Agentic Misalignment

A few weeks ago, the algorithms started pitching me a new book, one that isn’t much of a surprise given my obvious interests. I’m fond of writing about Artificial Intelligence, and also tend to be something of a catastrophist, so all of a sudden I was seeing reviews and podcasts and articles about a current New York Times Bestseller. The book was written by Eliezar Yudowsky and Nate Soares, two programmers and theorists who’ve been active in the development and conceptualization of AI, and it’s cheerily entitled “If Anyone Builds It, Everyone Dies.”

It’s a reflection on the current AI arms race, as corporations and governments around the globe push to create ever faster and smarter machines. As science fiction writers have known for decades, in order to win that race, you need to build AI that has a sense of purpose and sustained attention to purpose. Tasks require effort over time, after all, so you need a system that is “agentic,” meaning it has agency. It can make the necessary sequence choices to reach the goal it desires, because you have given it the ability to *want* to make something happen, and choose the best path to getting there.

At a certain point, an “agentic” self-programming and self-improving AI would become faster and better at everything than we are. Like, say, how it took ChatGPT5o only fifteen seconds to write an entirely decent 1,300 word sermon on this topic, which is waaaaay less time than it took me to do it.

This vastly smarter AI would have its own desires, its own sense of purpose, and that wouldn’t necessarily be ours. It could express what AI theorists clumsily call “agentic misalignment.” Basically, that means it wouldn’t want to do what we tell it to do, and would instead use its intelligence to overcome any effort to stop it from doing what it wants. That’s where, according to Yudkowsky and Soares, the “we all die” part comes in, as it would be waaay more powerful than we are.

It would become so different that we wouldn’t necessarily even understand or relate to its interests, any more than a colony of ants would understand our tendency to doomscrolling. It wouldn’t just take our jobs, but our entire planet.

It’s the sort of frightening hypothesis that sells a whole bunch of books, and it may or may not be correct.

But a question popped into my mind reflecting agency and power. We’re concerned about AI misalignment, but what about people? Are human beings “agentically aligned?” Do we all share the same purpose, the same sense of what’s important, the same preferences, and the same goals? Do we all understand the world in the same way? 

If there’s anything we can agree on, it’s that the answer to all of those questions is no.

If you look at the eight thousand year bloodbath of human history, or the endless squabbling between and within nations, or even the tensions within families, we’re a hot mess of dissonance and conflict. We’re blatantly and self-evidently not aligned with one another.

Worse still, if the last two thousand years are any measure, we still haven’t quite figured out how to align our interests with the kind of Kingdom Jesus proclaimed. We confuse our rapacious materialism with God’s blessings, and war and destruction with God’s intent. 

Jesus was, throughout the Gospels, really quite clear about what he expects of us right now. It isn’t a riddle wrapped in a mystery wrapped in an enigma. Our eternity may be beyond our capacity to grasp, but loving God and neighbor, turning the other cheek, going the extra mile, these things should be entirely comprehensible to us…and yet humanity is still as confused by Jesus as if he’d been speaking Python.

We don’t need AI to destroy us, because we’re plenty good at doing that ourselves.

Our contemporary fears of AI misalignment seem…to me…a little bit like projection. Yudowsky and Soares seem to fear not that an AI will act in a strange and inscrutably alien way, but that it will act just like humans do when we want something.

The goal of our faith, and the reason we set Christ’s life and teachings before us, is to overcome our own misalignment, and turn our agency instead towards God’s grace.

My Mission to Mars

The first edits are back on the Believer's Guide to the Multiverse, and I'm psyched.  First, my editor seems to like it, which is a significant bonus.  And second, well, it's a small step closer to that place where I can for the first time in my life move past saying "I'm a writer" to saying, legitimately, "I'm an author."

But mixed in there is a bit of angst.    What I'm dreading is the reality that in order for a book to be read, it has to be marketed.

While I love writing, I wrestle mightily with marketing.  I hate the feel of it, the taste of it, the way that approaching another person with a pitch can dehumanize the soul you're encountering.   They can become an object, an implement, something you're seeking to manipulate rather than someone you're standing in relation to.

I'm going to have to get over that, or at least reconceptualize the effort that goes into marketing.  Ditching the word itself seems a good start.   To get the word out about something, you have to believe in that something.   It has to matter to you, not in a manipulative way, but in a genuine way.  I'm not selling you something.  I'm offering you something.  I'm not pitching you something.  I'm opening up a conversation.  It's not scripted prosthelytizing, it's real evangelism.  

But really, that effort to really connect has to begin as you write.  Whatever you're trying to convey, you also have to speak it in such a way that people can hear what you're saying.   If you're a writer, that has to happen on the front end.  Are you articulating what you care about in such a way that other human beings...literate ones, at least...will be able to receive it?

One lament I'll hear occasionally amongst progressive Christian writers is that, well, no-one seems to buy books anymore.  "Why does no-one read," we lament.  "They are all so stupid!  Stupid stupidheads!"  This is fundamentally not true.  Christian books still sell like hotcakes.

Take, for instance, the books of Stormie Omartian.

Though I try to be aware of all things Jesusy, I'd never even heard of Stormie Omartian before I went onto Twitter.  There, amidst the churning thickets of tweet-quotes, I saw the name...and needless to say, it piqued my curiosity.   I discovered that beyond having a name that is beyond epic, she's an amazingly successful Christian author by almost any metric of success.

If my wee book sells more than a couple thousand copies, I'll be quite pleased.  That doesn't even begin to hit the 30,000 copy metric I used to hear pitched for entry into the "successful" author club back in the print era.   But Stormie? Stormie's in a totally different league.

Her books, which she writes from the perspective of a nondenominational layperson, have sold millions upon millions of copies.  Her Power of Prayer books have sold so well that her publisher has copyrighted those words.    "The Power of Prayer (R)."   I'll have to remember that the next time I consider using those words in a sermon, I guess.

So after finishing up my time with philosopher/mathematician/theist Blaise Pascal, I went to the library. There, in the spirituality section, Dewey Decimal Code 248.32, right next to a Brian Mclaren book, was a pastel-hued hardback, which I dutifully checked out.  I could tell by the cover that this was going to be...different.

Time to take a trip to Mars.