MIT professor and ThinkPad superfan: “Storytelling” machines key to unlocking artificial intelligence
When Prof. Patrick Winston earned his computer science Ph.D. from Massachusetts Institute of Technology in 1970, his thesis adviser, famed artificial intelligence pioneer Marvin Minsky, believed that computers would someday think like we think and even surpass us. Minsky was particularly impressed when Winston showed how a computer, like a human, could learn to recognize arches and other structures, learning something definite from each example.
Nearly 50 years later, Winston, himself now a popular professor and renowned artificial intelligence expert at MIT, believes the “fairy dust” that separates humans from other animals is the ability to understand and tell stories.
To give machines that magic, Winston has spent the last decade leading MIT’s Genesis Group, which aims to develop computational models that mimic human capacity for building and decoding narratives, whether they be Shakespear’s plays or the way in which signals move through circuits.
“If we’re going to understand human intelligence we have to understand the human capacity to tell stories, to understand stories and, especially, to make up new stories by combining bits and pieces of previous stories,” Winston said.
Computer programs with storytelling capabilities would be better at communicating with humans, and that could help people trust such systems, Winston believes. Machines could explain in human-like language, for example, the way in which they reach conclusions, allowing people to verify results. With the capability of understanding abstract concepts like revenge or nationalism, they could help decision makers understand how policy choices or news stories might be perceived around the world based on cultural differences.
“Such programs could model the interplay of varied factors in fields as diverse as education, economics, politics, health care, law, urban planning, law enforcement, defense, and business,” Winston said.
Trust, however, is a key ingredient.
“Some of today’s AI systems are powerful, often powerful at a human level or beyond, but as many have pointed out, they are dangerous in proportion to their power because they cannot explain how they reach conclusions,” Winston observed. “Often they make mistakes obvious to us humans.”
As it turns out, giving such skills to a machine is no easy feat. How can a machine be taught to infer and reach conclusions not explicitly stated in a text, for example? “Humans are not logical machines. Whenever there are elements in a story that can’t be explained with logical certainty, we typically find plausible explanations because we are explanation seekers,” Winston said.
In Winston’s Genesis Group, which is part of MIT’s Computer Science and Artificial Intelligence Laboratory, Winston and squadrons of his students have painstakingly built technology that can analyze roughly 100-line texts written for computers on subjects such as Shakespeare, international cyber conflict, and fairy tales. Genesis compares stories; detects concepts such as love or revenge, even when they are not named; concludes whether a short-term gain leads to a long-term loss; and explains acts based on personality traits. The system can even analyze a text through a filter of cultural bias, thus interpreting an event like the cyber attack on Estonia by Russia in 2007 from the point of view of people in one or the other country.
Genesis today, however, remains “rudimentary,” a reflection of just how complex human intelligence is, Winston readily concedes. Winston has been modeling human intelligence since before 1997, when he and seven colleagues issued a clarion call for interdisciplinary focus on the subject, saying progress in artificial intelligence had been less profound than expected because so little is known about human intelligence from a computational point of view.
Even at that time, Winston was putting his trust in ThinkPads to do his work. They have been so close to his heart throughout his career that he keeps nearly a dozen of his old devices in his attic “for old time’s sake.” Like many other users, Winston praised the ThinkPad’s toughness, though he also relies on Lenovo’s robust warranty option to repair or replace a machine when broken through accidental damage. “I’ve spilled coffee on a keyboard, I’ve dropped a machine off of a ladder. I’ve done just about everything it’s possible to do to a machine.” he acknowledged with a laugh. “Most of the time they just survive, but if you spill coffee on a keyboard, that’s when the accidental damage warranty option is good to have.”
Currently, Winston has a Lenovo ThinkPad T460 that he hooks up to a variety of displays at home and in his MIT office. “I take it with me everywhere,” he said. “It goes with me on vacations. It goes with me whenever I go to a conference. It’s a companion.”
In particular, Winston “can’t live without” the red button in the center of the keyboard, known as a “TrackPoint” nub, which he uses to move the cursor. “It’s something I just have to have to work. If I should have a keyboard without a little red dot in the middle I’m just completely lost,” he confessed.
Despite the daunting nature of his goals, Winston has a lot of fun, and his work has taken him in many very different satisfying directions. The author of 17 books, Winston is past president of the American Association for Artificial Intelligence. In 1986, he co-founded a company that makes scheduling and resource allocation software. He has also advised government agencies on security and for 18 years was on the Naval Research Advisory Committee, rising to chairman.
Through it all, Winston doggedly pursues the grail of human understanding. He laughs at the general overhyping of artificial intelligence prospects in the near term, but there is a real sense of urgency to his mission.
“It seems inevitable that intelligent systems are coming,” Winston concludes. “I want to be sure they can be trusted to help us make the right medical decisions, retrieve the right legal precedents, and suggest the right diplomatic moves.”