WHY AI CAN’T WRITE THIS ARTICLE (YET)
Roger Schank, a researcher and former professor, once proposed a novel goal for artificial intelligence: A computer should be able to watch West Side Story and recognize the plot of Romeo and Juliet. Schank and his students believed that stories are central to intelligence, reasoning, and meaning. By Schank’s measure, today’s AI isn’t intelligent at all.
The feature article on AI that HBR.org published earlier this week is, ironically, a good example of the kind of work that computers can’t yet do. It was written by two experts who drew on decades of experience to formulate a thesis, assemble evidence, and construct a narrative. And three editors helped to shape the nearly 5,000 words that made it into the final piece.
The fact that software can’t yet write an article like that isn’t a knock on AI, or evidence that it won’t be transformative. But that fact offers a window into how, exactly, machine learning technologies work, what they are and aren’t good at right now, and how they’ll develop as writing tools — or even writers — in the future.
Today’s AI works by formulating tasks as prediction problems and then using statistical techniques and lots of data to make predictions. One simple example of a text-based prediction problem is auto-complete. When I type “How’d” into a text message, my phone uses data and statistical modeling to predict what’s coming next. It offers “it,” “you,” or “the.” “It” is what I had in mind, and once I select that, my phone moves on to predicting the next word. This time it’s so confident that I’m going to select “go” (which is right) that it doesn’t even offer other options but instead moves on to the next word, suggesting “go with” or “go today.” In machine learning, prediction problems like this are called supervised learning. Given a data set containing the right answer — in this case lots of completed text messages — an algorithm learns to recognize patterns, such as that “go” often follows “How’d it.” (Another kind of machine learning, unsupervised learning, works differently, but supervised learning has driven most of the recent progress in the field.)\
hbr.org
|
