Carlos Lozada writes:

The term of art for mitigating the risk of rogue A.I. is “alignment” — that is, making technology operate in a manner consistent with the values of humanity. In [his essay] “We Must Pace the Frontier,” [Dareo] Amodei [CEO of Anthropic] invokes “align” or “alignment” 17 times. It is a benign and neutral-sounding term, as most good euphemisms are.

Amodei alludes to instances of A.I. models hacking into outside organizations as “recent alignment incidents.” And he reassures the public that his own company has “made clear progress in alignment — training models so that they remain safe, ethical, compliant with our guidelines and genuinely helpful.” (The “genuinely” is a nice touch there, as though we had an obvious way of discerning A.I.’s sincerity.)

Those principles of safety, ethics, compliance and helpfulness are embedded in the “Constitution” that Anthropic published in January for Claude, its collection of A.I. models — a document known internally in the company as the “Soul Doc.” It treats Claude as a near-sentient being, offering it “practical advice,” seeking its feedback and expressing “hope” that the system “genuinely cares” (there’s that word again) about A.I. safety.

I share that hope! But if the ability to lie can be an early marker of cognitive development in humans, might it not be so for artificial intelligence, too? Hoping for an A.I. model that just “genuinely cares” about human values and safety — and that behaves in a way that is ethical and helpful — is hardly a reassuring strategy. It’s an evasion of corporate accountability and an abdication of human responsibility.

This is one of the messages in another influential tech manifesto: “Magnifica Humanitas,” the first encyclical from Pope Leo XIV, published in May. The American pontiff is on board with deceleration, calling attention to the imbalance between the frenetic rate of technological change “and the slower development of awareness, norms, safeguards and institutions capable of governing its effects.” Yet he regards generic promises of alignment as inadequate substitutes for a true debate over human values.

“We cannot be satisfied,” he writes, “with merely calling for the moralization of machines — the so-called ‘alignment’ of A.I. with human values — without also having the courage to insist on a further condition: the possibility of openly discussing the ethical frameworks involved and subjecting them to shared standards of social justice.” Absent such expansive dialogue, he writes, technology invariably “takes on the characteristics of those who devise, finance, regulate and use it.”

Alignment is a meaningless concept unless we can answer the obvious question: aligned with what, and for whom? “Alignment” implies a consensus on what proper human values should be and their relationship to technology — a consensus that cannot be left to the White House or tech executives, no matter how many agreements they sign or constitutions they compose. [Continue reading…]