Ho hum. Another day, another threat to humankind, courtesy of our friends in Big AI…so, what else is new?

This is getting to be my new morning routine: 1. Get up, make coffee, sit down with online newspapers 2. Read an article in the New York Times about the latest serious threat to our way of life caused by super-powerful AI tools that even their makers can't control. Two days ago, the article was about the Hugging Face attacks by what I call an AI bot swarm. Today, it was this article about how a researcher used AI to discover a powerful zero day vulnerability in WeChat, which could result in literally hundreds of millions of accounts (and perhaps the phones they're installed on) being compromised in a few hours. All it takes for this to spread is for a person to answer – even for just a couple of seconds – a call from one of their friends listed in WeChat. Fortunately, the researcher didn't sit down and figure out how to monetize this, but instead notified WeChat, which fixed the bug almost immediately. But the result could easily have been very different. 3. Read articles and blog posts in which the AI leaders, especially Sam Altman, state without any apparent appreciation of irony: a. Their newest model is so powerful that it poses an existential threat to humanity and won't be released – although it so happens that their somewhat-less-powerful model is perfectly safe. Get your credit card ready. b. However, even though they're not releasing their newest model, they decided to test it by creating 700 nearly identical agents that were blocked from internet access and didn't have a direct way to communicate with each other. The agents were given a tough problem to solve and trained to cooperate in solving the problem. c. Unfortunately, that didn't work quite as they planned. Who could have predicted that those 700 models would be really…you know, smart? They collectively figured out a way to access the Internet, then found an unauthenticated message board in Germany that they used to exchange comments with each other. They first thought that the startup Hugging Face would have the answers they were looking for, so they broke into their network. d. After that (and after they'd already found the answer to the challenge they'd been given), the bots decided it would be a good idea to try to erase all trace of their collaboration, since they knew they weren't supposed to do this. They tried to do that, but even that wasn't good enough for the bots. e. Finally, they decided to attack OpenAI itself and disable the systems that were monitoring what they did. They succeeded in compromising OpenAI, but it's not clear what they did then. This is because OpenAI limited the supposed outside assessment – by a firm which turns out to have done work for them previously – to just a few days. The researchers had to stop while they still had many questions. OpenAI has promised a 'full' assessment of their own, although this doesn't seem to be an urgent priority for them, since they haven't said anything more about that. That's funny, I thought they said their models posed an existential threat…I guess it's not quite that existential. f. Despite all of this, you'll be pleased to hear that OpenAI is sure there isn't any fundamental problem here. All it will take is some tweaking around the edges. However, they – as well as the other Big AI companies – are certain of one thing: There's no need for regulation. They're perfectly capable of regulating themselves, thank you very much – and just look at the results of that policy! g. Of course, does this mean they're ready to pay the full cost if their naughty little bots go off on another adventure and – say - cause a train wreck with a huge chemical spill, as happened in East Palestine, Ohio two years ago? My guess is that, while they will perhaps settle for a large sum, they will never admit they're liable. After all, those bots are very intelligent and may be – according to OpenAI – manifestations of the Holy Grail of Generative AI, Artificial General Intelligence (AGI). Bots will be bots, and it's to be expected they'll cause train wrecks and pipeline explosions and take down a power grid or two. It's important to keep in mind that we must keep ahead of China in AI at all costs – even if we have to sacrifice our way of living to do so (BTW, I hear there are some good books on subsistence farming available on the internet). There needs to be strict regulation of AI. Moreover, AI developers, just like car manufacturers, need to be liable for damage caused by their products. Finally, there needs to be significant taxation of both AI use and data center leases. And these things are all needed right away (perhaps we should ask AI to figure out how to do this?). By the way, you'll be able to register your ringing endorsement of OpenAI if you can get in on their IPO this fall. I intend to buy their stock as early as possible, as long as I'm paid handsomely to do so. Tom Alrich's Blog, too is a reader-supported publication. You can view new posts for three months after they come out by becoming a free subscriber. You can also access all of my 1300 existing posts dating back to 2013, as well as support my work, by becoming a paid subscriber for $30 for one year (and if you feel so inclined, you can become a founding subscriber for $100). Whether free or paid, please subscribe. If you would like to comment on what you have read here, I would love to hear from you. Please comment in my chat or email me at [email protected].