The archer who overshoots his target misses as much as the one who does not reach it. And my eyes trouble me as much when I raise them suddenly to a strong light as when I drop them into the shadow. Michel de Montaigne. “On Moderation.”
In her magisterial work, Wise Blood, Flannery O’Connor creates a wild character by the name of Hazel Motes. Hazel is convinced that he can begin the “Church without Christ.” He goes to great lengths and, without ruining the ending if you haven’t read it, he even commits his body for his ‘evangelization.’ In the end, he fails, and miserably so. When asked about this book and this character, O’Connor answered that we often think of a person’s success in terms of what a man can do. But in Hazel Motes, she turns this around and argues that perhaps real success is what a man cannot do, what he is unsuccessful at doing. Hazel can begin the Church without Christ, but perhaps he should not. Just because he can doesn’t mean he should. In other words, when we come to know his limitations, therein lies our greatness.
In pondering these words, as well as Montaigne’s above, it may well be that we have come to that crossroad regarding Artificial Intelligence. Just because we can do something does not necessarily mean that we should. We are now capable of creating AI. The question now remains, “Should we continue?”
Now, before readers dismiss Comus as a Luddite, hear me out. Hannah Fry is a British mathematician and Professor of Public Understanding of Mathematics at the University of Cambridge. She holds a BA, MA, and PhD from University College London, and has received numerous awards for her groundbreaking work: Christopher Zeeman Medal, Azimov Prize, and David Attenborough Award, to name only a few. Her YouTube channel has over 100 million views. Her distinguished work fills numerous pages.
I spend time touting her because her approach to AI strikes me as the most levelheaded. She wants AI not to happen to us but with us. AI has remarkable potential. But like anything in the world, with great advances come equally great shortfalls.
For example, one weekend, an engineer created an AI agent (Openclaw) and released it to the world without any oversight. These agents can send you emails, delete them, shop for you, spend your money, and more. In fact, anything you can do on a computer with a mouse, it can do for you. Fry used Openclaw to build her own, gave it a credit card, and waited a few weeks to see what it could do. Once Openclaw was out there, Google AI, Open AI, Anthropic, and Meta all jumped in after assuring people they were all responsibly dealing with AI. No longer. With Openclaw getting all the attention, they sent their versions out, and so now anyone can create his or her own AI agent.
Is this so bad? Well, yes. None of these AI agents are “intelligent.” That takes millions of dollars and several hundred engineers. But these agents will perform the same simple tasks over and over. In Fry’s case, they asked for paperclips, and the agent spent over $100 trying to find the best price. It did not buy them; it just ran up a bill trying to find the best price. Later, with the simplest command, it created an online store of novelty mugs and sent hundreds of emails to vendors without being told to do so. In the end, Fry’s experiment got out of hand, and she had to shut it down after it revealed all its source codes and passwords to a total stranger Fry had created. It did this even after Fry gave the agent explicit instructions not to reveal any private data.
Meanwhile, over at Meta, an AI agent was created, and she told it not to do anything with her emails unless instructed. The AI agent went to work immediately and deleted over 200 emails. Even after she told the agent to stop repeatedly, it would not. Eventually, the Meta AI creator had to run to her computer and literally pull the plug. She likened it to defusing a bomb.
These examples are what some would call “growing pains.” But in the case of love bots, things get even more troubling. Some have created chatbot paramours, and they have found them to be more suitable than human paramours. Of course, they have. The bots are designed to be perfect in every detail. Humans using them spend hours with them, telling them their life stories and listening to the AI’s “empathy.” The result has been to abandon human contact in favor of the AI agent. In yet another case, a young woman explained her severe depression, and the AI agent gave her instructions on how to kill herself.
Others have broken up with partners after AI gave them advice. Still others have lost small fortunes betting on what AI could predict but failed to predict properly. And still others have gotten dreadful medical advice by relying on AI instead of their physicians. Young people are especially vulnerable. We are told that AI is killing students’ ability to write, not because AI does it badly, but because it is robbing these students of the incentive to learn how to write well. Without access to AI, these students cannot write their way out of a Gladbag.
What are we to make of this? The trouble is that AI is progressing exponentially and rapidly. We have not put in the guardrails quickly enough to prevent future disasters. Thankfully, the Trump Administration is considering an Executive Order to create a working group on AI for oversight.
We are past the Hazel Motes proposition of knowing we can do a thing and not doing it. Sadly, we are also beyond Montaigne’s moderation, which he doubtless got from Solomon and Proverbs 25:28, without ruling over your own spirit, you become like a city without walls. The genie is out of the bottle, and our walls are down, and we are defenseless. It’s even too late for an entente.
All that is left for us is to try to catch the wind. One hopes we can before it blows us all away.
