Teaching Robots to Read Harper Lee

Blog Stories

It started with Siri - Apple’s built-in Intelligent Assistant - who responds to the words that you say rather than the requests that you type. It can access any information on your device, and of course where that information isn’t enough, can refer you to more relevant resources on-line.

Siri has been known to respond with humour to unusual questions, or to reference pop-culture: “Who let the dogs out? Siri: “Who? Who? Who? Who? Who?.

It can even be updated with more relevant references, such as the upcoming premiere of the next Game of Thrones series. However these are pre-programmed responses.

Earlier this year Microsoft went a step further and released an Artificially Intelligent ‘chatbot’, but with disastrous effects. 

"Microsoft shuts down AI chatbot turned into a pro-Hitler racist troll in just 24 hours" - The Independent

Perhaps we should thank Microsoft for that and for pointing out that "we need to be more deliberate with how we interact with such kinds of AI technology, since these programs will only magnify the ideas and information we feed them.” 

One possible solution is story-telling.

Recent research by US social scientists found that those who read novels by the likes of Toni Morrison and Harper Lee do better at 'theory of mind tests' - "inferring others feelings". At the same time a team from The Georgia Institute of Technology are pioneering a system called Quixote, by which an artificial intelligence learns “value alignment” by reading stories from different cultures.

Stories teach us values. Take To Kill A Mockingbird for example. We don’t just learn that racism is unjust and systematically wrong but that it’s complicated, that it involves real lives, suffering, and courage. Stories inform and engage us by enabling us to experience a reality other than our own. They can evoke emotion in us, empathy for individuals, compassion or disgust. But they allow us to get there by ourselves.

What if we could teach robots in the same way?