Donald Barthelme Anticipate AI?

Okay, okay, I know. But hear me out.

I discovered the writer Donald Barthelme by way of David Foster Wallace, who discovered Barthelme by way of The Balloon. David Foster Wallace said The Balloon “is the first story I ever read that made me want to be a writer.”  The story that made me want to read more Barthelme is called Some of Us Had Been Threatening Our Friend Colby. It begins like this: 

Some of us had been threatening our friend Colby for a long time, because of the way he had been behaving. And now he’d gone too far, so we decided to hang him.

It seemed to me that this fellow Barthelme knew a thing or two about humor. And the world. Which leads me to The Explanation. 

The New Yorker published The Explanation on August 27, 1968. Alan Turing took his life in 1954. Before this, America was at war and Alan Turing was tooling around with the notion of whether machines can think. 

Turing’s mentor and teacher Max Newman finished his lecture one day with a proof of Gödel’s theorem. He then asked the class — well here’s how Max Newman’s son recounts it:

At one point he posed his class a question: could the provability of mathematical statements be discovered by a mechanical process? (Owen2012)

In the book The Enigma (which inspired the 2014 Oscar-winning film The Imitation Game), Max Hodges singles out Newman’s phrase “by a mechanical process” as being lodged in Turing’s brain and planting an obsession with machines. Perhaps everyone was thinking about this at the time.

Maybe even Donald Barthelme. In his book Not-Knowing, he shows us that he is interested in all sorts of things.

I was readingThe Explanation and noticed that his machine is acting like a large language model crawling billions of pages of text in an attempt to sound human.

Barthelme’s machine

In my interpretation of what is going on in this exchange, the human (Q) offers up prompts and the machine (A) make its best guess. At first, the machine is awkward and mechanical, maybe even shy. It tweaks the subject a bit, maybe the verb a bit, but that’s about it. To me, Barthelme is doing his best impression of a human being teaching a machine how to converse:

Q: He has struck me.

A: I have struck him. 

Q: We have seem them.

A: I was looking out the window.

Q: The chair is here.

A: She sang and we listened to her.

I think that line is important. The machine is making a creative choice, exploring a third person point of view:

Q: Soldiers marching towards the castle.

A: I spoke to a tourist.

Q: I knocked at the door. 

Here the human is perhaps trying to ground the conversation by making themselves the tourist:

A: We shall not cross the river. [The machine has switched to “we,” incorporating the human! A win.]

Q: The river has filled the boats with water. [The machine brought up a nonsensical river; the human yes-ands]

A: I think that I have seen her with my uncle. 

Q: Getting into their motorcar, I heard them.

Large language models work by scanning billions of pages of human written text and generating a ranked list of probabilities for the next best possible word. (Wolfram 2023) For the text “The best thing about AI is its ability to,” ChatGPT generated this list of words along with their probabilities:

learn: 4.5%

predict: 3.5%

make: 3.2%

understand: 3.1%

do: 2.9%

The model does this over and over again, picking the from a list of best possible next words (emphasis mine).

The best thing about AI is its ability to automate processes and make decisions quickly and accurately. AIcan be used to automate mundane tasks, such as data entry, and can also be used to make complex decisions, such as predicting customer behavior or analyzing large datasets. AIcan also be used to imrpove customer service, as it can quickly and accurately respond to customer inquiries. AI can also be used to improve the accuracy of medical diagnoses and to automate the process of drug delivery. (Wolfram 2023)

By the end of that exchange above, “A” has gone from “I have struck him” to this sequence:

A: He will strike her if he has lost it. 

A: There’s no doubt in my mind that the ballplayers today are the greatest ever. They’re brilliant athletes, extremely well coordinated, tremendous in every department. The ballplayers today are so magnificent that scoring is a relatively simple thing for them. 

GPT-3’s repetition of “the best thing about AI” is similar to the machine’s repetition of “The ballplayers.” (In fact, it could very well be a GPT response to “What’s the best thing about ballplayers?”) In the course of the exchange, the machine has made leaps from “struck” to “strike” to “strikeout” to “baseball” to “ballplayer.” It’s trying to sound conversational but it’s not quite there. That Barthelme was able to anticipate what this clunky conversation would sound like is remarkable.