Toil and trouble: How ‘Macbeth’ could teach computers to think

MIT professor Patrick Winston said stories are “a fundamental differentiating capability of us humans. And machines don’t have it yet.”
David L. Ryan/Globe Staff
MIT professor Patrick Winston said stories are “a fundamental differentiating capability of us humans. And machines don’t have it yet.”

Patrick Winston’s computer is learning about revenge, ambition, and murder. It knows that victory can make you happy. But it also knows you can’t be happy if you’re dead.

The computer had to learn these things in order to read “Macbeth” — or, rather, an extremely truncated version of Shakespeare’s blood-soaked Scottish tragedy. At just 37 sentences, the rough summary reduces the Bard’s immortal poetics to such clunkers as, “Witches had visions and danced” and “Lady Macbeth has bad dreams.”

But Winston, an MIT professor, thinks this bite-size “Macbeth” could help crack the biggest problem in artificial intelligence: how to build computer systems that can simulate the human mind’s unique powers of perception and insight.


Teaching computers to understand stories, as he sees it, could help shed light on crucial components of human intelligence that scientists don’t yet understand and cannot yet replicate in machines.

Get Metro Headlines in your inbox:
The 10 top local news stories from metro Boston and around New England delivered daily.
Thank you for signing up! Sign up for more newsletters here

“Stories seem to me to be the key to education, to social organization, to creativity, learning, consciousness, self-awareness — the whole works,” he said. “I think it’s a fundamental differentiating capability of us humans. And machines don’t have it yet.”

Now, as MIT makes a massive, schoolwide push for major advances in AI, Winston, 75, is hoping for a breakthrough on this problem, which has fascinated him since he was a doctoral student.

Computers can drive cars, beat people at complex games, and help doctors identify tumors by finding patterns and making predictions out of mountains of data. Some systems can even identify emotions from a person’s facial expression.

But even the most advanced artificial intelligence systems don’t know what it means to feel happy or sad. They struggle with metaphors, and they fall flat in conversations that haven’t been anticipated by their programmers.


“When you read your 3-year-old a story, they understand it well enough to laugh or to be frightened,” said Roger Schank, a well-known artificial intelligence researcher and professor emeritus at Northwestern University. “They can tell you what the story was about, which makes them smarter than any computer that we have.”

Many in the field are skeptical about whether science has advanced far enough to even begin to develop these capacities in computers.

“It’s a very difficult problem,” said Tomaso Poggio, a longtime colleague at MIT who works at the intersection of brain research and computer science. He says major advances in neuroscience must come first. “I don’t have the courage to touch it because it’s so beyond what I think I can do right now.”

Winston knows the challenge of simulated story understanding is an order of magnitude beyond the dramatic progress he’s witnessed in artificial intelligence since the 1960s. In the 25 years he spent as director of MIT’s Artificial Intelligence Laboratory, researchers there paved the way for innovations that were once far-off dreams, such as Apple’s Siri and the groundbreaking robots of Boston Dynamics.

But Winston remains convinced that computer science can plumb the complexities of human thought. Fifteen years ago, he started a project, now known as Genesis, that aims to use stories to break down human intelligence into tiny fragments of understanding.


He chose “Macbeth” as a primary text — Genesis has processed many other works — because the tragedy offers an opportunity to take big human themes, such as greed and revenge, and map out their components.

Getty Images/File
Actor Peter O'Toole played the title role in a 1980 production of “Macbeth.”

Shakespeare, he said, “was pretty good at his portrayal of the human condition, as my friends in the humanities would say,” Winston said. “So there’s . . . all kinds of stuff in there about what’s typical when we humans wander through the world.”

Genesis can digest short, simple chunks of text, then spit out reports about how it interpreted connections between events.

A web of rules and concepts helps it understand objects, actions, and how they relate to one another.

Some rules are always true: “If X kills Y, then Y becomes dead.” But others are possible explanations for why something might happen. Revenge, for instance, is described this way: “X is an entity. Y is an entity. X’s harming Y leads to Y’s harming X.”

Over time, these ideas have begun to imbue Genesis with rudimentary features of common sense. It can now use the concept of revenge to interpret a story.

In the final scene of “Macbeth,” the title character dies at the hands of Macduff, whose family was murdered by the Scottish king. Even though the story doesn’t say why Macduff did it, Genesis can point to revenge as the motive.

“We don’t kill everybody who angers us,” Winston said. “It’s only a connection that we’ll make if it makes the story more coherent and there’s no other explanation available for the killing.”

Winston has programmed Genesis to be able to answer written questions, as a student might, about what it has read. It can also create a visual representation of the events in a story and provide a description of how it believes they are linked.

Winston said Genesis is close to displaying something like self-reflection, where it would use its prior knowledge to explain its own process of interpreting a story. It can model the perspective of readers with differing cultural perspectives or points of view, a capability that Winston says could eventually help explain “why the same events can be told in two ways by two sides of an issue to justify very different reactions.”

But even after years of research, and contributions from some 50 of Winston’s students and colleagues, there’s still so much Genesis can’t do — learn on its own, for example. All the rules are supplied by programmers.

And Genesis is a long way from studying the actual text of “Macbeth,” or even the CliffsNotes. Winston had to dumb it down to a few dozen basic sentences.

So whereas Shakespeare’s play begins with the witches’ lyrical murmuring over the king’s fate (“When shall we three meet again?/ in thunder, lightning, or in rain?”), Winston’s starts with a bit of a thud: “Macbeth is a thane and Macduff is a thane. Lady Macbeth is evil and greedy. Duncan is the king, and Macbeth is Duncan’s successor.”

Kevin Coleman, director of education at Shakespeare & Company in Lenox, laughed when he heard this distillation of the play: “It’s like everything of value that Shakespeare is has been stripped away.

“It’s like I have fallen in love with this person named Mary, but how do I communicate my experience of Mary?” he said. “They’ve given me Mary’s height, and her weight, and the number of hairs on her head, and the length of her fingers. . . . That does not give you Mary.”

Taking in “the richness of Shakespeare,” in all its poetic and dramatic complexity, Coleman said, “is like waking up to our own human nature.”

In his own way, though, Winston is also using “Macbeth” to arrive at a deeper understanding of humanity.

“It’s explaining something big,” he said, “about where we are, who we are.”

Michael Fassbender played the title role and Marion Cotillard played Lady Macbeth in the 2015 film “Macbeth.”
Jonathan Olley/The Weinstein Company
Michael Fassbender played the title role and Marion Cotillard played Lady Macbeth in the 2015 film “Macbeth.”

Andy Rosen can be reached at [email protected]. Follow him on Twitter @andyrosen.