Our ability to answer questions currently separates us from machines in the world of artificial intelligence.
But researchers are working on algorithms to give computers this skill too.
Now, scientists at Facebook have come up with questions that they say artificially intelligent machines must answer, if they can ever match human brainpower – but some of them are a bit tricky, meaning some people may fail them too
The team from the social media firm’s AI lab in New York have come up with a list of 20 questions to separate humans from robots, because they say ‘many existing learning systems can currently not solve them.’
They test different types of reasoning and the ability to process language.
DO YOU PASS AS A HUMAN? TRY THESE QUESTIONS CREATED BY FACEBOOK TO TEST YOUR REASONING AND LANGUAGE SKILLS
1) The first question of the 20 listed in the research paper was the simplest for a computer to answer. It is:
John is in the playground
Bob is in the office
Where is John?
2) A harder example for a computer to answer is a question that requires it to differentiate three separate arguments, such as:
Mary gave the cake to Fred.
Fred gave the cake to Bill.
Jeff was given the milk by Fred.
Who gave the cake to Fred?
Who did Fred give the cake to?
What did Jeff receive?
Who gave the milk?
3) Another task tested a machine’s ability to perform simple counting operations by asking about the number of objects. For example:
Daniel picked up the football.
Daniel dropped the football.
Daniel got the milk.
Daniel took the apple.
How many objects is Daniel holding?
4) Basic deduction skills were tested by questions such as:
Sheep are afraid of wolves.
Cats are afraid of dogs.
Mice are afraid of cats.
Gertrude is a sheep.
What is Gertrude afraid of?
5) And reasoning about size was tested by this question:
The football fits in the suitcase.
The suitcase fits in the cupboard.
The box of chocolates is smaller than the football.
Will the box of chocolates sit in the suitcase?
Some of the questions require machines to recall facts, while others need reasoning, or to count the number of objects, expressed in a written passage.
The researchers say that humans should be able to score full marks.
While you may have got one answer wrong, none of the seven computer learning programmes that Facebook tested got all 20 questions right.
The experiment was designed to test whether a machine can answer questions ‘via chaining facts, simple induction, deduction and many more,’ the study published on arXiv says.
Scientists are keen to create this ability so that humans can talk fluently with machines, which could lead to robots looking after the elderly and more intelligent websites, for example.
The paper says: ‘One long-term goal of machine learning research is to produce methods that are applicable to reasoning and natural language, in particular building an intelligent dialogue agent.’
The experts aim to ‘classify these tasks into skill sets, so that researchers can identify (and then rectify) the failings of their systems.’
The results of the test show that humans are still ahead of the game when it comes to interpreting the world’s subtleties and language.
The questions are written in very clear and simple terms so there is still a way to go before computers catch us up.
