A debate has recently been sparked in Belgium over the use of artificial intelligence, as Petra de Suter, rector of Ghent University, used artificial intelligence to write her commencement speech for the academic year last September, but her text contains errors and two fabricated quotes – this is what is known as a "hallucination", Belgian public broadcaster RTBF reported.
According to Petra de Suter, Albert Einstein supposedly said, "Dogma is the enemy of progress." Except the scientist never actually said that line. Therefore, it's a hallucination, false information that artificial intelligence presents as true.
Imagine that you are using ChatGPT to write a speech. You ask it to include a quote from Albert Einstein on the topic of progress. Two scenarios are possible: it either gives you a real quote, or it produces a plausible quote, i.e. something Einstein didn't say but could have said.
Ig Bersini, professor and director of the artificial intelligence laboratory at a Belgian university, explained the phenomenon.
"The AI combined Einstein's quotes to create a new one that makes sense. Statistically, Einstein could have said it. ChatGPT combines words that were often associated together. But, of course, this quote is fake because Einstein never said it," Bersini said.
When we say that a quote makes sense, we mean that it is a statistical construct. It puts together different words that are part of a person's quote, but the order of those words does not exist. Einstein never said, "Dogma is the enemy of progress."
In the case of quotations, this is a fatal mistake, according to Igo Bersini.
"That's obviously something that absolutely has to be checked. And if you're the rector of a university, you certainly have to adhere to a certain code of ethics regarding knowledge, and when it comes to quotes, you absolutely have to check their authenticity," because a few clicks on a search engine are enough to prove an error.
Artificial intelligence giants are trying to reduce these hallucinations.
In some areas they are practically non-existent.
"When there are fewer words, like in programming and mathematics. The vocabulary is more limited for the exact sciences, so artificial intelligence works very well. It's more complicated when you use it for novels, essays, philosophical texts or speeches. Words are more diverse and can have multiple meanings," Bersini said, adding that extra caution is needed in these areas.
Petra de Suter's unfortunate experience serves as a reminder.
Artificial intelligence can deceive us. It also makes mistakes, having absorbed all the texts that are given to it. This is also inherent in human nature. Except that humans understand the concept of truth and lies, unlike artificial intelligence. It does not check its sources.
So it's up to you to carefully process the information you collect and to exercise your critical thinking at all times, Bersini added.
Bonus video:
