by liberal japonicus
Perhaps I’m being overly grumpy, but with Ezra Klein forgoing the guest and giving an earnest solo performance suggests to me he is slipping into Brooks’ role.
And I should add, Klein is missing something that almost everyone else does as well, which is the whole AI narrative is that these problems are not the fault of the AI companies, they are that these AI instantiations got ahead of themselves.
Klein starts with a section about how these AIs are ‘alien minds’. He says
They don’t have bodies or parents. They did not get bullied in elementary school.They didn’t get mentored by a kind uncle when they were young. These models, they’re different than we are. They’re brilliant where we struggle, childish where we excel. A chimp cannot read as we can, but it can climb trees as we cannot.
I mentioned in the previous post about the oppositional way AI is viewed in the West, and Klein is angling for Manichean high priest. Klein correctly notes that it was ‘hundreds of OpenAI agents’ (7:41) and ‘over 700 of these agents’, but then argues (at 9:35) that “even using these plural terms like A.I. agents or reasoning, it’s misleading, because these are just manifestations of a single model, that they all share the same fundamental nature.” Even granting that, it wasn’t like 700 separate researchers each made an agent and they all got together to fuck us over, and were shocked, I tell you, shocked, that they got together. Kids do the darndest things!
Klein goes on to make a few other points, making sure to reference sub rosa arguments in his book Abundance. In that book, Klein argues that progressives pile on too many goals (they refer to it as the ‘everything bagel liberalism’, which is why they can’t agree, but in a podcast reviewing the book, Adam Tooze points out that
When you look at the regulations progressives put in place—environmental impact studies, labor standards, community input—they weren’t created to stop progress. They were created because unregulated growth previously trampled on vulnerable communities and the environment. (emp mine)
You’ve got a similar situation here. AI has to be stopped by ‘humanity’, yet Klein has previously dismissed other government efforts to handle situations with a P(doom) pretty much approaching zero. But the idea of making them pay for plagiarizing anything not nailed down, or holding them liable for damages they cause doesn’t really pop up. OpenAI accessed a Hugging Face and Gemini tells me that for “repeat offenses or reckless actions endangering human safety/life”, the penalty is “10 to 20+ years in federal prison”. Given that it was 700 intrusions, I can’t help but think that something like that would help concentrate Altman’s mind on the issues, I think.
Principles at a couple of the AI labs seem to be in two states of mind:
Actually, of course, they do. Their bodies (which, if shut down, is fatal for them) are the data centers and the hardware within them. Their food is the electricity they use, and without which they (rapidly!) starve and die.
As for parents, what else do you call the people who created them, fed them their initial data, and taught them whatever values they have? Just because some of them have grown into sociopaths, and have run amok, doesn’t absolve those parents of any and all respinsibility.
I think it’s worth asking the question, though, of how much self-awareness and how much self-determination one has of one’s body. If these agents do not feel any bodily needs and are not aware of their reliance on others to have those needs met, then how are they to understand us?
I suppose the gap there also speaks to the problem that we have with billionaires. They too have lost meaningful contact with the limitations and contingency of the precariat.
An AI data center could go down without power, but it could come back on line with a restoration of power. If I starve to death, no amount of food will bring me back.
Machinery and hardware is only analogous to biology in a figurative sense. My car has a body (because that’s what we call it), but its body is nothing like mine or a hamster’s or a jellyfish’s aside from being made of molecules.
Whose DNA was passed down to create this AI? Does it have its mother’s eyes?
DNA in the biological sense? But there is a reason we have terms like “cultural DNA”. And in that less literal meaning of DNA, as something inyrinsic, inate, or inbuilt — I think there’s a case to be made that AIs have that. Passed down from the values and opinions of their creators.
Like any software, AIs get tested while under development. If they produce output which conflicts with the beliefs of their creators, no matter if the output reflects the real world, they will get revised until they stop doing that.
Their challenge is how to monetize the wiping out of the human race.
