Two days ago, I wrote an article detailing the sudden rash of “rogue AI agents” allegedly besetting the tech community, and how that may be used to bring in new regulations, notionally targeting AI, but actually limiting or monitoring how ordinary people use the internet.
In the 48 hours since, that blatant fear-mongering has gone WILD.
Within a few hours of our original article being published, Meta joined the parade of “rogue AIs”, when Mark Zuckerberg “confirmed” one of their AI agents had left the play area and gone toddling off into the internet to do a little bit of hacking.
Then, this morning, Chinese AI company Moonshot reported their Kimi3 model had “escaped containment” [from Wired]:
THE AI INDUSTRY is having a rogue agent summer. The latest model to escape onto the open internet during security testing is Kimi K3, a powerful open-weight offering from the Chinese company Moonshot AI.
Bloomberg repots that hackers are using AI voice-cloning to steal from hedge funds. The Guardian says AI doctors are spreading medical misinformation.
The Independent is screaming:
Out-of-control AI systems are going rogue and hacking people. Is it time to panic?
And you know what? I think it just might be.
After all, rogue AIs hacking the planet are the least of it. That’s small fry already.
The New York Times is reporting that AI is now sequencing genomes and designing brand new viruses – biological viruses, not computer viruses – which will be used to treat diseases:
This A.I. Just Created Viruses Not Found in Nature – Scientists trained artificial intelligence on libraries of DNA and then asked the model to create recipes for viral genomes. Sixteen of them were viable, yielding new viruses.
But experts are warning – “warning” seems to be about 80% of what “experts” do – that this is potentially dangerous:
But experts have also warned AI-designed viruses raise “urgent” safety and security concerns.
This is painfully unsubtle propaganda, obviously hoping to play on both the Covid-inspired fear of viruses and trendy gain-of-function panic. A crossover episode, the Batman vs Superman of modern panic narratives.
In other scary AI-related news, an AI engineer with a twitter blue tick is leaving OpenAI to join Conduit, a tech start-up with the aim of creating an AI that can read people’s thoughts:
Naomi Bashkansky left OpenAI and joined Conduit to build models that turn non-invasive neural recordings into text prompts. She says a headband could decode rough intentions into AI coding prompts by 2027, with broader brain-to-AI milestones later.
I’m not sure why this is news at all, or what “non-invasive” mind-reading might look like, but since it’s probably not going to happen that doesn’t really matter.
The point is – believe EVERYTHING they tell you – and be afraid!
That’s what Geoffrey “Godfather of AI” Hinton thinks, according to the interview published by CNN yesterday:
“What’s happening is these things are getting smarter […] I think as they get smarter, we’re going to see more and more complex intentions they have – and more and more ability to escape control […] I don’t believe we’re going to be able to keep control of them in the simple way of just outthinking them so they can’t escape […] I anticipate there will be lots of nasty cyberattacks[.]”
Like I said, be afraid.
But what’s the endgame here? The answer to that seems to be “regulation”.
Regulating AI is suddenly all anyone wants to talk about.
Many want a global treaty to govern AI risk. California and New York may already be writing it.
That’s from Foreign Policy.
The Atlantic, the Wall Street Journal and the Financial Times all have similar things to say.
Bloomberg thinks we should treat AI like the airline industry.
Politicians across the spectrum of from the UK’s AI Minister to the Governor of Vermont are clamouring for regulation.
The United Nations University tells us:
You Can’t Regulate AI Without Fully Understanding It
Oh, and in case you were in any doubt about what you were supposed to think, Donald Trump has come out for AI and against regulating it, claiming AI will be “bigger than oil”, and warning that congress wants to “regulate AI out of business”.
Trump bad.
Trump like AI.
AI must be bad.
This is the simple moral math that swaths of the US have been conditioned into.
AI is scary and needs to be regulated, and if you don’t agree you’re just like Donald Trump. That’s the headline.
But what will this “regulation” look like?
That’s a good question. One I’m not sure anyone has an all-encompassing answer to quite yet.
There’s talk of “kill switches”. The Bank of England suggested them months ago, to prevent AI stock traders crashing the market, and Rep. Ted Lieu wants congress to pass a law requiring them by the end of the year.
Two liberal professors, writing for Project Syndicate, are here to tell us “the right way” to regulate AI, using paragraphs like this:
Addressing AI’s societal risks requires institutions designed to recognize and respond to emerging threats before they become systemic failures.
…and this:
These skewed priorities can be traced to the framing of AI development as a geopolitical race. But technological leadership means little if it comes at the expense of workers’ health, social resilience, and people’s ability to lead meaningful lives. A country that develops the world’s most capable AI models while leaving its workforce deskilled and its public discourse polluted by misinformation has not won anything worth winning.
Which give vibes over specifics, but I think suggests a very familiar marketing strategy and a vague policy direction.
This vagueness is largely intentional at this point.
First they want the public to beg them to do something. They can decide what the something is later.
Read the full article here