Skip to main content

DPD switches off chatbot: “People love to hear a bot swear!” says AI expert

Parcel service DPD has switched off its chatbot that is supposed to help customers when they face issues.  The French-owned company that operates the service that delivers many a pacel in Flanders decided to take this unusual step after the chatbot insulted the company and swore during a conversation with a customer.  VRT has asked Flemish AI expert Ann Nowé (Flemish Free University of Brussels – VUB) to explain how it got to this!

Colin Clapson
Published:

DPD is a familiar brand in Flanders and has processed many a parcel for a Flemish household, but it was British musician Ashley Beauchamp, who triggered the present controversy.  Ashley says he simply wanted to know where his parcel was.  The company’s chatbot didn’t seem much help but was able to compose a poem extolling some of the horrors experienced by the company’s customers.  DPD then decided to take its chatbot offline.

The VUB’s Ann Nowé explains that chatbots learn from human feedback.  “There’s nothing wrong with that” she adds, “as long as that happens in a controlled environment. “If a company trains a bot using feedback from customers, you should be able to trust it.  They will provide positive feedback when the bot gives a polite and correct answer.  Once the bot is properly trained, you can freeze the autodidactic mechanism and roll out the bot” explains the head of the VUB’s AI Lab.

Is it the customer's fault?

“Let the bot continue to learn from conversations with customers and you get a different story.  Many customers love to hear a chatbot swear and are keen to teach it such unsavoury practises.  I don’t know exactly what happened at DPD, but I can imagine something like this happened”.

“It’s not such a good idea to allow a bot to learn from conversations with customers when there are no checks.  It has to be monitored.  I don’t understand how a big company like DPD isn't more alert to possible developments like this”.

Ann Nowé suggests DPD should return to the drawing board and start from scratch: “It’s inefficient and pretty hard to get a bot to unlearn bad habits.  It’s just like a dog that has learned poor behaviour!”  

More about

Most read

  • 1
  • 2
  • 3
  • 4
  • 5