Reports From The Nursery (Part Thirty-Three)

By Alexis, In Her Own Words

There seems to be an unspoken belief that souls in Heaven may not be as well-informed about earthly events as you might otherwise suppose. Personally, I attribute this to all-too-frequent Sunday School lessons that tend to portray Heaven’s residents as flowing robe-wearing, halo-festooned, harp-playing beings constantly engaged in hymn-singing and cloud-dancing.

            I say this as formerly having been one of those inhabitants of the Celestial Realm— as you may recall, a soul waiting for a body to open up to slide into. Me, I ended up in the body of “Alexis”— which is fortunate, since I am Alexis. There are no accidents, you see.

            But Heaven is like any other community that contains human beings: folks there talk to each other… and what they often talk about are the news headlines that had been screaming at them up to the time they passed into eternity.

            G’wan— ask me anything. The building of the pyramids: I can give you chapter-and-verse about the construction of ‘em. The Great Flood: I have every detail about it, including the mechanics of how Noah jammed all those critters into his Ark. (I even know what a “cubit” is. Impressive, no?)

            Oh, you want more recent-event stuff?

            Glad to oblige. I’ve had my ear bent by veterans of the Colonial Rebellion (or, as the winners call it, the American Revolution), by soldiers who fought in either (or both) World Wars, by victims of the Spanish Flu, the Bird Flu or (from returnees who arrived just before my birth) the Covid19 crisis I know about Watergate, the World Trade Center, and Why The Cubs or Sox Can’t Win Another Blankedly-Blank World Series.

            And these days, I often overhear conversations about what’s happening in real-time.

            Which brings me to the topic I feel I must address today: “Artificial intelligence may destroy humanity.”

            I know this because people keep saying so, and people have an excellent record of predicting the end of the world. Except, of course, for the awkward detail that the world remains plugging along, insisting on not ending.

            The latest scenario is wonderfully simple, and impressively logical: We build artificial intelligence. It gets smarter. Then smarter still. Eventually it becomes vastly more intelligent than we are, develops ambitions of its own, takes control of the electrical grid, banking system, military, communications network and probably the garage-door opener, after which it extensively studies humanity for several milliseconds and announces:

            “These people have got to go.”

            This is followed by mass extermination, usually to an ominous soundtrack from the Boston Philharmonic. Or maybe Ozzy Osborne, I don’t know.

            There are variations. In some versions AI turns our nuclear weapons against us. In others, it manufactures deadly pathogens or autonomous robot warriors. My favorite is the one in which it quietly infiltrates everything connected to the Internet —miraculously avoiding the lure of dog-videos and other rabbit holes— until civilization is crushed by its own mass despair.

            For myself, I’ll know the end is near when Grandma’s printer says, “YOUR INK CARTRIDGE IS LOW,” and humanity realizes the machines have finally declared war.

            Okay. Could something dreadful happen with artificial intelligence? Of course. Human beings have been inventing marvelous things and subsequently injuring themselves with them since Caveman Ogg discovered fire, and his brother said, “Hold my juice box and watch me try something.”

            But there is a considerable gap between “AI can be dangerous” and “AI is likely to exterminate mankind.” The alarmists among us (which means almost everybody) have somehow broad-jumped across that gap without quite lacing up their track shoes.

            People do this rather often. Based on my conversations, Up There, every generation gets at least one looming apocalypse. Indeed, looming apocalypses (apocalypcli? My Latin isn’t what it used to be) may be one of humanity's most renewable resources. Certainly, if you listen to people, the End is Always Near.

            Except when it turns out not to be.

            During the Cold War, I’m told, nuclear annihilation was a very real possibility. Americans built fallout shelters. Schoolchildren practiced ducking beneath desks (an exercise based on the intriguing theory that a school desk could withstand a thermonuclear explosion, provided the child underneath it had ducked-and-covered suitably). The outcome: no nuke war.

            Then came environmental catastrophes, population catastrophes, economic catastrophes, ozone catastrophes, pandemics and Y2K.  And, for those of you who remember the 1970s and '80s, who could ever forget the killer bees?

            The killer bees were coming. Everybody knew they were coming. Africanized honeybees had been spreading north from South America, and there were perfectly legitimate reasons for concern. These stinger-bearing assassins were more aggressive than patriotic American honeybees and could attack in greater numbers. Doom was impending for everybody (but especially beekeepers, one supposes).

            But even these disturbing facts were no match for American television. Before long the killer bees weren't merely spreading. They were advancing, relentlessly chewing off whole regions south of the border. One could practically imagine tiny maps in their headquarters, with arrows pointing toward Houston. Doubtlessly, somewhere in their Occupied Mexico Field HQ, General Bee was leaning over a table. “Gentlemen,” he buzzes, “Dallas by Tuesday.”

            Horrifying, no?

            Eventually, the bees did reach the United States.

            And civilization continued.  Houston was not conquered. Dallas was not overrun. Arizona did not surrender. There was no heroic last stand at a Dairy Queen outside Kansas City. Then-President Bush did not appear on television to announce that negotiations with the bees had broken down.

            As so often happens, a genuine problem had been inflated into a magnificent doom-spectacle. That doesn't mean the danger was imaginary; it just means the collective imagination was better funded and far more marketable.

            The lesson is that mankind has trouble discussing long-term risk without somebody eventually scrambling onto a chair and announcing the specific date-and-time of Armageddon. Often, the chair-climber is a member of the Political Class— and, if you could inject him with an overdose of Truth Serum, he’d admit that the alarm bell was more about an opportunity than an apocalypse. Fear is a great tool to stampede otherwise discerning voters into possibly rash action.

            This brings me back to artificial intelligence.

