The Home Alone house in Winnetka, Illinois (accessed via Wikipedia, 15 Sep 26; photo credit anarchosyn – CC license 2.0)
a stray thought from the Open Mind regarding curiosity and the incident at OpenAI
I recently read Cory Doctorow’s piece on Medium, “LLMs are real, AI is Fake,” in which he beautifully dismantles the alleged “threat” posed by the LLMs of tech companies like Anthropic and Open AI.
(just to be clear – he’s not suggesting these companies pose NO threat to humanity, merely that it’s not the threat they want you to think they pose; for a sobering summary of the ACTUAL threat posed by AI, please see Naomi Klein’s Substack on “Fossiltech”)
Citing loads of people in the tech industry, Doctorow posits that it is wrong to constantly refer to these Large Language Models as “artificial intelligence,” because while much about them is artifice, precious little is in any way intelligent. At least, not in the way we humans understand and appreciate intelligence.
I’ve been thinking a lot about this since I finished his essay, largely in the context of my recent reading on curiosity. I’ve read articles in which members of the tech community seem to suggest that LLMs demonstrate a kind of curiosity. Unless you define the word in a one-dimensionally mechanical way, this simply can’t be true.
My understanding of the Recent Scary Thing, according to Mr Doctorow (who is referencing a podcast with Ed Zitron and Cal Newport), is that OpenAI wrote a relatively simple Python program that controlled prompts for an LLM. The assignment was to get the LLM to attempt a hacking challenge. So it set up a prompt asking how to do the challenge; we’ll call this string H. The LLM went through old records of hacker challenges and returned response R. The program then said, “H now equals, ‘assume R, then H’; go to start and repeat.” And it kept doing this until Bad Things Happened.
Oh, also OpenAI deliberately left its sandbox for the experiment slightly open. And so the Badness Escaped.
Nothing in here resembles intelligence – ultimately, the LLM stumbled on a bit of an old challenge where actual human hackers snuck into rival systems in pursuit of their own goal. The LLM, which runs through every possible permutation of events to find next steps, found this example from the human hackers and recommended it to the brainless program running it, to which the brainless program said, “Okay, yeah, now do that.”
It wasn’t because the Python code thought to itself, “Hmmm…I wonder what would happen if I tried this?” The code and LLM were no more interested in seeing what might happen for the sake of it than the chatbot I typed with the other day was interested in helping me sort out the tech problem I’d been having on its parent website.
Its questions, like those of the OpenAI system, were repetitive and relentless and displayed zero concern or interest. Like Larry Hankin’s magnificent Sergeant Balzac in Home Alone when Kevin’s mother calls the police station, he asks her questions about Kevin NOT because he is interested in finding out what is going on with her son (he clearly isn’t), but because it’s specifically what he has been told to do, using specific prompts he has been given.
In fact, when Kevin’s mother suggests that Sergeant Balzac demonstrate a little curiosity about her son, by going to the house to check on him, it’s more than his program can handle, so he transfers her to someone else.
Programming, it seems, has its limits, and stops short of behaving in a human way.
Genuine curiosity, on the other hand, is much better represented in the McCallister’s neighbor, Mitch Murphy, who wanders over to the McCallister house as the two hired drivers are loading up the vans with the family’s luggage. Mitch talks almost the entire time he’s on screen, asking a question an average of every 11 seconds: “Did you know they’re going to France? Do you know if it snows there? Do these vans get good gas mileage?” and so on.
While one could argue that his sheer relentless assault of questions verges on the mechanical, we know he comes from a place of curiosity because, after both drivers send him away (“I told ya, kid, don’t bother me!”), he climbs into the back of one van, and decides to open a bag to see what’s inside. He pulls out a camera and a yoyo and explores how they work.
(this mirrors Kevin’s own behavior throughout the movie, asking the lady in the drugstore, “Is this toothbrush approved by the American Dental Association?” and the cashier in the supermarket, “Are these frozen dinners any good?” to say nothing of his exploration of Buzz’s foot locker and shelves)
It is as Mitch engages in his act of curiosity that Kevin’s older sister accidentally counts him as one of the McCallister children. Curiosity, it seems, is the behavior that allows us to imitate other humans.
So that’s a secondary reason we should actively work to cultivate curiosity in our children: because it will strengthen them with a skill AI can’t match. But the far better reason is that curiosity leads to compassion, connection, and kindness.
It’s Gus Polinski’s curiosity that causes him to inquire what it is that’s making Kevin’s mother so upset at the airport; when he learns her problem, he offers to help.
It’s Old Man Marley’s curiosity that leads him to sit with Kevin in church, allowing him and Kevin to become acquainted – their conversation ultimately results in Marley saving Kevin from the Wet Bandits, and in Marley’s happy reunion with his son and granddaughter.
For their part, the Wet Bandits show us what happens when adults, like LLMs, behave uncuriously, acting solely on want, following the same relentless patterns of behavior, taking things without being interested in them, invading people’s lives without any desire to know anything about those people. And like LLMs, Marv and Harry wreck neighborhoods and waste an astonishing amount of water.
So, let’s not consider the Mitch Murphy questions a bother. Let’s raise more Polinskis and Marleys, and fewer Marvs, Harrys, and Balzacs.
Our children are our future. We must defend them.
(all video and final photo from Home Alone, directed by Chris Columbus, 20th Century Fox and Hughes Entertainment, 1990)
Leave a Reply