Conversation with an AI

Published 11 Mar 2021

By Aileen Wang



As AIs become increasingly powerful, they become more and more able to imitate, and generate human speech. Last time, I went on a long ramble on the fictional hypotheses about the societal, moral and philosophical questions that might eventually arise around AI. This time, I thought I might put the fictional to the test – by personally engaging with an AI myself.

You may have heard of GPT-3, short for Generative Pre-trained Transformer 3. It’s, in the words of Wikipedia, an ‘autoregressive language model that uses deep learning to produce human-like text’. In short, it’s an AI language generator that speaks in much the same way a human would. If you’d like to see some of its, frankly astounding, capabilities, I highly recommend this video here.

However, unfortunately for a money-short, time-poor university student like myself, the product is commercial, and although OpenAI (the developer of the AI) releases an API to access the model, there’s a long waitlist. Fortunately, GPT-2, the younger sibling and previous model of GPT-3, is available for limited access through this site , creation of Canadian engineer Adam King. Through it, you can prompt GPT-2, after which it will autocomplete whatever phrase you input.

I have to admit, I arrived an incredible skeptic. However, this exercise has proved both one in hilarity as well as surprising competency on the part of the AI. Perhaps the word ‘surprising’ is ill-used here, but I can think of nothing else to describe my feelings when a coherent, genuinely interesting snippet of text appeared before my eyes, while knowing that the hand behind was not flesh, but machine.

The first prompt

Prompt: In hopes of a brighter day, with a rosier dawn

Answer:

In hopes of a brighter day, with a rosier dawn ahead of her, the Princess slowly opened her eyes and sat up to get her bearings.

She was in the forest, and that meant one thing - running.

So her feet began to move to that end, and she noticed with some alarm that she was stark naked.

Good Goddess!

Where was her cloak?

It was night, and she would have to wait until morning to go in search of the bridle and bridle - stable, or find some safe - if a bit hidey - hole - to change clothes.

Tiptoeing through the forest in the dark,

And here it broke off, having clearly hit a character limit. In the interests of public decency, I’ll finish it off:

Tiptoeing through the forest in the dark, she stumbled upon a soft pile. It was her cloak, dusty and muddy, but still wearable. Grateful for anything in the cold, she shucked it on and continued her way towards the dark, wooden heart of the forest.

And so we leave our Princess to an unknown, hopefully fortunate, fate. Who knows what she’ll find, if it’s a Prince Charming sleeping in a glass coffin, or a beast that needs slaying, or a dragon that needs taming? Maybe she’ll set up a hut and become a witch.

The second prompt

Prompt: Rain fell like sleet on the glass

Answer:

Rain fell like sleet on the glass of the huge old mansion, but Rolf didn’t mind.

The shadows under the trees were long and foreboding and he kept his eyes peeled for any sign of wildlife.

The few bushes and plants were wreathed in mist and the crunch of the trees and the rustle of dry leaves seemed closer somehow.

Rolf found himself hoping he’d find someone to run with him, some reason to get moving.

He was startled out of his thoughts by a knock at the door, startling Rolf out of his reverie.

“Nah, not much of one.”

A sort of Gothic horror setting itself up, if not for that weirdly casual, oddly Australian bit of dialogue at the end. Is the skeleton of a kangaroo with the desiccated skull of a man in its short paws going to appear on the other side of the door? Or perhaps Rolf is already insane, and in his insanity his mind has somehow drifted all the way down the Pacific.

The third prompt

Prompt: He poured the tea onto his own head

Answer:

He poured the tea onto his own head, dousing himself.

“You asked for a story.

So here you go.”

The figure laughed, and it was a dark, throaty laugh.

“There once was a man, the man from your stories, the one that stared off into space.

This man was a very sad man, sitting on his little balcony day in and day out, watching his flowers die in the street.

He sat there and did nothing but watch the flowers, and never spoke a single word to anyone, not even his wife.

She was always off with some new man from the city, and this

Again, an unfinished snippet. This time, it seems like a weird cross between a fae exchange, a fireside tale, and some poorly-aged Aesop about the dangers of cuckolding. In the interests of not leaving this poor man hanging, again, I’ll finish it off.

She was always off with some new man from the city, and this grieved him as much as it relieved him. He was too cold to love her, and to selfish to fully accept that she too, did not love him.

Now, I was going to finish it here, but I figured it’s not exactly the happiest note to leave things on. So, like a gambler hoping for a lucky strike, here we go one last time.

The fourth prompt

Prompt: The roses in her arms were almost as radiant as her smile.

Answer:

The roses in her arms were almost as radiant as her smile.

Sunshine, rainbows and kittens, oh my! Kody Clemens is the fourth of seven siblings, but he stands out in more ways than one!

By day he’s the meanest linebacker on the good ol ‘Trojans team.

By night, he’s the hero in the eyes of every little girl who’s ever been loved by a hero.

Kody is just trying to stay out of trouble, make good grades, and hold onto the only friend he’s ever had.

That is, until he gets the chance to fulfil the fantasy of every little girl who’s ever been loved by a

Now this is interesting! From the looks of it, if the AI had been allowed to go on for longer, we might have been caught in a rut. The rut of ‘every little girl who’s ever been loved by a hero’, which is either a lot or very few, depending on the definition of hero. Finally, we see a hint of the mechanical creaking on the other side – although still coherent, repetition and rapid changes in context and tone are starting to make themselves known as potential problems.

Well, it’s not quite the joyful, uplifting end I was hoping to get, but a laugh is better than nothing, right? Go in peace, safe in the knowledge that Kody Clemens, fourth of seven siblings, is looking out for you!


Tags: Data Science Machine Learning, Deep Learning and Neural Networks