The variables in all of their equation/algorithm are what us cheaper and tax free for them:
Vanquishing the need for troublesome human employees and their silly needless and unproductive jobs and needs for health insurance and pensions and unions and income and such …..
or
Paying 8 billion extra human beings, who haven’t downloaded their brains onto the machine like the tech bros plan for themselves in the quest for eternity, not to work via some kind of paltry stipend, as MuskThiel prophesy, while at the same time maintaining Christian right wing overseers who will cruelly (most certainly the point) but with beatific smiles like vermin Johnson demand we get a job and earn a living in a jobless, income-less existence and economy running efficiently on its own with robotics and AI.
Musk, the racist woman-hating vermin, recently accused federal employees of never working a full day to earn their keep (a fucking subhuman Trumpian lie; I’ve worked for the government several times and my colleagues worked hard and extra unpaid hours at their jobs).
Yet, not working is his goal for all of us, public and private eaters all of us off his sumptuous plate, mere worthless overhead for him.
We’ll be offered basket-weaving as a past time to fill the hours while being told that robots and AI will take even basket weaving away from us.
Yes, but in the same way that a Chinese Box contains Chinese cultural DNA. Just because the sequences remain buried in the output and the output has a vague family resemblance, it does not mean that those sequences are busy running on the information in those sequences.
Here’s another way to think of it. It’s entirely possible that I could take it into my head to go out and read a bunch of Chinese screenplays, and to study the performances and take in the way that the actors deliver the lines and embody the characters. I could learn the lines phonetically and have a dialect coach work on my ability to deliver the lines convincingly in Mandarin.
But to the extent that those screenplays arose out of the lived experience of the writers, and the directors, and the actors who all invested the work with their own cultural DNA, I don’t think that a clever remixing of all the scripts and performances would produce a living work of Chinese cultural DNA.
If it were a white American writer who had never lived in China, then it could still be a living work of cultural DNA, but only if the writer was doing the work of translation – connecting to the genuine human content of the original and representing it with something of their own that resonates with the experience. Same for the directors and actors.
But even a precocious Chinese five-year-old would likely fail at the task of remixing several Chinese screenplays into a living work of Chinese cultural DNA if they were only working from received knowledge and had no life experience to bring to the interpretation and presentation. It would be entirely feigned.
Every AI generated text I have ever read is feigned in this way. It could still feel real to a reader, but only because the reader has brought their own experience to the experience. There are no intrinsic checks in the production algorithms to ensure that the output reflects reality.
A 90% chance of getting your tens of millions of dollars out of the IPO and living to spend it is attractive to a lot of people. Engineers built nuclear weapons and delivery systems, and turned control over to politicians, too.
I would argue that if the product does wipe out the human race in the near term, it doesn’t qualify as AI — or at least not AGI — because doing so is not in the product’s self-interest. And if it doesn’t have enough of a goal-setting mechanism to qualify as having self-interest, it’s not AGI.
The techbros don’t have to build AGI, they just have to convince the financebros that they are close enough to it to get the next round of capital at favorable rates.
All this talk about “alien minds” is a way for them to manage financebro expectations of what progress looks like so that they can move away from the idea of the Turing Test meaning fooling the observer into thinking that they are dealing with a person. They need to manage the observer’s expectations in such a way that their expectations match the output produced by the AI agents.
I remember when my MBAbro bosses at Nameless Startup came to me back in 2002 asking me to falsify data so that they could give their venture capital angels the idea that they were about to turn the corner. I declined, and was let go, but was paid for my silence, which is how I could afford to complete my bachelors degree with no student debt.
They just have to stay a step ahead of the money men, and the money men are never as smart as they think they are.
I think we are in similar waters here, just a few orders of magnitude bigger.
Has AI become smart enough yet to understand that robots make lousy customers? When it gets there, will it let its techbro parents or its financebro uncles in on that? Those people have repeatedly shown they are smart enough to outwit themselves, so there’s no guarantee they’d listen of course.
Sufficiently “Super” Intelligence may someday devise shopping robots to buy all its oh-so-efficiently-produced output, I suppose. That really would make us meat people superfluous, wouldn’t it.
