Michael Shadlen: Eyes, Angel Dust, Alan Turing, and Decision-Making in the Brain

Michael Shadlen. Credit: John Abbott/Columbia-Zuckerman Institute 

There is something very wrong about their eyes.

Clearly overdosing, the patients enter the hospital’s emergency room confused, paranoid, and often ranting. Hovering over them, psychiatrists, police, and doctors feverishly work to pinpoint the chemical tormenter. It’s a hallucinogen, that’s obvious, but which one? The patients’ eyes tell all. They dart to take in a stimulus, but instead of returning to their former location they drift, slowly, eerily back to center. They cannot control their eyes.

It’s a form of gaze nystagmus, a kind of involuntary twitching of the eye, and it’s a telltale sign of the street drug phencyclidine, also known as PCP or Angel Dust, a powerful dissociative hallucinogen and frequent culprit in this California ER. But for neurologist and AAAS Fellow Michael Shadlen, the patients’ eyes point to something more than PCP. The eyes take him on a path that helps unlock the neurological foundations of decision-making.  

Shadlen remembers his overdosing patients during his hospital residency in San Jose in the early 1990s, “It was like they couldn’t hold onto a thought to connect the immediate past to the immediate future. And I started thinking, ‘You know, what we’re seeing is a really basic function of the brain that involves evidence accumulation.’”

Over the years, Shadlen has been accumulating his own evidence. He is now a leading brain researcher with his own lab at Columbia University. His abiding interest is decision-making. Through studying the brain’s response to external stimuli, Shadlen has essentially reverse-engineered the calculations happening inside the skull when a decision is made. His starting point is the eyes; conscious, controlled eye moment, the very thing his “Angel Dusted” patients were incapable of doing.

From the hospital corridors, Shadlen’s research traveled a winding path that included monkey brains and the statistics that helped defeat the Nazis.

“I didn’t care about vision per se,”  Shadlen said about his initial decision to study perception. His experience in the emergency room aside, Shadlen said his interest in eye movement was largely practical because examining eye movements is a great way to study decision-making and the brain.

“Eye movements are super fast and super precise, and they feel effortless. They can reduce decision-making to an elaborate sensory-to-motor conversion,” said Shadlen.

During the same period that Shadlen was completing his residency in San Jose he also worked at nearby Stanford University in the lab of pioneering brain researcher William Newsome. By attaching electrodes to the neurons of monkeys, Newsome and his associates were examining the signal-to-noise relationship inside the visual cortex. (The monkeys, incidentally, aren’t hurt by this procedure. There are no pain receptors in the brain.)

Once attached and connected to the lab’s equipment, the neurons made a kind of clicking noise, reminiscent of a Geiger counter. To parse out the clicks’ spacing and amplitude, the researchers developed a series of mathematical models. Shadlen said this effort created the foundation for his future work.

“The computational ideas from that led to thinking about how the brain might be interpreting the data, or evidence you might say, from the visual system in order to reach a decision. Bill [Newsome] and I started that together,” said Shadlen.

To understand how this works, consider a typical Shadlen experiment.

A monkey is placed in front of a screen displaying a series of flashing dots. The dots indicate movement in a particular direction, for instance right or left. The monkey has been trained to indicate a direction. He does this by resisting moving his eyes until he’s confident that he knows the direction of the motion.

Meanwhile, electrical activity starts picking up inside the monkey’s brain, which is recorded by an electrode that’s been attached to a single neuron that tracks directional motion. The neuron itself is in the association cortex, a part of the brain that intervenes between sensation and action. The dots flash. The neuron starts firing. The electronic clicking starts. This is the sound of the monkey’s brain weighing the available information.

The clicks start slow and then pick up speed. Within seconds, they are rising to a climax. Evidence about the direction of movement has reached a critical point of accumulation. A decision has been reached. The monkey moves his eyes either right or left, indicating the conclusion his brain has come to. Uncovering the mathematics at work inside the monkey’s brain is another story.

It’s now very much en vogue to talk about the brain as a kind of computer that runs the equivalent of statistical models to understand and interact with the world. But in the late 1990s, when Shadlen, then at the University of Washington, picked up a statistics book written by mathematician I. J. “Jack” Good, he did so more out of general interest than an attempt to prove anything.

“The reasons I had for reading Jack Good had nothing to do with what I got out of it,” said Shadlen. What he got out of it was quite a bit.

Good was a colleague of Alan Turing, the famous British mathematician considered one of the fathers of modern computing. Good and Turing worked together during World War II at Bletchley Park, the secretive British government headworkers where they and other code breakers cracked the Nazis’ Enigma machine, a kind of mechanical, proto-computer that encrypted messages about the location of German submarines and other strategically important information.

Good’s book outlines how Enigma was cracked by employing a variation of Bayes’ theorem, a formula used to calculate the probability of a given outcome as evidence accumulates. In a nutshell, Bayes’ theorem allows for beliefs to be updated as new information comes in, it also allows for accurate predictions to be made based on imperfect information, a must have for code-breaking of all kinds. Reading Good, Shadlen saw a clear analog between Bletchley Park’s code-breaking and his monkeys’ decision-making. He then took the next step: He applied the Bletchley Park equations to the clicks he was hearing in the lab. They fit. 

Shadlen published his findings in 2002 in the journal Neuron. Since then, Shadlen and colleagues have published multiple studies that build on this statistical decision-making model. These have included studies revealing how the brain calculates statistical confidence levels about choices, how the brain makes trade-offs when both time and information are limited, how the brain perceives time itself, and even how decision-making might relate to humans’ ability to create and appreciate art, music, and storytelling.

Shadlen said he hopes that the simple decision-making processes he has helped reveal inside the visual cortex will help scientists better understand more complex decision-making processes throughout the brain. Always the clinician, however, his ultimate desire is to help people, like his emergency room patients years earlier, who suffer when basic neurological decision-making goes awry. In 2010, he summed up his thoughts during a lecture he gave for the Allen Institute:

“I view the 25 million years that separate our brain from the monkey brain not as time in which we developed really brand new ways of doing neural computation, but rather by having bigger brains, we can make decisions about decisions, effectively cascading the processes.”

Meet More AAAS Members