Things seemed less dire as this week began. Trump walked back a few steps from the edge of the gangplank he hastily, clumsily and stupidly built for himself. World trade, the dollar and our investment accounts won't collapse it seems, at least not before the All Star Game in July.
Joy is fleeting, and by late afternoon this past Monday I turned some shade of blue after reading a piece about what AI will soon do to our lives. Soon, as in very soon. It was written by an economist and someone who works for an AI company called Anthropic, a start-up responsible for Claude, an AI competitor of ChatGPT and Google's Gemini.
It's full of plausible and scary stuff, predicting how AI will change work, leisure, healthcare and status. In our soon-to-be AI-controlled world, machines will be the smartest people in the room. "Since we will all be ranked below some other entity on intelligence [these entities being the ChatGPTs and Geminis of the future], we'll need to find new and different outlets for status competition." Just what we need.
Tyler Cowen says in that article that AI may be very good at answering questions but has limited capacity in formulating them. His claim is that we’ll always need people like professors, or management consultants, to pose the right questions, which the AI machines can then answer. A friend of mine (a retired management consultant) raised the important point that while AI will proliferate, so will any number of anti-AI programs and software.
Curious about Cowen’s claim, I asked ChatGPT: "What's the most important question to consider when analyzing international balance of payments?" Here's the reply I received.
The most important question to consider when analyzing the international balance of payments (BoP) is whether the current balance of payments position is sustainable in the long run?
Why this question matters:
Sustainability speaks to whether a country can continue financing its current account deficit (or surplus) without leading to economic instability.
It encapsulates concerns about:
Foreign reserves: Are they being depleted or growing?
Capital flows: Are inflows (FDI [foreign direct investment] portfolio investment) stable or speculative?
Exchange rates: Is the currency under pressure to depreciate or appreciate?
Debt accumulation: Is the country relying too much on external borrowing?
Supporting Questions:
To assess this central issue, analysts often explore:
What are the main drivers of the current account balance (e.g., trade, income, transfers)?
Is the capital account financing the current account effectively?
Are there signs of external imbalances leading to currency crises or inflation?
How is the government managing macroeconomic policies in response to the BoP?
I'm no economist. There may be more salient questions to ask about the balance of payments, but AI today appears quite capable of asking some that address the issue in non-trivial ways. If the questions AI poses today are not entirely on point, it's likely that the future iterations of the software, available in the coming months (not years) will become more and more on point.
There may be some good news in all this for blue collar folks. Will AI plumbing and heating happen? Assume the MIT Robotics Lab designs an HVAC repair machine that can crawl into your basement, take apart your furnace, diagnose its problem and fix it with spare parts. A very big assumption. An iPhone has around 1,000 parts. How many will that plumber require? And what will it cost to make? Once you’ve perfected the prototype, you’ll need an assembly line to produce that robot in large quantities. How many buyers will there be for a robotic plumber?
There’s a department at MIT, CSAIL (Computer Science & Artificial Intelligence Laboratory), and one of the professors there has been studying sperm whales, tracking them to record their vocalizations to learn whether there’s something like a language they use to communicate with. They’ve attached sensors to the whales, but no robot is yet following them. The Lab is developing a soft robotic fish “that can swim in three dimensions to continuously record the aquatic life it is following or engaging.” From which I conclude that developing and building a robot that can perform complex tasks is a tall and very expensive order. And will remain so for the foreseeable future.
AI gardeners won't happen, even though you can buy an AI mower, just as you can an AI vacuum cleaner. We've had an intelligent vacuum cleaner, an Eufy ,for five years now. It has performed its assigned task with a grade no better than a C on a good day. We haven't used it for several years now. A late model Eufy costs $750. I have no plans to upgrade.
A friend recently had his hip replaced at a hospital in Boston. In the pre-op interview, the surgeon told him that while he would certainly be in the operating room for the procedure and in command, much of the actual surgery would be performed by a “robotic arm,” able to work at greater levels of precision than the human eye. This seems an advancement, as greater surgical precision must be a net social good. Fine, but one still needs to know whether the benefits are worth the costs. There’s no such thing as a cheap robot. My wife had two hips replaced in recent decades with no robots involved and has no complaints.
AI won't take over our leisure time, either. A dinner in a good restaurant, or a night at the theatre won’t be replaced by machines….ever. McDonald's can and will hire AI cooks and AI food wrappers to make and deliver your food. There may well be an audience that’s eager to buy tickets to a rock concert performed by robots. I might even do that, for the novelty of the experience.
My optimism about culture and creativity, however, may be misplaced. I may be dead wrong about robotic plumbers and living in ostrich sand about all the other looming, soon-to-be realities than AI is bringing our way.
That link which you may have ignored, attached to the word misplaced, is to an essay published just a few days ago by people who have worked in AI for a quite a number of years: AI2027. Ross Douthat of The Times this week published his conversation with one of its authors. The byline is: The Forecast for 2027? Total AI Domination. (I would have re-positioned that question mark, by the way).
Douthat’s conversation is a Must Read. It’s considerably more disturbing than the article in The Free Press I mentioned earlier. Robotic plumbers may be just around the corner, but they may be the least of our problems.
As these machines expand their capabilities, AI2027 argues they’ll soon be able to train themselves to complete all manner of tasks; to ask and answer questions that the machines themselves formulate. This is not artificial intelligence, but superintelligence, as it’s called.
