On pseudo-facts – Regina Rini in TLS:

‘Chatbots frequently render pseudo-facts – “hallucinations” in industry lingo – with the same apparent confidence as genuine information. In late February a Vancouver lawyer was scolded by a judge for filing citations to imaginary cases, allegedly the legal stylings of ChatGPT. Remarkably, stories of professionals unwisely outsourcing their expertise to robots have become common. The lesson should be obvious: don’t trust bots with anything important.’

(…)

‘Do you have any protection against a hallucinating service bot? Thankfully, the emerging answer is yes. At least in a small claims tribunal decision in British Columbia last month. The case concerned a man named Jake Moffatt, who in November 2022 booked a last-minute flight to Toronto for his grandmother’s funeral. An official Air Canada chatbot said he could book the flight at full price, then later file for a bereavement discount. When he tried to make good on that promise a human employee overruled the chatbot; prior approval was required for bereavement rates.’ (…)

‘Modern AI is powerful because it extracts patterns from enormous data sets, allowing it to adapt to new circumstances rather than stay within strictly defined algorithmic rules. But this suppleness means that it is hard to know who to blame when things go wrong. No one programmed the Air Canada chatbot to lie about corporate policies; presumably it tricked itself after imbibing text from unrelated documents. So how could its designers, or the company, be at fault?’ (…)

‘We are probably only a few years from widespread adoption of autonomous digital personal assistants. In 2018 Google demonstrated a smartphone voice assistant capable of calling the hairdresser to book an appointment for its owner. What happens when you tell your AI-PA to book an affordable holiday and it instead signs you up for a shady timeshare in Ibiza? This is a looming problem, not only for courts, but also for ordinary people. We want it to be easy to make large corporations accountable for their AI agents, but no individual wants to be accountable for using the same technology on our own behalf.’

Read the column here.

Alas, not only chatbots make up pseudo-facts.

And fairly often human employees promise things that never happen.

Who can we trust with important things? The expert fell out of grace many years ago, but when it comes to your teeth, to give just one example, a reliable dentist appears to be the best choice.