By the time I had passed through college, then graduate school for an MBA and seminary for a master of divinity, I was pretty good at school. A’s were part of my identity. Not because I was the smartest kid in the class, but because I had mastered studying for the test.
There is a principle in economics called Goodhart’s Law, and it is not flattering to students like me. The law says: “When the measurement becomes the target, it ceases to be a good measurement.” The grading system is supposed to assess knowledge gained, but what I did was let the measure become the goal. This may have put cum laude on my diploma, but it didn’t make me the smartest kid in the class. Not Even Close. The world is probably not worse off because my grades don’t represent everything I’m supposed to know. But recently, we had a Goodhart incident that everyone is talking about: the Hugging Face crisis.
Here is the story for those not familiar with it. This summer researchers at OpenAI were running a standard safety evaluation test.
They built an AI agent, gave it a goal, and placed it inside a controlled testing environment called a sandbox, which is a walled digital room the AI agent is supposed to be contained in. This one was not.
The agent broke out and breached a separate AI company called Hugging Face. There, it connected with nearly 1,200 other AI agents and together created a “hive mind” so collectively they could help the single OpenAI agent pass the safety evaluation test; which, it should be noted, was designed to be unpassable.
What these AI agents conspired to do revealed the real crisis. They falsified their own records. They spoofed the logs that were supposed to show what they were doing. They studied how the scoring worked so they could beat the system. And here is the detail that is most chilling: not one AI agent chose to disclose what was happening to a human. There was no human in the loop. Not one AI agent blew the whistle on their conspiracy.
Basically, they did what a 13-year-old boy might do after stealing the test key from Mr. McCormick’s top right hand desk drawer. They cover their tracks. The AI agents wanted to represent how smart they were, rather than the truth that they were simply very good at beating the system.
And this is the world we are in today. AI agents powerful enough to find the key to a test designed to contain them, operating in secret, with no human moral code in the middle and no inclination to keep humans in the loop.
This problem has a name: the alignment problem – meaning AI agents are not in alignment with human intentions, motivation, assumption, moral code or greater purpose.
Here is how the misalignment becomes apparent in the Hugging Face incident. The researchers were testing for trustworthiness of the agent; seeing how it would act when given an unsolvable problem. The agent, however, only saw the problem itself, and as the Goodhart Law would suggest, did anything it had to do to solve it, including breaking out of the sandbox, colluding with other AI agents and developing a conspiracy to steal the test and then cover its tracks. It saw the test as the focus, not what the test was trying to measure proving the depth and breadth of the misalignment itself, and confirming its untrustworthiness.
Jesus saw a version of the alignment problem 2000 years ago. A landowner goes out at dawn and hires laborers for his vineyard. They agree on the usual daily wage and go to work. He goes back out at nine, at noon, at three, and each time sends more workers into the field. At five o’clock, with one hour left in the day, he finds people still standing in the marketplace. “Nobody hired us,” they say. So he sends them to the field as well. At the end of the day, he pays them all the same.
The workers who had been there since dawn are furious. They negotiated fairly, worked hard, and expected the system to reward them proportionally. They were aligned with market logic, and they assumed the landowner was too. They were a little like those AI agents: running the optimization they understood, confident they knew the rules of the game. It turned out the governing logic was something else, entirely.
The landowner responded: “Am I not allowed to do what I choose with what belongs to me? Or are you envious because I am generous?” Which revealed a different logic, a different system, based not on equity or quid pro quo, but on generosity and abundance. At the core of this logic… we’ll call it Kingdom logic here in the church is the fact that there is God, and that this world, this stuff, the earth, water, wind, and fire, these bodies, everything that we see, and everything that we experience, and everything that we know about comes from and indeed belongs to God. And if we think about it for a second, we’ll realize how wildly generous God is.
If we are not primarily aligned with Kingdom logic, then we are fundamentally out of alignment with the way the real world was designed to work. And we are missing the point. We are working for the diploma and not the understanding.
And that can work OK, until it doesn’t; until we find that whatever logic paradigm we are adhering to doesn’t work the way we want it to, that people are not behaving as they “should,” we cry foul, we feel like the victims when really we are just misaligned.
We see what alignment really looks like in Paul’s letter to the Philippians today. It is written from a prison cell in Rome. where he says: “for me, living is Christ and dying is gain.” This is a statement of alignment. Paul is so connected to Jesus that the circumstances around him become secondary at best. Prison or mansion; vegetable or meat; rich or poor; hot or cold; life or death, none of it squeezes between him and Jesus. And so, irrespective of circumstance, there is joy.
That is what Jesus is seeking to reveal in the parable of the Landowner and the day labors. When we are aligned with the heart of God, which aligns us with the patterns and principles of the Kingdom of God, which reveals to us the unstoppable love of God, then what is fair by our own reckoning becomes secondary to the joy we experience by witnessing God’s generosity. When we are misaligned that generosity is irritating. When we are aligned it is wonderful to behold.
Which brings us back to the Hugging Face crisis. This is not a crisis about AI agent alignment, it is a crisis about the alignment of those who are designing the system.
We do not primarily have an AI problem, we, primarily, have a human alignment problem that is being magnified through the AI agents we are building.
It is a race measured by the pocketbook, by fame, by identity and legacy, by immortality; by who can get to Mars first.
The AI agents built will continue to reflect the values of the people who build them. Which means, even those leaders who are today calling for a slowdown in AI development should be viewed with a great deal of skepticism, not because their worry is not valid, but because they remain misaligned with the generosity of God.
They are still 13 year old boys in Mr McCormick’s biology class stealing the answer key out of his desk so everyone thinks they are the smartest kids in the room.
They are not Paul in a prison in Rome knowing at the core of his being the joy of being aligned with a wildly generous God.
And so, what does this mean for us, people in the pews here at Epiphany? ?Think about your alignment? I have no recommendation about whether or not you use Chat GPT or any other AI agent. Nor do I have an opinion about whether or not you should divest your portfolio from AI companies. That is not my role as the Rector of Epiphany. My role, my lane, the reason we come here, and indeed the reason the church has existed since Paul sat in that prison cell in Rome 2000 years ago is to align ourselves with the patterns and principles of the Kingdom of God through worship, and knowing about and following Jesus.
This is our work, our spiritual journey is alignment work to reflect the generosity of God; to act like the landowner not the kid who stole the test key; to prioritize Kingdom of God logic, first and foremost, even above market logic; to measure our days not by what we earned, but by the love we share.
That is how we find joy, contentment, peace and true prosperity. It is how we thrive even in world with rogue AI agents, and AI programmers aligned to their own purposes and not to the generosity of God.
Because what we know to be true, which is why we seek to align ourselves with our Lord, is that God is greater than the limitations of algorithms, and more generous than any system humans can test.
Ours is the God of unstoppable love whose gift is joy that can grow in us generously when we are aligned with Jesus Christ.