The Grand Illusion
The other morning in the course of my daily search, I came across this as part of an “AI Overview” on a Google search:
• The Saga of Recluce: A popular fantasy series that follows Rahl, a young apprentice who becomes a powerful mage
As those who’ve read the Recluce books know, Rahl is the protagonist of Natural Ordermage, and the Recluce Saga isn’t about just one protagonist. In fact, Rahl is the ninth protagonist (in publication order) of the saga.
The summary also states that I attended Williams College, which is slightly misleading because I graduated as well as attended.
Since that first occurrence, that same “AI Overview,” or one similar, has reoccurred on several occasions.
Obviously, such errors irritate me, but, more than that, they disturb me because an artificial intelligence (supposedly) is providing incorrect information at the same time that Google is touting its AI capabilities.
I didn’t even ask for a summary in my search. It was provided unasked for. So…not only is the information thrust upon me, but it’s wrong, and likely provided incorrectly to other searchers as well.
I’m also fairly certain that other erroneous information is being supplied by other AI overviews on differing subjects, simply because these AI overviews are based on internet-posted information, much of which isn’t fact checked in any way, but such “overviews” lend a credence to dubious or erroneous “facts.”
I wonder if DeepSeek would do any better.
Luckily, we’re in the Misinformation Age, where accuracy no longer matters, so results like this are perfect responses! No one needs to know anything, they can just believe whatever they want, and it’s so good to see Big Tech doing their part by crafting AIs that can’t get anything right, but just maybe can confirm existing biases!
(Removing sarcasm hat now…)
The large language models erroneously referred to as AI just look for the combination of words most likely to occur on the totality of all digitised writing in English. They don’t do nuance or complexity. So, if Rahl is the most referenced (or the most recent of equal references) that’s what the LLM supplies. In particular, if an incorrect statement is more frequently stated than a correct version, that’s what the LLM supplies- and in so doing probably increases the frequency count of the incorrect version, thus firming it up in the’minds’ of itself and all the other LLMs. They are machines for propagating urban myths and conspiracy theories
a PS example: the IBM AI for putting up update messages on coverage of the Wimbledon tennis tournament: nice constrained remit, should be a doddle. BUT tennis players have two important numbers associated with them – their ranking in the world and their seeding for the tournament: often the same, but if (eg) the #2 drops out everyone else moves up one on the seeds. The IBM AI rarely associated the right number with each word: it doesn’t know what the words mean, it just looks at how often the number and the word are together. The frequency count thing also means that if a player was #1 ranked for a year, but last week dropped to #5, there are a lot more references to #1 in its ‘mind’ …
When I ask Deepseek to “summarize the saga of recluse” I get as a final summation paragraph:
Overall Summary:
The Saga of Recluce is a richly detailed fantasy series that delves into the philosophical and ethical implications of power, magic, and societal structure. Through its diverse cast of characters and non-linear narrative, it explores how individuals and societies grapple with the forces of Order and Chaos, often revealing that neither is inherently good or evil—it is their use and balance that matter. The series is known for its deep world-building, complex characters, and thought-provoking themes.
That’s a damned sight better than what Google comes up with.
Just for fun, asked ChatGPT to write the first chapter of a new Recluce novel (with some additional parameters):
From The Balance of Iron and Order by Master Smith Lurgan of Fairhaven, 1193 AF
“A smith does not shape iron through strength alone, nor does a mage shape Order through mere will. Both require precision, patience, and an understanding of balance. To strike too hard is to shatter. To strike too soft is to weaken. In this, as in all things, Order endures only where discipline prevails.”
Chapter One: The Blacksmith’s Path
Garric brought the hammer down hard, the impact ringing through the smithy. He didn’t much care for finesse—so long as the iron held its shape and did what it was meant to do, that was enough. His father, of course, disagreed.
“You’re favoring your right again,” Veylor said from the doorway, arms crossed over his chest. The smith’s apron he wore was old, stained with years of work. “You think the metal won’t notice?”
Garric grunted and flipped the horseshoe with the tongs, setting it back against the anvil. “Seems fine to me.”
His father only shook his head. “It’s not about what seems fine. It’s about what is fine.”
Garric sighed, adjusting his grip before bringing the hammer down again. He figured he had maybe a glass before midday, and he was already thinking about the cool green berry juice waiting at home. Nothing better after a morning in the forge.
The sound of hooves outside broke through the steady rhythm of hammering. Garric turned slightly, catching sight of a rider slowing to a stop in front of the smithy. The man wore the black-and-white of Recluce, his cloak marked with the sigil of Fairhaven.
Veylor straightened, wiping his hands on his apron as the rider dismounted. “What brings a Council man out this far?”
The rider’s gaze settled on Garric. “Garric, son of Veylor?”
Garric set the hammer aside, feeling a pit settle in his stomach. He had a bad feeling about this.
“He is,” Veylor said.
The rider nodded once. “You are summoned to the Council.”
Garric exhaled. There went his afternoon.
And, most likely, his future.