Canadian MP Reads a Chatbot's Prompt Aloud During a Speech in Parliament

Progressive Conservative MP Bill Oliver, a member of the legislative assembly of the Canadian province of New Brunswick, gave a speech last month on the risks of creating human rights agencies. He rightly noted that citizens often pin hopes on such bodies that exceed their actual powers.

The trouble is that Oliver then went on to read aloud a service aside plainly addressed not to parliament at all, but to himself: “Here's a more natural, flowing version of that passage that reads like a piece of legislative oratory rather than a set of short talking points.” Anyone who has ever used a language model will recognize the telltale turn of phrase: this is how a chatbot proposes an alternative style in response to a request. The MP, by all appearances, did not realize in real time exactly what he was saying.

From Unnoticed Slip to National News

At the moment it was delivered, the odd sentence went by almost unnoticed. This week, however, the video began spreading across Reddit and Threads, and the story was then picked up by major Canadian media — right up to the public broadcaster CBC and the newspaper The Toronto Star. The latter saw in the episode a symptom of “a growing rift in society: between elites who happily hand their duties over to the Borg, and the masses who are outraged by it.”

Oliver is, of course, far from the first politician to publicly read out a text prepared by someone else, and certainly not the first to enlist a chatbot's help in drafting a speech. The novelty lies elsewhere: to utter a blatantly machine-generated insert out loud, in such a public setting and, seemingly, without noticing the catch — this, as Ars Technica quips, is a fresh milestone in the annals of awkward exposures of AI use.

To Conceal or to Confess

The episode fits into an already noticeable trend. Lawyers, writers, journalists, and academics have repeatedly landed in embarrassing situations when the use of language models came to light — most often because of obviously invented “facts” and hallucinations in texts passed off as written by a human.

It is hardly surprising that many prefer to keep quiet: according to a study last year by Duke University, employees tend to conceal their use of AI, because colleagues perceive such work as a sign of laziness and see the author as easily replaceable. The story of the Canadian MP shows what the opposite approach leads to, when a chatbot's prompt is read out straight from the transcript.

Source: Ars Technica