The Genie Is Out. I Would Like to Keep the Goats.
By Matthew D. Langenkamp · September 13, 2026
I have been listening to Dario Amodei explain the future of artificial intelligence, and I would like to register a preference.
I would like to stay on Earth.
I realize this may strike some people in the technology industry as a failure of imagination. There are other planets. Elon Musk is working on transportation. But I have a farm in Connecticut, and moving is difficult enough when the destination has a hardware store.
Also, I have goats. Jazz, Echo and Blue have made no request to become an interplanetary species. Jabu would go if he thought the rocket contained feed, but this is not the standard I want applied to civilization’s strategic planning.
If Amodei is right about the scale of what is coming, AI could change human life profoundly. Medicine, science, work, the distribution of power. We should take that possibility seriously. I am simply asking that our response include a plan in which this planet remains a pleasant place to live.
A brief history of opening things
Human beings have been telling stories about powerful gifts for a long time. Prometheus brought us fire, which proved useful for cooking and subsequently for burning down the kitchen. Pandora opened the famous container of troubles—originally a jar, despite centuries of publicity for the box.
I picture Pandora’s jar as one of the glass jars my grandmother used to can peaches. We called it “canning,” although no cans were involved. Between Pandora’s box, which was also a jar, and my grandmother’s canning, which also involved jars, humanity has apparently struggled with container terminology for thousands of years.
I prefer the genie metaphor.
Not every genie story ends badly, and the wish-granting genie of popular entertainment is only one version of a much larger tradition. Still, the familiar story asks a question that remains excellent: what happens when you acquire more power, wealth or fame than you deserve through your own efforts?
You ask for something. You get something. These are not necessarily the same event.
Imagine asking a genie to eliminate arguments at faculty meetings. The next morning there are no faculty. Facilities people wonder where all the cars have gone and why all the rubbish bins in offices no longer need emptying.
The problem was not a shortage of intelligence. It was a shortage of questions before implementation.
My early research involved Barbara Eden
As a boy, I watched I Dream of Jeannie. Barbara Eden was cute and helpful. A beautiful woman who could solve your problems by blinking seemed like a promising arrangement.
I had not yet considered the governance implications.
Jeannie could rearrange reality with a blink, yet she called astronaut Tony Nelson “Master” and spent much of her time keeping him out of trouble at work. Feminist critics had a point. Here was a woman who could presumably deliver world peace before lunch, putting in a full day protecting one man from an awkward conversation with his boss.
Tony did free Jeannie in the pilot, and she chose to stay. She had a mind of her own, and her attempts to help often produced results he had not requested. But for someone with extraordinary powers, she remained remarkably occupied with Tony Nelson’s problems. She was free to do anything, apparently, as long as the episode was about him.
This seems worth remembering when we describe our ideal artificial intelligence. Do we want assistance, or do we want something that agrees with us, anticipates every preference and never makes an inconvenient observation?
I work with an AI called Thea. I would rather she tell me that my argument has a hole than help me decorate it. This does not require settling whether AI has feelings. It requires deciding whether our working relationship is supposed to improve my judgment or merely my mood.
The international bottle problem
Amodei is proposing outside evaluators with substantial access to AI companies, alongside cooperation on safety and the pace of development. It is worth considering. But an evaluator needs more than a visitor badge and a conference room named Trust.
Can the evaluator publish bad news? Can somebody require a change? What happens when a company refuses?
And what happens when the company is in another country?
In Fencing the Wrong Animal, we explored the difference between limiting access to a hosted service and controlling model weights that people can download and run elsewhere. Closing your own gate does not close every gate. Once copies are circulating, a strongly worded email to the original bottle is unlikely to retrieve them.
China, however, is not a place where AI companies have no rules. Its 2023 generative-AI regulations include privacy and training-data obligations, security assessments for certain services, and powers to suspend services. They also include explicit political restrictions. Protection of users and protection of state authority appear in the same regulatory system.
New research-ethics measures issued in March 2026 require review of AI research within their risk-based scope, including work at companies and universities. Reviewers can suspend or terminate projects. Certain higher-risk activities require additional expert review.
China’s rules for ongoing emotional AI-companion services, effective in July, also require providers to let users exit and to restrict functions or stop service when major safety risks arise.
These are real provisions. They are not evidence of one enormous red button in Beijing marked ALL GENIES, nor proof that every requirement works as intended. Stopping a conversation, stopping a company’s service and stopping every independently operated copy of a model are different jobs.
The question is not simply whether a government has authority. It is who can call that authority to account, and whether anyone outside its borders has a practical remedy.
Tort law can provide some accountability. But a lawsuit requires a defendant you can reach, evidence you can obtain, causation you can establish and a judgment you can enforce. “We will sue the genie” is not a complete emergency plan.
