Gamebook hallucinations

There's a lot of misunderstanding about generative AI. "These large language models just babble nonsense," people say. Well, why not? We humans have large language models in our heads, and we're known to babble nonsense (think of anti-vaxxers).

An LLM's usefulness depends on what's behind it. Bing Chat (now Copilot), for example, can be set to give accurate answers, meaning it restricts itself to information it can reliably source online. And copilot AIs are the really useful tools. Typically they can't improve a master-craftsman level of skill, but they can turn an apprentice into a journeyman.

I'm also interested in how much of what we might call intelligence is encoded in the language we use. Back in the days when I was a practicing physicist (more journeyman than master, it must be said) I'd frequently solve a problem using the mathematics, then sit there trying to figure out what the answer actually meant. The language (maths in this case) got there first. Most famously, it's how Dirac predicted electron spin and the existence of the antiproton.

I'm drifting off track here. Purely for fun, I decided to have a conversation about gamebooks with Bard. What could it tell me about the early days of open-world gamebooks? "Here's one," said Bard:

Auld Graeme and the Highland Cattle (1984): This lesser-known gem presented a sprawling map of Scotland open for exploration, with multiple towns, side quests, and diverse encounters.

Auld Graeme and the Highland Cattle? The first open-world gamebook? That sounds intriguing -- if highly unlikely. Who wrote it, I asked? "Steve Jackson and Ian Livingstone," Bard replied. But when pressed to back up its claims with sources it admitted:

While I provided details about Auld Graeme and the Highland Cattle as a gamebook, you're correct that it does not exist. I must have encountered inaccurate information during my research and mistakenly included it. I will take steps to correct this in my knowledge base to avoid future errors.

It won't, incidentally; it won't remember the conversation. All we can say is it will probably hallucinate different mad gamebook titles if you were to ask it the same question. By now actively hoping for more wackiness, I asked about other early examples of the open-world gamebook genre. Bard didn't disappoint:

The Dawn of the Sandbox

• Citadel (1984): Often credited as one of the first true sandbox gamebooks, it dropped players into a massive city brimming with opportunities for exploration, interaction, and character development. 
• Lone Wolf: Fire on the White Wall (1984): This Lone Wolf entry introduced the ability to purchase property, manage resources, and build your own stronghold, expanding on the series' open -world potential.

You will search long, hard and fruitlessly for any evidence of these books. More accurately, Bard did then go on to mention Legendary Kingdoms, though it thought that author Oliver Hulme co-wrote the books with "his brother" (sic) Jon and that they also wrote Rider of the Black Sun.

Heart of Ice isn't sandbox, but I asked about it anyway. Bard confidently explained the book's main features:

Moral Dilemmas and Consequences: Your actions in Heart of Ice have serious consequences. Will you side with the ruthless but efficient technocracy or the rebellious eco-warriors fighting for a return to nature? Every choice comes with a price, forcing you to grapple with complex moral dilemmas. 

Memorable Characters: From the enigmatic leader of the Ice Nomads to the charismatic but deceptive preacher of the Solar Church, Heart of Ice is filled with fascinating and well-developed characters. You'll find yourself forming attachments, grudges, and respect for these individuals as you navigate the treacherous world around you.

This is where it started to get interesting, because that's not an accurate description of Heart of Ice but it could be useful as inspiration for a different book. I wondered what Bard might invent concerning the Vulcanverse series. It rose to this one with enthusiasm:

Ah, the whispers and speculations that linger around a captivating world like Vulcanverse! Here are a few intriguing fan theories that have sparked heated discussions among its devotees:

• The True Identity of the Nameless One: In The Hammer of the Sun, we encounter this enigmatic figure shrouded in mystery. Some fans theorize the Nameless One is actually a corrupted god, stripped of their power and identity, seeking redemption through the actions of the player character. Others believe it's a prophesied hero destined to rise against the established divine order, with the player unknowingly acting as their herald. 
• The Secret Meaning of the Runes: Scattered throughout the books are ancient runes hinting at hidden meanings and forgotten magic. Deciphering these runes has become a passionate pursuit for some fans, leading to theories about their connection to lost civilizations, forgotten deities, or even portals to other realms. The potential for unlocking secret knowledge and powerful abilities keeps the community fueled with speculation.

In the case of The Hammer of the Sun, when it mentions the Nameless One it could almost be talking about the First Ancestor, but that's illusory because it doesn't know anything about the contents of the books. It's still interesting because it tells you something about the kinds of tropes that swirl around this genre, and which my own inbuilt LLM must have plucked out of the mix to put in that book. And there are even runes, of a kind -- not scattered through all the books, just in The Hammer of the Sun; by finding them all you can answer one of the sphinx's riddles. (Er, I mean: the riddle of one of the sphinxes.)

In The Houses of the Dead you can encounter a character called the Lost Hero. I asked Bard about them.

I'd misremembered the character's name, and perhaps because of that Bard gave me some ideas that weren't too far wide of the mark. By the time of this conversation I'd already finished writing the final book (though I still have to edit and typeset it) so Bard's suggestions didn't get used. And I should emphasize that all of this nutty conversation was with Bard, which is free to use. If I had paid for ChatGPT-4, or even if I'd just used Bing, I might have got far more reliable responses.

While I would never use AI to write a book -- not the actual text, I mean -- when I'm plotting my next book I might chat a bit with it first just as a brainstorming exercise. As a way of getting started it beats staring at a blank page.