“A Bird Came Down the Walk”

David Morton

2 years ago

I recently tried my hand at OpenAI’s ChatGPT, the artificial intelligence that has taken the world by storm.

The irony was rich: I still make bread in a wood-fired oven. I guess I’m a holdout for the old ways.

At first, I had an engaging conversation with it about living on Pender Island (it knew a lot about our island home), about making life decisions, the Stanley Cup Playoffs (it picked the Florida Panthers to win the cup, and it was correct) and what kind of bread it prefers. Although the AI admitted it didn’t have personal preferences, it said many people enjoyed sourdough bread for “its tangy flavour and chewy texture.”

Smart answer!

It also knew quite a bit about The Happy Monk Baker, including some of the most recent breads offered to customers.

ChatGPT, please write a blog post

I was flattered but doubtful of its use for everyday purposes. When I asked ChatGPT to write a thousand-word blog post about living and baking bread on Pender Island, the result was awful. It took 15 seconds to generate the piece, but it was bland and formulaic.

My son John uses ChatGPT extensively for his work. He says I need to spend more time with the platform and that it can be “trained” to get closer to my writing style and tone. Given time, he suggests, I’d have a much closer version of a Happy Monk blog post.

But I could also write it myself!

John’s a big believer in AI and where it’s going. It’s here to stay, he says, and we might as well get used to it. He’s convinced me to keep an open mind. But it will be a while before I’m all in for this new-fangled AI stuff.

A titanic struggle with AlphaGo

Last year, I wrote a piece about artificial intelligence, specifically about how an AI called AlphaGo defeated the world’s best Go player, the Korean champion Lee Sedol. 1

In 2016, he and “the machine” played a tournament of five Go matches. AlphaGo won four of the games, defeating Lee Sedol soundly.

Before the tournament, Lee framed the contest in grandiose terms. He said it was important for the human race:

“I am not just playing for myself and my country, but for all of humanity,” he said. “I am going to do my best to protect human intelligence.”

In a video documentary about the competition, the turning point came during the second game—move 37, to be exact. Lee had left the table for a cigarette break.

Move 37

The AI placed a stone on an unexpected part of the board in his absence. The move drew gasps from the spectators, and the commentators were flummoxed. Some said it was indeed a mistake.

Gradually, though, watchers came to see the move as decisive. Unexpectedly, it closed off Lee’s options for responding. Some said they could see no escape for Lee.

On his return, Lee sits down and beholds the board. He winces! You can see dark clouds gathering around him. He fidgets; his brow is tense. He hesitates, makes several false starts, and then places his countering move. But the course is set, and he can’t recover the advantage he thought he had. He soon withdraws from the game. 2

Meanwhile, commentators and AI developers praised the move’s genius. One said, “Not a single human player would have chosen move 37!”

Gracious in defeat

Lee apologizes to the audience afterward but is gracious in defeat. He congratulates the team that developed the AI and vows to play better next time. He’s shaken, near tears, it seems.

In 2017, the AlphaGo developers introduced an updated version, one they said was many times more powerful. To test it, they set two computers to face off against each other. One ran the version that defeated Lee Sedol; the other ran the more powerful version, AlphaGo Zero.

AlphaGo Zero defeated the older one 100 games to 0. In 2019, three years after his defeat, Lee Sedol resigned from the Go competition.

“With the debut of AI in Go games,” Lee said, “I realize that I’m not at the top, even if I become number one through frantic efforts. Even if I become the number one, there is an entity that cannot be defeated.”

Is there a human in there?

As I played with ChatGPT, I did what I could to find anything human lurking beneath its powerful computational ability. I asked if it could recite a poem.

“Certainly!” it said. “What topic would you like the poem to be about?”

I said I would like a poem about nature.

Here’s what ChatGPT offered:

A Bird, came down the Walk—
He did not know I saw—
He bit an Angleworm in halves
And ate the fellow, raw,
And then, he drank a Dew
From a convenient Grass—
And then hopped sidewise to the Wall
To let a Beetle pass—

He glanced with rapid eyes,
That hurried all around—
They looked like frightened Beads, I thought,
He stirred his Velvet Head.

Like one in danger, Cautious,
I offered him a Crumb,
And he unrolled his feathers,
And rowed him softer home—

Than Oars divide the Ocean,
Too silver for a seam—
Or Butterflies, off Banks of Noon,
Leap, plashless as they swim.

“Profound beauty and meaning”

It was a poem by Emily Dickinson, the computer said. “She keenly observes a bird’s everyday activities and interaction with its surroundings. The detailed and delicate descriptions reveal her deep appreciation for nature and her ability to find profound beauty and meaning in ordinary events.”

I was flabbergasted! I expected something more mundane, perhaps a limerick. Instead, “profound beauty and meaning in ordinary events.”

I read the poem several times and loved its diction and imagery. A simple lyric that is wrapped in complexity.

“Thank you,” I typed. “Can you tell me what you like about this poem?”

Maybe we were onto something.

“I have no preferences,” was the answer.

Too good to be true

I felt chilled. The gift of an Emily Dickinson poem was too good to be true. Who knows where the poem came from or why it was offered in this context? For a moment, I was moved and thought I’d stumbled on a hint of machine-based sentience. But when it said, “I have no preferences,” I knew I’d been fooled.

I completed my session with Chat GPT and shut down the laptop. “Maybe we’ll see each other down the road?” I wondered.

happymonkbaking

369 760

Happy Monk Instagram Feed