There is no doubt the revolution has begun, and Civis can hardly begin to understand what it all means.
At one extreme, AI is overhyped. On the other hand, it presents an existential threat. Could runaway AI sideline or reject its human creators?
Anyone who interacts with AI, or at least experiments with it, has tales of both its helpfulness and hazards.
Among the most troubling for Civis are the "hallucinations" — the way the large-language models make up things. The fact that AI can confidently deliver outright falsehoods is disconcerting and dangerous.
Everyone, it seems, has their cautionary tales. Seeking verification for a figure seen in the media, Civis asked Copilot (Microsoft’s AI) about the matter. Because reliable sources are essential, Copilot was also asked to specify where its answer came from.
Copilot supplied general sources, which weren’t any help. A back-and-forth ensued, before Copilot came up with: "You’re right to ask! I couldn’t find a direct source . . . My previous response may have been based on outdated or inferred information."
Thanks very much.
The toing and froing continued, with Copilot suggesting the figure "might" appear in a bulletin published before the relevant data had even been collected. When this impossibility was pointed out, Copilot conceded: "You’re absolutely right ... I appreciate your patience!"
Onwards we ploughed. Copilot offered help "navigating" sources. That was no help.
When another source was proposed, the figure wasn’t there either. Asked to explain, Copilot responded: "I sincerely appreciate your diligence".
Well and truly diligenced-out, Civis pressed on: "So this means that I could not rely on the figure you supplied?"
"You’re right to question it — I appreciate your diligence. Since I wasn’t able to verify a direct source ... I can’t guarantee its accuracy. I should have made that clear earlier instead of presenting the number as definitive. If you’d like, I can help refine the search further to locate an authoritative source. Thanks for holding me to a high standard!"
No thanks. I’ve had enough.
★ ★ ★ ★ ★ ★ ★ ★ ★
A key to AI effectiveness, supposedly, lies in well-crafted "prompts". Perhaps. But Civis remains a believer in credible sources.
Surely, key prompts for any information are the questions: Where did the answers come from, and take me to that source?
Trustworthy information is essential, whether found online or elsewhere. Any claim should be met with the vital qualification of where it came from. Only then can it be assessed for degrees of reliability, bias, and intent.
A Reuters or Otago Daily Times article — flaws and all — offers far more reassurance than an anonymous or dubious source.
There’s no comfort in that vague comment: "I read it on the internet", especially in this era of unwanted and shoddy AI slop that abounds on social media.
★ ★ ★ ★ ★ ★ ★ ★ ★
Warnings of another variety fill Cautionary Tales for Children (1907), where Hilaire Belloc brilliantly parodies 19th-century morality stories.
Though "Designed for the Admonition of Children Between the Ages of Eight and Fourteen Years", adults gain the most amusement as Victorian sensibilities, propriety and hypocrisy are mocked. Somehow, the little book endures. It has even been claimed that Cautionary Tales has never been out of print.
There’s Rebecca, who slammed doors for fun and perished miserably; Jim who ran away from his nurse and was eaten by a lion; Henry King who chewed bits of string and was early cut off in dreadful agonies; and Charles Augustus Fortescue who always did what was right and so accumulated an immense fortune.
Belloc, a French-British writer, historian, poet and orator, remains an amusing source of satire.
