One of the ultimate aims of artificial intelligence is to create machines we can chat to.
A computer program that can be trusted with mundane tasks- booking our holiday,reminding us appointments -but also one that can discuss the weather and answer offbeat questions.
Alan Turing, one of the first computer scientists to think about artificial intelligence, devised a test to judge whether a machine was “thinking”.
He suggested that if, after a typewritten conversation, a human was fooled into believing they had talked to another person rather than a computer program, the AI would be judged to have passed.
These days we chat to machines on a regular basis via our smart devices.Whether it be Siri, Google Now or Cortana, most of us have a chatbot in our pockets.
Such conversations can be frustrating and are often little more than voice-activated web searches, leading people to ask questions they know the virtual assistant can’t answer.
Now, after several years of virtual bullying by humans, the machines are starting to fight back
Ask Siri “what is zero divided by zero?” and you get the following rather sassy response: “Imagine that you have zero cookies and you split them evenly among zero friends. How many cookies does each person get? See? It doesn’t make any sense. And cookie monster is sad that there are no cookies and you are sad that you have no friends”.
That is a pre-programmed joke, but increasingly the firms behind virtual assistants are trying to get away from scripted answers to ones that use more artificial intelligence.
Google is determined to offer a more human-like interface and has been feeding data into neural networks – artificial computer brains – to teach machines to learn for themselves how to have more natural conversations.Google is also determined to offer a more human-like interface and has been feeding data into neural networks – artificial computer brains – to teach machines to learn for themselves how to have more natural conversations.
This summer two Google engineers – Oriol Vinyal and Quock Le – released some of the chats humans had had with its neural network. Among other things, the network had learnt from thousands of old movie scripts.
Google’s chatbot wasn’t entered into this year’s Loebner Prize .The highest-scoring chatbot at the most recent contest, held last weekend, was Rose.
Although it was ranked the most human-like bot by three of four judges, it failed to fool any of them into thinking it was a real person.
The big technology firms don’t tend to take part in such competitions, in part because it remains incredibly hard to create convincing bots.
Read more at BBC