When Artificial Intelligence Met Little Red Riding Hood
Little Red Riding Hood by Jacob and Wilhelm Grimm is a well known morality tale. It is the story of a young girl who is waylaid by a wicked wolf who tricks her and her grandmother and eats both of them up. It is a deceptively complex story that begins in blissful innocence then descends rapidly into deception and a double homicide before ending in a heroic finale.
In the two centuries since the Brothers Grimm wrote their version of Little Red Riding Hood, the number of documents churned out by printing presses and electronic publishers has grown prodigiously. The typical Fortune 1000 company is said to maintain 20,000 to 40,000 active contracts at any given time. Until now, analyses on this mountain of documents have been an entirely manual task with associates, managers and executives poring over contracts, agreements and reports for hours at a time. But help is at hand. Machine learning and artificial intelligence are uniquely placed to transform such tasks by helping us derive meaning and insight from text in a highly efficient manner.
But how well do such techniques do? Basic sentiment analysis which assigns a sentiment, say positive or negative, to a given word, works up to a point. When applied to Little Red Riding Hood, such simple analysis fails to capture the grisly nature of the second half of the story and suggests it to be as uplifting and idyllic as the first. (See illustration). Clearly we need a more robust approach that understands concepts as opposed to merely chasing down key words. This is important if machine driven text analyses are to have value in fields like business and law where the cost of misinterpreting a document or a contract can be prohibitively high.
In a watershed study, 20 top US lawyers went toe to white-shoed toe with the LawGeex Artificial Intelligence algorithm. The challenge was to see whether the lawyers or the algorithm would perform better in spotting risks and inconsistencies in an everyday set of legal contracts. The AI was trained on tens of thousands of contracts using a combination of deep learning technology and human experts. This allowed it to “understand” legal concepts while wading nimbly through legalese - “technical legal language that is often complex and counterintuitive.”
The results were sobering. The AI achieved an accuracy of 94% as compared to 85% for the human lawyers. But that’s not all. While the lawyers took an average of one hour and a half to review the set of five legal agreements, the AI was done in only 26 seconds.
This is a powerful illustration of the promise of artificial intelligence. When applied thoughtfully, AI frees up humans from mundane activities and allows them to devote more time and energy to higher-level tasks that our creativity and cognitive abilities are much better suited to address. In doing so, AI can bring costs down dramatically while giving rise to impressive productivity gains. And that’s a story worth telling.