Air Canada Must Pay Passenger Damages over Chatbot's Bereavement Discount Lies
gizmodo.com
gizmodo.com
This seems entirely reasonable. But I have to wonder the limits. Clearly the customer in this case wasn't trying to scam Air Canada, but on the other end of the spectrum there's a case where a chatbot can be tricked to sell a car for a dollar[1]. What happens in the middle ground (perhaps where a user prods the bot for info about return policies and gets responses the user wants to hear)?
[1] https://twitter.com/ChrisJBakke/status/1736533308849443121
Easy example first: if you're trying to fast-talk someone into agreeing to sell an expensive car for $1, you're trying to scam them. No judge (I hope) is going to uphold that even if the words "Yeah, sure" came out of the salesperson's mouth.
It gets more complicated when it's a contract people might actually sign. If you call up your ISP and the low-level sales representative agrees to give you 50% off for the next six months as credit for a couple days of service outage -- that might actually be a binding contract, even if it's well outside what the company authorized their agent to do. Same with agreeing to take a return, etc.
However I am unsure whether thinking of chatbots as legal agents of their owner/operator is correct (either as a question of law, or just being useful as a metaphor).
Just like you shouldn't be employing sales people that sell items with negative margin.
This is actually a nice verdict, it effectively says 'if you ask a computer to represent you then what it says as your representative is binding on you'.
More discussion yesterday: https://news.ycombinator.com/item?id=39378235