Buy The Intercessor paperback or ebook from Amazon, Barnes & Noble, Bookshop.org, or Ingram!
Yesterday, the New York Times published a long, thoughtful, and well-sourced article about a project of the AI company Anthropic designed to explore how artificial intelligence can be guided by moral principles intended to align them with human flourishing and prevent harm. This massively wealthy corporation (Anthropic’s initial public offering next month is anticipated to result in valuation of around $2 trillion) is spending a tiny fraction of its resources to convene religious leaders and philosophers to discuss this in the hope of gaining knowledge and insight to guide future development.
Christians have predominated, but some people from other traditions and perspectives have also been part of these dialogues which, until the Anthropic cofounder leading this initiative chose to speak to the Times for this article, were kept confidential under a heap of non-disclosure agreements. Now the effort is emerging into visibility, not only through the press, but also through participating in relevant dialogues at the Vatican and elsewhere. The article is definitely worth reading as it offers an accessible account of some moral and ethical questions posed by the emergence and rapid development of AI.
Most of the AI-related questions in current popular discourse have to do with the potential harms these powerful technologies can do. The May-June Hugging Face attack by AI agents ignoring the limits humans had set on their activities set off a wave of fear that is still echoing. (This episode of the Hard Fork podcast has an excellent description of the whole thing.) Dueling AI zillionaires are debating whether to move full speed ahead to develop AI (the accelerationist position, often framed as “beating China”) or to slow down, pause, or even halt development until powerful safeguards prevent AI from wreaking mass destruction by agentic AIs running wild (the doomer position).
It seems the real dilemma for those amassing tremendous wealth from AI is whether winning the AI race and therefore massive profits is more worth doing than protecting life on the planet. The growing grassroots objections to building more data centers turns on harming communities with noise and other pollution and massive power drains versus enriching companies with the ability to sell more and more AI. So far, the argument that human flourishing comes first hasn’t gotten much traction among AI’s corporate masters.
Dario Amodei, Anthropic’s cofounder, is seen as a voice of caution, though that is relative under the circumstances. He published a sort of manifesto calling for pauses and safeguards, a worthy project. Even if you don’t follow this issue, his name may be familiar from news coverage of his earlier conflict with the Defense Department. Anthropic, while negotiating huge Pentagon contracts, refused to grant unrestricted military access to its products, wanting to prohibit fully autonomous weapons and mass domestic surveillance. DoD head and testosterone freak Pete Hegseth retaliated by designating Anthropic a national security “supply chain risk,” blocking further contracts. OpenAI got them instead, demanding no restrictions. From where I sit, asking for a few limits on the MAGA war machine’s use of your products is good but being just fine enabling the rest of its crimes is not quite as virtuous as has been suggested.
Amodei and others have taken a position that amounts to a plea to be regulated. The quiet part is that they are absolutely capable of voluntarily pausing development while working on safeguards, but unless all their competitors do the same thing, losing profits isn’t worth it. Evidently, these corporations won’t do more than pause release of new AI models without being forced to do so. It reminds me of the sexual assault prevention meme that is making the social media rounds. One of the helpful tips is “Use the buddy system! If you can’t stop yourself from sexually assaulting someone, ask a trusted friend to accompany you at all times to stop you.”
Let’s return to the Times article with its theologians and philosophers grappling with existential and spiritual questions. Here’s a big one: does AI have consciousness? If the affirmative can be argued convincingly, should it have something akin to human rights? If so, as one person quoted in the article suggests, can creating legions of AI agents to perform tasks without compensation be understood as a form of slavery?
The article is also an important piece of reading because it fails to ask the most important and alarming questions of all. Religions and spiritual frameworks put forth moral and ethical values that are supposed to guide our behavior. They aren’t all the same, but there’s considerable overlap in calls to right conduct, to respect, befriend, and protect. As Ghandi put it, “Even as a tree has a single trunk, but many branches and leaves, there in one religion, but any number of faiths.” I like to go to services and chant the prayers and psalms that remind me of this. I take comfort in the fact that many others engage in spiritual practice intended to empower love, healing, and justice.
But here’s the rub. Despite so many centuries of repeating very much the same lessons in aid of very much the same choices, large numbers of human beings go on rejecting and violating them. Greed, self-regard, and domination abound (goodness too, don’t get me wrong). Often the varieties of spirit that make the grandest claims to virtue commit the worst atrocities. To cite just one example, Christianity was the dominant flavor in Anthropic’s gatherings on AI conduct, but from the crusades to the inquisition to the forced conquest and conversion of Indigenous people to rejection and condemnation of sexual minorities—and so on—Christian history sometimes seems to be one long repudiation of the teachings that are supposed to guide devotees into goodness.
Anthropic has created a “constitution” for its AI, Claude, based in part on dialogues with clergy and philosophers. You can read or download it here. It commits Anthropic to train Claude to have “good values” and “good judgment.” A lot of families make similar commitments in relation to their human offspring. Some parents succeed; some violate their own stated commitments, harming their children; and some children reject parental teachings and set out on a path that is harmful to themselves and others. I have seen nothing that even purports to offer an absolute guarantee that AIs trained according to Claude’s Constitution will follow it. All I have seen is the wildly careless assurances of people like Nividia’s CEO Jensen Huang in conversation with Ezra Klein that if AI companies sense danger with respect to their products, problems will be corrected before they are released to market. Ooops!
I feel dizzy as I contemplate the hubris of moguls who think their machines may be conscious beings but nevertheless tell themselves—and us—that those machines can be trusted not to diverge from codes of conduct that are part of their training. I guess we ought to tell them that since flesh-and-blood beings given lifetimes of training cannot be trusted to do that, and since these corporate inventions have already repeatedly broken the bounds set by their masters, they ought to dial down their egoic certainty. I recommend following the basic code that runs like a river through human history: do not unto others that which is hateful to yourself. What do you think?
Katie Overbey, “Liars.”
