AI ushers in the Misinformation Age

3d rendering ai robot think or compute

It started as a simple personal damages claim. The plaintiff alleged he suffered injury while flying on the Colombian national airline, Avianca, when he was struck in the knee by a food and drinks service trolley mid-flight.

A statement of claim, asserting negligence by one of Avianca’s employees was filed by the Plaintiff’s attorney in the District Court of the Southern District of the State of New York in February 2022 and the case file number, 1:22-cv-01461 was attributed to it.

Lawyers for Avianca filed a motion to dismiss based on the company having filed for bankruptcy in 2021 at or around the time the alleged injury had occurred.

Lawyers for the plaintiff filed a motion with the court seeking to proceed despite the company’s bankruptcy arguing that Avianca should have continued to hold public liability insurance.

The court then requested that the attorney for the plaintiff file a motion providing case law examples.

After requesting an extension of time, the attorney for the Plaintiff did file an affidavit citing case law, eight exhibits with another example the attorney said existed but could not be located in case law.

And that is when the excreta hit the oscillator.

The plaintiff had used the AI chat mode, ChatGPT, to come up with the case law histories. The ever-accommodating ChatGPT did just that. But it made them all up. The cases themselves, the file numbers, even the jurisdictions where the cases had gone to trial were entirely fabricated.

Of the nine cases cited in the affidavit, eight did not exist and the ninth was deemed irrelevant.

When discovered, it left the defence attorney with more than three decades of experience as a trial lawyer, deeply embarrassed and a New York District Court Judge incandescent. The Judge issued a show cause notice why the defendant’s attorney should not face sanction which is a gentle way of saying disbarment.

In a verbal submission, the red-faced attorney told the judge that he had never used ChatGPT before and had assumed that everything the AI chatbot had told him was true. He told the judge that he sought confirmation as to the veracity of the material the bot had provided. In its HAL-like voice, the bot assured him it was.

The attorney now awaits the judge’s verdict on his professional career with a small army of his own lawyers assembled. His fate will be determined on, or after, June 8.

The dystopian future of the rise of self-aware machines scanning charred landscapes with what is left of humanity huddling in the debris out of sight of death rays remains the stuff of science fiction.

Colombian airline Avianca. Picture: AFP

In the early days of AI, the prevailing experience is that misinformation will flourish. Forget the death rays, AI promises to drown us in inaccuracy, “alternative” facts and bland deceit.

CNET, a technology media company that publishes various online magazines including one dealing with finance and an accompanying real estate magazine had published 78 AI-written articles under eponymic by-lines without disclosing it to readers. The jig was up when the ChatGPT ersatz reporter was asked to explain compound interest and got it horribly wrong.

The article implied a savings account initially containing $10,000 at a 3% interest rate, compounding annually, would accrue $10,300 in interest after a year. The earned interest would amount only to $300.

Bad robot. But it was worse behaviour from the publishers. As a matter of course, anything written by AI on any topic by a commercial publisher should acknowledge AI as the writer. CNET has since apologised to its readers and determined that it will inform readers where AI articles have been published in future.

We are not talking about the future of journalism, although there is that. The reach of AI generated copy means anyone who earns a living or undertakes educational pursuits which would involve hammering away on a keyboard can now be replaced by a blinking cursor that vomits up half-truths and outrageous lies.

In a conversation with a principal at a large private school in Melbourne, I asked how schools will deal with AI. There are obvious pitfalls, not least of all plagiarism. The principal had just returned from an AI in education conference in London. He rolled on enthusiastically about a new age of learning but ultimately, he conceded that AI was coming and there were little educators could do but accept it and learn to live with it.

While AI in its formative years might offer a convenient, fast alternative to human endeavour, its time and labour-saving features start to fall away when more time is required to fact check every assertion made.

It’s unsurprising that AI is only as good as its expanding database. ChatGPT was the first chatbot but there are others like Alphabet’s chatbot, Bard which appears more robust and reliable, but it still makes basic errors and passes them off as fundamental truths.

The threat is that the Information Age, initiated by the invention of the Gutenberg press in the 15th Century is coming to an end, to be replaced by an AI-driven Misinformation Age. The best way to counter AI’s excesses is to regulate the hell out of it but that’s a tall order given our legislators here and overseas are for the most part carting around laptops and smartphones they are barely capable of using.

On Wednesday morning, I ran the bogus case histories and the fake case numbers cited in the Defendant’s affidavit in the New York case through ChatGPT. It drew a blank. Apologies were extended but it could not find any “specific information” of the nine case histories ChatGPT had contrived little more than a month earlier.

It’s learning.