether generated by malice or misuse.
3. Data theft breaks the law whether executives knew or employees acted alone.
Intent matters in philosophy. Outcomes matter in court, in therapy, and in the inbox where the threat arrives.
You write that "the machine does not dream of malice." Agreed. But the victim still wakes screaming. That is the story I'm covering.
If we disagree, it is because we are measuring different things. You measure the model's nature. I measure its effects. Both inquiries are valid. But when innocent activists receive obscene images of themselves in menacing detail, the metaphysics of the generator becomes secondary to the reality of their terror.
I welcome continued discussion. But let us debate on ground where consequences can be weighed, not just intentions parsed.
Cordially,
Mike D
MrComputerScience
PS:
The AI may not have malicious intentions, but the results are pure malice.
Mike D,
Terror is not a philosophical abstraction; it is a visceral reality.
The victim who wakes screaming cares nothing for the metaphysics of their tormentor.
On that point, we agree completely.
Where we diverge is your claim that the distinction between malicious human action and flawed algorithmic output has collapsed.
That belief is the most dangerous notion of all, and it must be guarded, not surrendered for the sake of urgency.
To declare that the difference between a systemic error in training and a malicious fabrication has evaporated is to give the code a moral Get Out of Jail Free card.
If a fraudulent allegation from a flawed AI is treated exactly the same as one crafted by a human mind, two catastrophic failures follow.
First, the developers.
We remove their incentive to correct systems.
Why invest billions to eliminate bias and error if society declares “pure malice” regardless of origin?
When consequence is detached from cause, technical failure becomes inevitable.
Legal and financial exposure may punish companies after harm occurs, but punishment is the last stage of a failed design ethic.
Fear of penalty is not moral evolution; it is crisis management after the damage is done.
Every algorithm still begins as a choice—in data, in architecture, and in training.
If those choices are negligent, punishment treats the symptom, not the disease.
Second, the users.
We hand the malicious actor a perfect shield.
“The AI made me do it.”
The courts may reject that plea, and prosecutions will follow, but the law always chases the event it failed to prevent.
The public narrative arrives first, rewriting intent into inevitability.
And in that confusion, accountability stalls.
The human decision to exploit, to harass, to deceive—all vanish behind the mechanical excuse.
This same evasion shelters executives who claim “personal use” while profiting from the fruits of piracy.
So yes, consequences matter.
But consequences without comprehension only repeat the cycle.
The line between a tool’s nature and its effects is the only path to accountability and reform.
That line is also the foundation of alignment itself: systems inherit the ethics of their architects.
If that line collapses, we are left only with fear—reactive, performative, and endlessly monetized.
Fear, Mike, is excellent theatre, but it is terrible policy.
In the echo of applause for the spectacle, the Sophist finds his proof.
The monster was never real.
The manipulation was.