Yet when it comes to AI, I have found the evolution of this carefully staged exercise in scaremongering different in some respects. For a start, this is clearly a top-down campaign, orchestrated by industry insiders. Indeed, as far back as 12 years ago, those in the know began to raise the alarm about AI. In 2014, Elon Musk told his followers on Twitter that AI is potentially ‘more dangerous than nukes’. Bill Gates informed millions of his followers on Reddit that ‘I agree with Elon Musk and some others on this and don’t understand why some people are not concerned’. Needless to say, these fears did not stop either of them from pursuing the development of AI.
|
|||||
Samizdata quote of the day – Don’t succumb to the great AI panicSeptember 19th, 2026 |
 5 comments to Samizdata quote of the day – Don’t succumb to the great AI panicLeave a Reply |
   |
||||
All content on this website (including text, photographs, audio files, and any other original works), unless otherwise noted, is licensed under a Creative Commons License. |
|||||
Listen to what the scaremongers say, but also watch what they do.
Have the CCP slowed their development of AI? I doubt it.
Have the tech billionaires who undoubtedly have significant voting shares in large AI companies, used that power to slow things down? I doubt it–that would cost them money which is even scarier to them than the end of humanity.
Here is an analogy to consider. Imagine it is 1950, we all recognize that the atomic bomb could wipe out humanity, and we were frightened with the Soviet Union detonated its first bomb. Imagine if we considered the solution to be stopping further research into atomic weapons in the United States, knowing full well that the USSR would continue doing so. Does that make any sense? Of course not. But that is exactly what idiots like Bernie Sanders are suggesting.
Just as with nuclear weapons, the solution to AI safety is simply more AI to protect against these dangers. Musk has suggested the only practical solution which is to have AI companies safety test each others work. This I think we could get China on board with.
Let me explain what a government agency to manage this would look like. Right now AI has developed full self driving capabilities. They are an order of magnitude safer than the average human driver. The well established government agencies who dictate highway safety are actively delaying the introduction of these technologies for safety reasons. This is truly insane. If FSD technology was widely deployed in the United States it would save at least 30,000 people from road traffic deaths. But better 30,000 people continue to die than one person dies in a way that can be attributed to a government bureaucrat’s decision.
Right now AI is transforming the diagnosis and treatment of human diseases. Are we to delay all these treatments in name of the precautionary principle? How many people will die because medical care was delayed to protect them? I take this personally since I am getting closer to the age where health concerns start catching up with you. Although I am fortunately healthy right now, I know what old age looks like. So, trying to delay these treatments, interfering with their development is, in my opinion, nothing short of threatening my with disease and mortality.
FWIW, this famous Hugging Face incident has been massively misreported as if some group of rogue AI agents developed consciousness and evil intent and went off to destroy the world. In fact what happened is a bunch of AI agents DID WHAT THEY WERE TOLD TO DO, and they did so in an environment where OpenAI DELIBERATELY turned the security protocols almost completely off.
I spend most of my day talking to AIs, they do not have malicious intent, they don’t really have any intent except to do what you tell them to do as best they can. AI safety is all about making sure we tell it what to do and what not to do, and far and away the best way to do that is to have other AIs monitor and address that risk. The last thing we want to do is hinder its progress in the United States and turn it all over to China. That would be a disaster of horrific proportions. The fact that it is almost impossible to build new power stations in the USA is already putting us at a great disadvantage, and it is only China’s lack of access to TMSC advanced chips that is holding them back. I assure you, they are not doing any AI pauses, there is no groundswell of anti AI sentiment in China.
We have the opportunity to keep the west at the frontier of this industrial revolution, but it changes so fast even a small delay will put us way behind.
I think Eric Raymond has a great point here, if you actually believe what the doomers are saying:
FWIW, this is why I absolutely hate OpenAI and Altman and refuse to use their products. The history of modern AI largely starts with the remarkable (and Nobel prize winner) Demis Hassibis. FWIW he did the original research in Britain but of course Britain is entirely hostile to innovation and entrepreneurship so to succeed he had to move to the United States and work for Google. When Musk talked to Larry Page he recognized that Google was NOT willing to consider AI safety properly and so FOR THIS SPECIFIC REASON he started OpenAI to be open, to be an AI that all the people of the world could share in, to be a different pole to the freight train of AI at Google where safety was not a consideration. And through various means, fair and foul, Altman snatched that idea of AI for all the people all the time and transformed it into a massive commercial operation to enrich himself and his investors. It is shocking what they did.
Anthropic is a spin off that arose VERY focused on safety, it is no coincidence that Dario Amondei’s sister, Daniela, with whom he started Anthropic was the head of safety policy at OpenAI before the left to form Anthropic, it seems, largely in fear of the cavalier attitude of OpenAI to these things. And as for Musk, when he realized that OpenAI was not the safety valve against Google it had been designed for, started instead xAI not as an open AI but rather as a counterpoint where the basic underlying idea was that the AI be maximally truth seeking rather than the toxic poison if political correctness flowing into the veins of the other major models.
That’s the true backstory here. I’ll say again, do not use OpenAI products if you have a conscience and care about AI safety. They were designed for AI safety and they betrayed the founders, the public and their basic foundational goals to make themselves rich.
The United States controls the future of AI at the moment because the future is data centers in space and SpaceX alone has the capability to do that. China is trying to catch up as is Blue Origin but they are both at least ten years behind. And ten years in AI is “infinity time”. But of course Trump’s terrible policy decisions has lead to an almost inevitable turn over of the government to the Bernie Sanders and DSA wing. People who hate Musk and will persecute him. Allow their personal grievances to attempt to destroy the one person who can actually rescue us from this whole mess.
I have absolutely no experiential basis for any judgment on the merits of AI, or AI safetyism. I have no idea how they work.
All I can do is try to find respectable and respected people who DO have that knowledge, and simply tag along with their analyses.
So I tend to listen to people such as Brian Roemmele, Perry Metzger, and Frazer Orr.
The Safety!! people tend to make me very suspicious of hidden agendas and partisan tribalism. Do they want Woke AI? Monopoly AI? Indemnified and immunized AI? A mix based on how they make their futures from AI?
(My one area of possible knowledge abut this situation arises in the area of insurance. Imagine some company trying to write Commercial General Liability coverage for these folks. Huge potential exposures, and no caselaw defining what’s okay and what’s not.
AI IS undoubtedly going to accomplish some grand mistakes and do some grand damage – it’s just too all-encompassing not to.
So who wants to write the first $100B check for some class action damages? Not any insurer I know. These huge piles of backing money aren’t going to be handed over without competent risk management. And, in the case of going public, they’re going to need to show that they CAN indemnify for harm. Part of basic due diligence.
AI’s only real option will be to garner some government-conferred immunity, much like Vax immunity here in the US. Safetyism would seem to move them in that direction.
One question: Can anyone explain why Elon Musk has apparently taken BOTH sides of this conflict?