LOS ANGELES (KNX) — While our machines have become more interactive throughout the years, the idea of a machine or a piece of technology being ethical or moral sounds out of this world.
Just ask Siri a moral or ethical question and more often than not, she’ll respond with “Interesting question” or “I don’t understand.”
Then there’s Delphi, an artificial intelligence system that tests ethics and morality in machines. Yejin Choi, the system’s creator, told KNX In Depth that while AI plays an important role in our lives, it doesn’t have the ethics and morals like humans do.
“AI’s already inside a lot of what we use today, like search engines, social media, content recommendations, and even home devices – all have AI inside,” Choi said. "Currently, these systems are not particularly taught with the social norms, ethical norms that we abide by when we interact with humans and as AI becomes a more integral part of our lives, we thought it’s important to research.”
So does Delphi come up with answers?
“It’s basically learning through examples, lots of examples,” Choi explained. “It’s taught with 1.7 million examples of what people might judge.”
When it comes to how the system reacts to people’s questions, Choi said the questions people ask have been surprising.
“We were very surprised by the volume and the creativity of the questions that people entered into our system,” she explained to KNX In Depth. “In our previous research on racism, sexism, or toxicity of newer language models or common senses models, we always had this web demo available but usually people don’t really play with it whereas in this particular case, we have received more than three million user queries and they’re very good and very adversarial and really pushes the limit of what AI can and can not do.”
Choi added that Delphi doesn’t really understand anything. “Currently, the demo will hash a little bit if it has low confidence in answering. Often times, if you give convoluted sentences, especially if you present a weird situation where there are two things at odds at the same time, along which is more important, it might not answer.”
Listen to Choi’s full interview with KNX In Depth up above.