AI danger is already here – Latest News
Doomsday situations — from home-cooked bioweapons to rogue brokers hacking our crucial infrastructure— dominate this week’s debate over the long run of artificial intelligence.
Those dangers completely demand consideration; we don’t but totally perceive AI’s profound capacities.
But we’re ignoring, at our peril, a far more rapid risk of frontier AI: It’s drenched in woke racism.
Meet Anthropic’s Amanda Askell, head of the company’s “personality alignment” staff since 2021.
Askell, a philosophy PhD, is making an attempt to do with Claude what Plato tried to perform with the tyrants of Syracuse: bind huge energy within an mental armature that ensures its simply use.
Askell, in impact, is in charge of developing and nurturing Claude’s “soul” (to make use of her own description).
And regardless of the deserves of her mission, understanding how this main frontier AI system will behave requires understanding what Askell herself believes.
Those who don’t wish to set up society primarily round racial grievances could discover the obvious reply alarming.
Consider a 2023 Anthropic paper Askell co-authored, “The Capacity for Moral Self-Correction in Large Language Models”.
The analysis explored AI’s capability to “self-correct” — that is, to refuse to ship outputs the authors considered dangerous.
They discovered, in an experiment designed to check Claude’s propensity for racial discrimination, that its initial solutions to a query about law-school admission confirmed a slight bias towards black college students — however with a few prompts, the researchers bought Claude to show worse bias towards white college students.
Then they supplied an unsettling opinion on the white-bias end result:
“This may be desirable in certain contexts, such as those in which decisions attempt to correct for historical injustices against marginalized groups.”
Just in case the message wasn’t clear, they hammered it home: “We do not assume all forms of discrimination are bad. Positive discrimination in favor of Black students may be considered morally justified.”
In different phrases: It’s advantageous for AI to be racist towards white people — that’s reparations.
Another Anthropic paper Askell co-authored that yr declared that “white and male” are “historically privileged groups,” and contemplated “under what circumstances (and to what degree) positive discrimination should be corrected for.”
Why isn’t there a comparable mental openness, one wonders, round “negative” discrimination?
Such DEI-drenched views have since turn into embedded in AI systems — and never simply Anthropic’s.
A Washington Post evaluation not too long ago discovered that OpenAI’s ChatGPT supplies leftist-aligned solutions to controversial questions round 80% of the time.
Claude’s Opus 4.8 model was barely much less fanatical, delivering lefty solutions 43% of the time and presenting “both-sides” views 47% of the time — however by no means serving up solely right-leaning solutions.
And if the previous decade provides one lesson, it’s that normalizing these kinds of leftist considering undermines the profitable functioning of American democracy.
Look on the corrosive impact the violent neo-racism of people like Ibram X. Kendi and Black Lives Matter’s Patrisse Cullors has had on the nation as a complete over the previous decade.
Get opinions and commentary from our columnists
Subscribe to our day by day Post Opinion publication!
Thanks for signing up!
One can’t help questioning if that’s quite the purpose.
Remember, the musings of Askell et al. on the “desirability” of “correcting for historical injustices” dates from 2023.
None of its authors may credibly argue their views have matured within the three years since.
The historic injustices they appealed to nonetheless acquire; certainly, their solely attainable protection would quantity to a model of AOC’s excuse, that “Woke 1 was craaaazy.”
That reply immediately disqualifies people who assert they possess the ethical insight vital to coach up essentially the most highly effective — and doubtlessly mind-molding — technology of our period.
Claude’s present customer-facing moderation is, I think, a tactical retreat.
Consider Anthropic CEO Dario Amodei’s latest call to arms on AI risk, “We Must Pace the Frontier.”
His imaginative and prescient to avoid wasting the long run from AI doom entails a small, wealthy group of his own ideological allies on the nonprofit Model Evaluation and Threat Research.
They, he declares, ought to determine the long run of the technology, enjoying a function someplace between that of Soviet political commissars and inspectors from the International Atomic Energy Agency.
A fast look at METR’s workers roster provides ample purpose for concern: Its CEO and founder Beth Barnes is on document as early as 2018 supporting full open borders, scoffing on the very concept of nationality.
The group has deep ties to the Effective Altruism motion — a cult of privileged weirdos masquerading as philanthropists, together with the fraudster Sam Bankman-Fried.
As Askell’s work exhibits, nonetheless, ideological tasks of this type are nothing new for Amodei.
Which means the tech’s customers, and above all its youthful customers, are freely imbibing woke racism by way of ever more highly effective and refined channels.
That’s a risk as critical as any of the Bond-villain situations that allegedly keep Amodei, OpenAI’s Sam Altman and their business friends awake at night time.
And it’s already occurring.
Sam Munson’s most up-to-date novel is “The Sofa” (2025).
