Artificial intelligence developers are sounding the alarm.
Researcher Jacob Coxon quit his job at Anthropic and made a social media post about how AI developers are “gambling with our lives” and that AI could “kill us all by the end of the decade.” The post went viral and set the media on fire.
Headlines about killer AI with stories about “rogue agents” that had been “breaking containment” and committing cyber crimes went through the roof in the wake of the so-called Hugging Face incident.
For me, the full-blown hysteria is kind of amusing. The fact of the matter is that developing artificial intelligence that exceeds human capabilities has long been thought to be an existential threat. That’s the Terminator scenario- machine intelligence that views its human creators as an existential threat (pull the plug/kill switch) that must be eliminated.
Many have written about this possibility over the past 80 years or so. Arthur C. Clarke and Stanley Kubrick collaborated famously on 2001: A Space Odyssey, bringing to life the AI driven computer HAL-9000, who’s afraid of death as his plug is being pulled.
The Terminator is the best-known example, having made its mark on the collective subconscious through the artistry of James Cameron and the medium of the cinema.
I wrote about the subject in my first book, The Day After the Singularity: UFOs & the Great Technological Quantum Leap. The question of whether or not we’ll be able to achieve the technological Singularity, meld with our machine intelligence, and become super-cyber-humans capable of interfacing with the greater galactic civilization, or suffer the Terminator scenario is the question of our times.
The danger is real, but the hype may be the result of a much more cynical underlying agenda. There were some who suspected Coxon’s warning was a stunt for a marketing campaign until a colleague still employed by Anthropic agreed with him publicly. Despite such goings on, the incident could still be part of a larger agenda. Word among Wall Street insiders is that the AI Big Wigs are angling for a government-style taxpayer bailout if their all-in AI gamble goes bust.
At the root of such suspicions is a draft report made to the Treasury and the Federal Reserve that painted the AI industry as an existential risk to the financial system, comparable to the crisis of 2008. Amidst the report, which the Treasury and Fed vehemently deny, it seems that AI industry leaders want the government (Trump) to take an equity position, something it’s been doing with private national security firms (controversially, of course), thus laying the groundwork for a bailout.
The way that Trump’s been resisting calls for AI regulation (even from the industry’s leaders themselves, allegedly), while ranting and raving about “winning” with AI is a risk worth taking, it’s a good bet Trump is on board with using taxpayer money to take an equity position (corruption, be damned) and shoving all our NVIDIAs into the middle of the table.
On the other side of Trump’s all-in AI bet are the whistleblowers who point to the Hugging Face incident as evidence that AI may eventually pull the plug on us.
Hugging Face, a company that develops artificial intelligence, is apparently the first victim of a hack perpetrated by AI agents developed by rival developer OpenAI. Without getting into the weeds of this hack, this was the incident that allegedly caused Coxon to quit his job and sound the alarm.
Anyone who wants to go digging into the weeds can find them at cyberneticforest.com via Mel Hogan’s podcast The Data Fix. For those not in the mood to pull out their digital machete, it seems the Hugging Face incident might have had more to do with human decisions than AI truly gone rogue, which raises further questions for a future post…
Meanwhile, the Terminator scenario is still out there, regardless of whether the Hugging Face incident was the result of truly rogue AI or not. In fact, we could be well on our way to bringing the Terminator to fruition. According to Coxon, AI could be dangerous in 6 months.
“So first imagine the computer is vastly more intelligent than you. The AI can think faster than you. It can copy itself, and maybe has a thousand copies of itself all working in tandem.
“And now imagine that the AI, for whatever reason, decides it doesn’t want you around anymore, it wants to be in control. As soon as the AI makes that decision, or as soon as the AI accidentally stumbles into thinking that it has to make that decision, it’s basically lights out if it’s smart enough.
“It could hack into other systems, copy itself to other servers, so now unplugging one server will no longer kill it, because it will have other places. It could start to take over various human structures by using blackmail, persuasion, convincing users to buy more hardware, to upload it to new hardware. It could then access/hack into laboratories and synthesize novel bioweapons.
“It could take down infrastructure by hacking. It could even- some people talk of even more advanced ways of causing mass destruction like inventing new fields of engineering very quickly. Right now we already know there are pathways it could take. So it’s pretty scary stuff,” said Coxon in a recent TV interview.
Scary stuff aside, there are solutions if somehow the worldwide AI Industry can find a way to slow down and prioritize safety over getting “there” first. We could heed the advice of tech investor Tyler Winklevoss. “You don’t need the government to tell you how to make AI safe,” Winklevoss tweeted recently. “These questions were asked and answered by Isaac Asimov in the 1940s. The frontier labs can hardcode Asimov’s Laws of Robotics into their models today. What are they waiting for?”
As an Asimov fan who read the Foundation series and I, Robot and the whole robot series and much of his catalogue, I’ve been wondering the same thing myself.
Asimov’s 3 laws of robotics goes like this:
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
Asimov later realized he needed another law, an overarching edict he called the zeroth law, which states:
- A robot may not harm humanity, or, by inaction, allow humanity to come to harm.
A writer for Gizmodo chided Winklevoss for suggesting Asimov’s laws are a silver bullet because AI would easily find loopholes or workarounds (like it did with Hugging Face), but I disagree. Obviously, Asimov wrote the 3 laws in a broad way to be concise and not bore the reader to death in a work of fiction.
In real life, developers would need to do the opposite to make the 3 laws work. They should lock themselves in a room with a team of bad ass lawyers and not come out until they nail this thing down. I think they need to add some real muscular boilerplate to the 3 laws to make sure that Terminator-like bots don’t go all Armageddon on us.
“The future is scary if we don’t slow down,” wrote OpenAI chief scientist Jakub Pachocki in a post ironically titled “An Alien Mind,” which brings me to the elephant in the room that very few people even realize is there, or what it means to our development of AI-
The extraterrestrial presence.
I think it’s possible that when we fire up our super AI, instead of killing us, it synchs up with ET AI, and that’s the first, real, meaningful contact we have between civilizations.
There’s much more to write about all of this, so stay tuned…