The “AI is going to kill us all” news cycle revved up, as OpenAI’s Sam Altman and xAI’s Elon Musk joined Anthropic CEO Dario Amodei’s call to invite independent reviewers into AI labs to ensure the world’s smartest computer scientists don’t accidentally cause the extinction of the human race.
asserted
scientists → go → race
A calm read of Amodei’s proposal may lower your blood pressure.
uncertain
read → lower → pressure
He’s proposing things every AI lab needs to do anyway, mostly for commercial reasons: Operational excellence, alignment, interpretability, testing, and evaluation.
asserted
lab → propose → reasons
Imagine if an auto company CEO wrote a blog post suggesting that the industry should try to make reliable cars that react in predictable ways when you turn the steering wheel and hit the accelerator and brakes.
asserted
you → imagine → accelerator
Oh, and maybe try to figure out how these piston thingies make the wheels turn.
asserted
wheels → try → ?
Of course AI companies need to do these things.
asserted
companies → need → things
Who would put AI in charge of anything important if they didn’t know what the technology was going to do, or why?
asserted
technology → put → what
Why wouldn’t AI companies invite independent evaluators in to help?
asserted
companies → invite → evaluators
If philanthropists want to provide free labor, why put up a fight?
asserted
philanthropists → want → fight
What Amodei didn’t address is what happens when independent evaluators decide that a frontier lab they are monitoring should simply shut down operations, leaving investors holding the bag.
asserted
they → address → bag
There is a risk that Anthropic, or any other AI company that gets this feedback, will simply find a different independent evaluator.
asserted
that → be → evaluator
Amodei proposes evaluating the danger level of a model by looking at checkpoints.
asserted
Amodei → propose → checkpoints
“If models have capability X, then they need to be accompanied by certifications of alignment properties Y and Z,” he writes.
asserted
he → have → Y
But model capabilities aren’t the most instructive measure of whether they pose a danger to humanity.
asserted
they → pose → humanity
If this were the case, today’s panic would have begun 29 years ago, when Deep Blue beat Garry Kasparov at chess.
asserted
Blue → begin → chess
Humanity would have no chance.
asserted
Humanity → have → chance
Thankfully, artificial intelligence doesn’t work like human intelligence.
asserted
intelligence → work → intelligence
Just because a model can solve an extremely complex math problem doesn’t mean it could plot and execute checkmate against humankind.
uncertain
it → solve → humankind
Even the ability of an AI model to improve the creation of new AI models (called recursive self-improvement) is not the thing that endangers humanity.
asserted
that → improve → humanity
To do that, AI models would need to reach a level of sophistication that allows them to operate independently on tiny computers, and then to continually learn in secret.
asserted
them → do → secret
Without that ability, humans can simply unplug them.
asserted
humans → unplug → them
Frontier labs are certainly not building for that scenario: They’re spending hundreds of billions of dollars on AI data centers with the assumption that AI will continue to require massive, powerful computers for many years.
asserted
AI → build → years
If they build a super-intelligent AI model that can run on any computer and continuously learn, they will be putting themselves out of business.
asserted
they → build → business
There are plenty of real, even immediate, AI risks, but human extinction can’t really happen as long as there’s an off switch.
asserted
extinction → be → risks
There’s no mention of the doomiest scenarios in Amodei’s blog, or by people like the former Anthropic researcher Jacob Coxon, because even the most concerned people inside these companies don’t see it as an immediate threat.
asserted
people → ’ → threat
The subject of AI safety, however, may have escaped containment in the laboratory that is San Francisco, spread by a breathless media for whom mass extinction is — let’s face it — an incredibly fun story.
uncertain
’s → escape → it
Now that it’s infected the networks of Washington and Brussels, the biggest risk may be that hysteria replaces thoughtful debate.
uncertain
hysteria → infect → debate
A new list on the rationalist site LessWrong, long a hub of worries about AI, offers “some ways AI could kill us,” including deadly viruses, killer drones, and blocking out the sun.
uncertain
AI → offer → sun
Notable
The “AI freakout” has reached a tipping point, the Wall Street Journal declares.
asserted
Journal → reach → point
Altman told Fortune the company had delayed its IPO to settle safety concerns.
asserted
company → tell → concerns
Momentum toward a bipartisan AI safety bill stalled in Washington Friday as some Democrats worried it wasn’t tough enough, Semafor’s Ashley Gold scooped.
asserted
Gold → stall → Washington
The politics of AI are changing fast, and a conservative leaders last week described the technology as unleashing a horde of unwelcome “algorithmic immigrants.”
asserted
leaders → change → immigrants