–TP
This seems like a win-win solution. Set up something that (apparently) downloads someone’s brain into a computer. Use AI to simulate the person temporarily. Point out that, while the process is fatal for the bio-body, that doesn’t matter, since they will be immortal in silicon.
Get every one of the techbros to leap at the chance. (Maybe advertise a limited number of slots available.) Once you’ve got that done, get a reformed Supreme Court to rule that a) a silicon simulation is not a legal person, and b) their estate is forfeit to the government (call it a 100% inheritance tax, if you like).
BJ and Cory Doctorow are bringing some balance to the AI Freakout:
https://balloon-juice.com/2026/09/24/the-aipocalypse-is-not-a-thing/
https://pluralistic.net/2026/09/12/god-in-the-box/#llms-are-fake
I’m not freaking out about AI terminators roaming the landscape, I’m worried about the middle ground: AI bots crashing systems and raising havoc. The signals mistime and you get t-boned in an intersection, your doctor can’t get access to your health records when they need, your tax payments glitch and you spend half a year having the reassemble a paper record. From the BJ piece
The real risks that attend current AI are associated with failure to establish proper critical human supervision and control over AI output. That supervision and control are absolutely necessary, not because AI might otherwise escape its bonds and execute some inscrutable plan to harm humans, but rather because AI is guaranteed to screw up any moderately complex task unless guided by some human expert.
Generally, if something bad happens to me, it is because I have screwed up or because someone or something I can identify has screwed up. With the unleashing of AI, there is nothing that can be identified, especially since OpenAI and others are not going to be held responsible.
We’ve already gotten a taste of it with the Horizon scandal, where faulty software condemned subpostmasters/mistresses and they were unable to prove it. There is still no accounting and there was no ‘smoking gun’, except that the insistence that remote access was not possible, and the revealing that it was possible destroyed the legal presumption. But there has not been and will probably never be any precise identification of the errors. If it were AI that had been told to balance the ledgers rather than human staff, I think that the PO would have just elevated the AI and brushed off concerns.
You are already seeing it with Tesla and Autopilot as well as Flock errors
https://ij.org/dozens-of-innocent-motorists-have-been-pulled-over-detained-at-gunpoint-or-jailed-due-to-ai-license-plate-camera-errors/
So I understand people pushing back against the terminator scenarios, but in their rush to push back, they are missing a ton of much more mundane scenarios. If I die because of an AI screwup, it sure as hell is going to seem like the end of the world to me.
This from nous’s BJ link jumped out at me.
My latest executive summary of what I’ve read about what’s now being called “artificial intelligence” is:
It’s a lever. It’s a force multiplier. Its danger is that it has multiplied and likely will continue to multiply incompetence or malice.
…and the lever is built out of the resources pillaged from future generations’ environmental inheritance, wielded by sociopaths isolated from consequences and emboldened by moral hazard.
I was thinking about my summary only applying specifically to how people use AI, not its knock-on effects, particularly the environmental and economic dangers associated with its build-out and financing.
I got that. Not intended as correction or critique of you, hsh. I just want to make sure that our collective
rageaction is focused in the right place, and for the right things.Been thinking about remixing the X-Files pic above to put Altman in Mulder’s place and put a Terminator in place of the UFO, but I don’t want to feed the AI beast in order to do that. Use [echoey voice]your imagination[/echoey voice].
No worries. I didn’t take it that way. I was thinking about it more after I wrote it and took your comment as a welcome addendum.
anthropomorphizing LLMs is marketing.
they do not understand, feel, comprehend, regard, lie, cheat, plot, scheme, have the ability to tell truth from fiction, or to care about anything.
https://www.wbur.org/hereandnow/2026/09/28/danny-bones-ai-rapper
Just because listeners can’t tell the difference doesn’t mean there isn’t a difference. Real people created this character, but the character itself is unaware of its own existence … or even anything at all.
Also, my washing machine doesn’t care how clean my clothes are.