Computers certainly aren't dumb. However, if you’ve ever talked to a chatbot or a virtual assistant, you'd be forgiven for occasionally thinking so. For all their brainpower, machines are still surprisingly clunky and awkward in the art of conversation, especially when it comes to answering questions
To overcome this weakness, computer engineers at the University of Maryland have been feeding machine learning algorithms questions designed to challenge them, hoping they will become better trained at communicating with humans using language.
As reported in the journal Transactions of the Association for Computational Linguistics, the research has generated a collection of over 1,200 questions that completely bewilder even the best computer answering systems today, despite being comparatively easy for people to answer.
Here are a select few examples of the trivia questions:
- “Name this European nation which was divided into Eastern and Western regions after World War II.”
- "Name this river that defines the border between Texas and four Mexican states."
- "Identify this metal that is used for decorative coatings and many musical instruments."
- "Name this type of symbiotic relationship exemplified by a flea which sucks blood from its host."
- "Identify these pieces of protective gear which were, contrary to popular belief, more commonly worn by samurai than by Vikings."
- “Name this scene painted by Leonardo Da Vinci in which Christ indicates that Judas will betray him.”
- "Name this South African leader who became president in 1994 after spending 27 years in prison."
- "Name this current Israeli prime minister who oversaw Operation Protective Edge in early 2014."
- "Name this photographer who took many black-and-white landscapes set in the American Southwest."
(Bear in mind, these are some of the easiest ones and are still relatively tough.)
But how can a computer, with all its memory and processing power, be foiled by such simple questions?
The reason is much more to do with language than knowledge. Computers break down and answer questions using a very different method to humans. It’s notable that the questions are phrased in a bit of an odd way. That’s because they’re laced with six different language phenomena that consistently stump computers, but don’t tend to phase humans.
These tactics include unexpected contexts, such as a reference to a political figure appearing in a clue about something unrelated to politics. While a computer might be misled or "distracted" by this additional context, it might actually spark a valuable thought in a human brain. Alternatively, the question might require some form of reasoning skill, such as clues in the question that require the mental triangulation of elements in a question or putting together multiple steps to form a conclusion.
“Humans are able to generalize more and to see deeper connections. They don’t have the limitless memory of computers, but they still have an advantage in being able to see the forest for the trees,” Jordan Boyd-Graber, associate professor of computer science at UMD and senior author of the paper, explained in a statement.
“Cataloguing the problems computers have helps us understand the issues we need to address, so that we can actually get computers to begin to see the forest through the trees and answer questions in the way humans do.”