In one version of where this all ends up, the superintelligent machines decide to kill all the humans.
I didn't make that up; it's discussed in AI2027.
That sounds like a movie script, but it may be more than that.
Now on to lighter fare. I asked ChatGPT to write a story with the prompt: "It was the best of times. It was the worst of times." (Last month I asked it to write one that began with: “I woke up this morning and a zebra was in my bed.”)
ChatGPT replied with: “Sure! Here’s an original short story inspired by the iconic opening from A Tale of Two Cities, exploring the duality of joy and despair in a modern setting:
The Mirror Season
It was the best of times. Ava stood on the stage beneath a canopy of lights, her cello resting against her shoulder like an old friend. Applause surged around her like a tidal wave, sweeping away the nervousness that had haunted her for weeks. She had made it—her first solo performance with the national orchestra, a dream realized at last. Tears sparkled in her eyes, hidden by the glittering spotlight. Her parents stood in the front row, faces beaming with pride.
It was the worst of times. Just a few hours earlier, Ava had sat in a hospital corridor, watching the life drain from her grandmother’s face through a pane of glass. The doctors said she held on longer than anyone expected—long enough to hear Ava’s recording played from a phone speaker, the same piece she had just performed on stage. Her grandmother’s hand twitched in recognition. Then it went still.
Outside the concert hall, the city buzzed with contradictions. Lovers kissed under streetlamps while sirens wailed down alleys. A child laughed at a magician’s trick as a homeless man wept behind a row of trash bins. In the center of it all, Ava walked alone, bouquet in hand, her heart divided—half blooming with triumph, half crumbling under grief.
Later that night, she opened her grandmother’s last letter, left unopened on her kitchen counter. It read:
My darling girl, some seasons will feel like the sun and storm are dancing over you at once. Do not run from either. That’s where the music lives.
Ava set the letter down gently, then turned toward her cello. She began to play—not for applause, not for the world, but for the memory of joy, for the ache of absence, for the strange and sacred place where both belong.”
After that famous first sentence, Dickens wrote :"It was the age of wisdom, it was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of Light, it was the season of Darkness, it was the spring of hope, it was the winter of despair, we had everything before us, we had nothing before us, we were all going direct to Heaven, we were all going direct the other way."
Thus it's not surprising that the AI machine picked up on duality as the core idea to replicate for my story.
However, A Tale of Two Cities, as you may remember from 9th grade English, is a story about darkness and light, or dualities, in only the most simplistic, reductive sense. Lucie Manette, Charles Darnay and Sydney Carton make up one of the extraordinary triangles in English literature. Dickens puts them in difficult situations from which they must extricate themselves or suffer dire consequences. No machine could have written that, and machines that try to copy it do a very poor job of it. e.g. The Mirror Season
Othello/Iago/Desdemona, or the Bennet sisters and Darcy in Pride in Prejudice are other stories, triangular or larger in their interlocking connections, that a machine will find it impossible to improve upon. The machine can transpose plots, character conflicts and situate them in new settings, but as The Mirror Season shows the results reek of the formulaic. It's the kind of story you might read in a magazine, in the old days, when waiting for your hairdresser or dentist. There's little evidence I’ve seen that AI storytelling will have that creative spark, that capacity to imagine new configurations, scenarios, plots and characters which all good stories require.
There are, however, two problems with my proclamation about creativity. One is that much, maybe most, so-called art today is a commercial undertaking, requiring finance. The financial decision could well be that while AI doesn’t produce great or memorable stuff, its products are good enough.
You don’t have to look hard to find support for this. Many of the big grossing movies in recent decades are recycled; they’re sequels or prequels of comic book plots, or adventures of super-heroes—-all of which can be cloned and transposed by AI script writing until the Messiah comes. The Writers’ Guild will object, as they have.
Robert Benton, who died this week at 92, and won two Oscars for screenplays (and should have won a third for his work on Bonnie and Clyde), was by all accounts a creative genius. (If you’re unfamiliar with his work, watch Kramer versus Kramer, winner of the Oscar for best picture in 1980). Hollywood today may have some interest in hiring the next Robert Benton, but as a business AI script writing is very appealing to the studios. Think of all those episodes of CSI, Chicago Fire, or Law and Order that AI could have written.
The second problem with my proclamation is that it may be no more than wishful thinking.
Tolstoy, as good a storyteller as the world has ever produced, believed there are only two basic stories: a man goes on a journey and a stranger comes to town. Others who have studied Jung and the role of archetypes came to similar conclusions, such as there are only seven basic plots—-not just in storytelling, but in history as well.
If there are only two or seven stories to be told, an AI machine is all we need here.
There's a grain of truth, maybe several, to these reductionist claims.
You could say that King Lear is about a man who goes on a journey. But if that was all you wrote about the play in your term paper, your teacher should give you a failing grade. You’d get a C,, maybe, if your paper focused on elder-abuse in Lear, but to get an A you’d need to immerse yourself in Shakespeare’s language, where he reveals what drives his characters to act as they do. And that language and those characters are not replicable or replaceable by a machine. Logan Roy from Succession is not Lear.
Why is that true? Tolstoy gave us the answer with the so-called Anna Karenina Principle: “All happy families are alike; each unhappy family is unhappy in its own way.”
By the way, if there are only two or seven plots available to choose from, which one applies to that situation in which those superintelligent machines kill all of us?