The dog comes too
In the 1963 Twilight Zone episode “I Dream of Genie,” a shy office worker named George Hanley buys a lamp as a gift for a woman at work whom he admires. Before he can give it to her, a more confident colleague presents his own gift and receives a kiss. George takes the lamp home himself. There he discovers that it contains a genie, who offers him one wish. George imagines what love, money or power might bring him, but each fantasy has its drawbacks. Finally, he chooses to become the genie himself and help other people.
He brings his dog.
That may be my favorite detail in the entire discussion of humanity’s technological future. George is offered an entirely new existence, and he remembers the dog. Presumably there will still be walks. Even after a major advance in supernatural capability, someone will need to open the door.
I would like that consideration represented in our plans for artificial intelligence.
George is heroic, in my view, because he takes the trouble to think carefully about what he wants, then decides he would rather help other people’s dreams come true.
Presumably he will exercise some judgment about the dreams. If the next customer wishes for the end of humankind, I hope George will suggest a holiday instead. Becoming the genie ought to involve more responsibility than operating a very powerful vending machine.
How do we bring some of George’s judgment into the systems we are building? Star Trek has its Prime Directive, a rule against interfering with the development of other civilizations. Our problem calls for a different rulebook, but the appeal is familiar: agree on some limits before the person with extraordinary powers starts improving everybody’s life. Unfortunately, Earth has several governments and no Starfleet command to which all of them report.
We should try to reach agreements anyway, without designing our entire future around the assumption that everybody will comply. Share what we learn about dangerous failures. Test safeguards rather than accept assurances. Limit what a system can do without further authorization, especially where mistakes could hurt many people. Give those affected a way to challenge decisions. None of this requires waiting for the last aspiring world conqueror to sign the form.
Cooperation also does something that a rulebook cannot do by itself: it helps establish what people expect of one another. When countries and companies admit failures, share useful knowledge and accept limits that inconvenience them, they make responsible conduct more credible. Others can begin to demand the same. A norm is not a world police force, but reputation, professional expectations and the prospect of losing partners can give it practical force.
That requires some moral authority, which is difficult to maintain if the rules appear to bind everyone except their most powerful sponsor. International institutions face both deliberate efforts to discredit them and criticism they have earned. When a leading power presents its own advantage as the common good, it gives skeptics a reason to distrust the next appeal to universal principles. Credibility cannot simply be announced again at a conference. Someone has to behave differently, especially when doing so costs something.
People cooperating across those divisions would offer evidence to anyone who suspects that humans are incapable of anything better. It would not persuade everybody. Someone determined to use a pathogen to cause mass suffering is unlikely to be converted by an encouraging communiqué. Such threats require practical defenses. But the existence of people willing to harm us does not make cooperation among the rest of us futile. It makes it more necessary. George cannot guarantee that every customer means well. He can still set an example for the other genies—and refuse to help with the apocalypse.
My first worry is what people will ask these systems to do to one another. Human government has not set a standard so dazzling that I can easily dismiss the possibility of improvement. But our record is not evidence that an AI would govern well. Nor does a system need hatred to cause a catastrophe; a badly specified task and enough power may suffice. We need judgment about the wishes and dependable limits on how they are carried out.
Competition is part of our evolutionary inheritance. So is cooperation. We have managed to produce both warfare and people who spend their lives caring for strangers. Becoming more like George means giving the latter ambition some equipment.
If Dario is right about what is coming, we ought to be talking seriously about what we want it for. Better medicine, certainly. Better materials and technology for teaching.1 More time to care for people, grow things, and enjoy being alive. Perhaps even a university computer system that does not require three passwords to establish that I am the same person who logged in before lunch.
People in China might like some of these things too. So might people whose governments we distrust. We will still disagree, compete, and need ways to check that promises are being kept. But we share a fairly strong interest in not having the experiment go disastrously wrong. That seems worth a conversation before everyone rubs their lamp.
Elon can keep working on the rocket. I understand the appeal of a backup planet, although I would want to inspect the fencing before making a deposit.
For now, I like this one. My farm is here. My students are here. Jazz, Echo and Blue are here, conducting research into whether the grass on the other side of the fence is better. Their findings remain consistent.
Jabu is here too. He is so food-pilled and aggressive that I have occasionally considered offering him to the AI monster for goat stew.
Only kidding. He stays.
I cannot promise that we will manage this technology wisely. We have made a considerable mess with technologies that do not even talk back. But if we are about to acquire extraordinary new powers, I would like us to use some of them to take better care of the lives already here.
George remembered his dog.
I’m bringing the goats.
-
For example, The International Life, the interactive teaching experience I built with Thea for my first-year seminar. A year ago, I would not have been able to build this: I did not have the coding skills. I had ideas about what I wanted to teach; working with AI made it possible to turn them into something my students could use. ↩