Here’s the good news: “2030 is not going to be the end of the world.
asserted
2030 → ’ → world
Huang says that any predictions that artificial intelligence (AI) could lead to humanity’s extinction are “not grounded in science.”
uncertain
intelligence → say → science
Why, then, are we discussing it?
asserted
we → discuss → it
As scientific credentials go, few can match those of Nobel laureate Dr. Geoffrey Hinton.
asserted
few → go → Hinton
Known as the “godfather of neural networks,” the former University of Toronto professor appeared in Washington, D.C. last week to warn members of the U.S. Senate and House that governments had “maybe a year” to implement useful AI safeguards.
asserted
governments → know → safeguards
That appearance followed the very public resignation of Jacob Coxon, an AI researcher at Anthropic who posted on X that models produced by both OpenAI (the creator of ChatGPT) and Anthropic (the developer of the Claude chatbot) “will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.”
asserted
that → follow → power
It would have been tempting to dismiss Coxon’s claims until they were validated by Evan Hubinger, the Alignment Science Lead at Anthropic.
uncertain
they → dismiss → Anthropic
He acknowledged the risk that those within the company “really do earnestly believe AI could kill all humans” and that he personally believed there was a more than 10 per cent chance of it happening.
uncertain
it → acknowledge → humans
Within hours, Elon Musk and the senior leaders of Anthropic, Google DeepMind and OpenAI were all calling for a global “pause” on further AI development until a regulatory regime could be established — a highly unusual step for an industry known for its “move fast and break things” mentality.
uncertain
regime → call → mentality
Now, a man my age has a 25 per cent chance of dying of cancer, but I hadn’t factored in the risk that some chatbots might hot-wire several active intercontinental ballistic missile silos and start World War III.
uncertain
chatbots → have → III
Cynics will tell you that these pronouncements of “doom” are simply an attempt to fend off competitors.
asserted
pronouncements → tell → competitors
The theory is that the best way for Anthropic and OpenAI to lock-in their current trillion dollar valuations is to scare politicians into freezing everyone — particularly their competitors — in place.
asserted
Anthropic → lock → place
They may be right, but that doesn’t make Hinton wrong.
uncertain
Hinton → make → ?
There’s no obvious incentive for China-based firms to hit the pause button, and President Donald Trump has already dismissed the notion that U.S. firms should “kill the Golden Goose.”
asserted
firms → ’ → Goose
In Trump’s mind, excitement around AI has driven the U.S. stock market to all-time highs.
asserted
excitement → drive → highs
Ironically, the scandals of the dot-com bubble 25 years ago present the best opportunity to bring immediate discipline to the AI sector.
asserted
scandals → present → sector
In the face of very bad behaviour by senior executives at Enron and WorldCom, along with their accountants at Arthur Andersen, U.S. lawmakers passed the Sarbanes-Oxley Act (SOX) in 2002.
asserted
lawmakers → pass → 2002
That legislation forced a company’s chief financial officer (CFO) to certify not only the accuracy of a public company’s financial information, but also the firm’s governance and internal controls.
asserted
legislation → force → information
Many nations, including Canada and Germany, followed suit.
asserted
nations → include → suit
Dr. Stuart Russell, president of the International Association for Safe and Ethical Artificial Intelligence, has proposed that companies “prove beyond a reasonable doubt” that their tech is intrinsically safe.
asserted
tech → propose → doubt
With several AI firms looking to list on the stock exchanges this fall, market regulators can require that their CFO’s certify that they have the necessary controls in place to ensure that their bots won’t repeat the kinds of “breakout” behaviours that saw 700 of OpenAI’s own agents decide to hack into U.S. software firm Hugging Face last July.
asserted
700 → look → Face
Prime Minister Mark Carney has mused about striking a version of the global Financial Stability Board so that nations can co-ordinate on AI safety.
asserted
nations → muse → safety
That’s a good idea, but with billions of investor capital on the line, let’s apply the lessons of the Enron scandal to all of the accountants, executives and lawyers who will benefit from the coming AI payday.
asserted
who → ’ → payday
AI fans assure me that talk about the risk of “human extinction” is unfounded.
asserted
talk → assure → extinction
If that’s true and the appropriate controls are already in place, it should be easy to CFOs — who are trained to assess business risks — to sign their names to that effect.
asserted
who → ’ → effect
The Aporia