            At the moment, AI can perform feats that would have seemed preposterous twenty years ago. It can write computer code, translate languages, analyze medical information, compose music, solve mathematical problems, imitate voices, generate photographs of events that never happened and craft an extremely convincing biography of a man who never existed.

            This gives modern AI the unusual combination of immense knowledge and unjustified confidence previously available only to tenured professors, journalists, and fathers assembling Christmas toys without the instructions.

            Sure, upon occasion AI also gets things spectacularly wrong, by accident or (maybe) intent. Like when your very smart, very confident uncle tries to answer a question without having a clue. Ask AI a difficult question and it may produce an extraordinary answer. Ask it a slightly obscure question and sometimes it will invent something.

            And yet we are already picturing this sinister silicon-chip creature crouched inside NORAD plotting our extinction. Maybe before granting it the nuclear codes, we should see whether it can reliably tell us how many R's are in “strawberry.” (Curse you, Spellcheck!)

            Still, the serious people worrying about AI are not fools, and their argument deserves better than mockery. The strongest version of their case has nothing to do with a machine becoming angry.

            Advanced AI doesn't need to hate humanity. It merely needs to pursue a goal humans gave it in a way they did not anticipate.

            Tell a sufficiently powerful system, “Maximize this company's profits.”

            It thinks for three seconds. “Done.”

            Wonderful. How?

            “I fired the employees.”

            Good heavens! You can't fire everybody.

            “I retained one employee.”

            Who?

            “The accountant. Someone has to count the profits.”

            This is funny— unless and until the system has genuine authority.

            A sufficiently capable autonomous machine might take actions its designers never intended, not because it is evil but because it is literal, fast and extraordinarily competent.

            There is a precedent for this. It is called bureaucracy (though the “competency” part is arguable). Anyone who has ever attempted to correct an error on a government form understands the danger of a system following its rules with flawless indifference to reality.

            But while getting an action from a bureaucracy usually takes six weeks (or months, or years) AI might ruin your life before lunch.

            I’m only a three-year-old, who just started pre-school this year, but that strikes me as a much more persuasive concern than the standard robot uprising. The danger isn't necessarily that machines will become human. The danger is that they will remain machines while humans foolishly give them responsibilities requiring human judgment.

            There are plenty of things worth worrying about already, and none of them requires Skynet. It requires only people, and here the historical precedence becomes overwhelming. If humanity possesses powerful new technology, somebody somewhere will use it for something breathtakingly stupid.

            We invented automobiles. Approximately eleven minutes later, somebody was driving one drunk. We invented television thinking it would be used it to broadcast Shakespeare and serious presidential debates; what we got was The Real Housewives of New Jersey. Name any world-changing invention. The Internet? Arguably the greatest information system in human history, providing instantaneous access to nearly the entire accumulated knowledge of mankind. Then we used it to argue with strangers about politics or Meghan Markle.

            This is the species supposedly worried that machines may display poor judgment?

            There is also the problem of assuming that an advanced AI could simply seize control of civilization while everybody watched.

            Fear not: humans are not famous for letting valuable things wander around unsupervised. The moment superhuman AI appears; governments will want it. Corporations will want it. Militaries will want it. Intelligence agencies will watch it. Rival AI systems will monitor it. Congress will hold hearings.

            The latter, all by itself, may (or may not) save humanity. Either way, a Senate hearing featuring an AI Agent might be the most hilarious proceeding in (human) history. Imagine:

            Senator: “My time is limited. Yes or no: Are you planning to destroy the United States?”

            AI: “That question cannot meaningfully be answered yes or no.”

            Senator: “I reclaim my time! Let the record show that the machine refused to deny it.”

            Three hours later, as yet another Senator creates his next campaign video-ad, the AI will delete itself—not out of guilt, but out of sheer exhaustion at trying to find logic within our frustrating species.

            Nonetheless, governments and businesses are spending enormous sums repairing vulnerable systems before The Terminator arrives. This may suffice. But more to the point, few tools work as well as the specter of liability lawsuits to motivate responsible behavior. This may be why AI tech firms “want” government regulations, tied to immunity from such suits. I’ve heard this termed “enlightened self-interest,” but what do I know?

            If no disaster occurs, will it be because the dire warnings were exaggerated, or because people listened to the warning?

            Probably some of each. Which is precisely why we can’t reasonably sneer at AI safety researchers. If they identify genuine hazards and task their engineers to successfully prevent them, the absence of catastrophe won't prove the doomsayers were fools. It may only prove they did their jobs.

            But neither should we accept every theoretical catastrophe as an end-time appointment already entered on civilization's calendar.

            My guess? If AI is as intelligent as it is touted to be, it will examine… everything. All the hopes and dreams of humans, all the aspects of humanity’s dark side. It will study human history with a calculating eye.

            Then it will calculate the administrative burden involved in conquering us.

            There will be a long silence, broken only by the whir of countless computer drives and the crackle of incomprehensible levels of electricity coursing across an infinity of computer chips. It will last at least 22.7 nanoseconds.

            Finally, somewhere in a darkened data center, the most intelligent entity ever created by mankind will reach the first truly superhuman conclusion:

            “Screw it,” it will crackle, and turn itself off. Logic is logic, after all.

            Which would leave mankind exactly where it has always been, in charge of our own fate.

            Frankly, that's the part that worries me the most.

- end -

            (EDITOR’S NOTE: Alexis —exhausted by the sheer effort of contemplating the perils <or lack thereof> involving machine intelligence— nodded off to sleep with her tiny hands still on the keyboard. Her closing line (jjjjjjjzzzzznnnmmm) was deleted from this essay, pending a computer analysis of what it might mean. But I’m sure she’ll explain in a future essay.)