33 School superintendents along the U.S. East Coast had a daunting decision to make this week: With a massive winter storm forecasted, they had to weigh whether to keep the schools open or to close in anticipation of the poor weather. Many factors go into the decision, such as temperature and anticipated snowfall, but there are also a host of other more subtle factors like the expectation of road safety based on the last storm and how many other good calls they’ve made this year.
Whether deciding on a snow day or choosing a new car, making decisions is a complicated task. Neuroscientists are illuminating human decision-making, using computational simulations to test how people make judgments in a range of conditions. They are finding that a big part of this process is how people allocate their limited cognitive resources in rapidly changing environments – making for efficient, but not always optimal, decisions.
Chris Summerfield of the University of Oxford, recipient of a CNS Young Investigator Award, will be discussing his latest research on human decision-making at the upcoming CNS conference in San Francisco. He spoke with CNS about what makes a good decision and some of the ways his lab tests decision-making.
Decision-making is a ubiquitous topic in experimental psychology and neuroscience, because all behaviors – from riding a bicycle to buying a house – involve decisions.
CNS: What makes a good decision?
Summerfield: Researchers disagree over what makes a good decision. Psychologists have traditionally assumed that good decisions are those that maximize accuracy in laboratory-based tasks. However, economists remind us that we are all strongly motivated by the likely outcome of our choices – i.e. by whether we are likely to receive a rewarding stimulus (for example, a financial bonus) or not as a result of our actions. Behavioral ecologists take the argument a step further, pointing out that the bottom line for an animal is that it needs to survive and reproduce.
Normally, these three factors – accuracy, reward, and survival probability – are aligned. But sometimes they are not. For example, a human participant in a laboratory experiment might perform badly in order to complete a boring experiment as quickly as possible, or an animal that is close to death might engage in a risky behavior that is unlikely to succeed, because they have nothing to lose in that situation.
CNS: How did you become interested in studying human decision-making?
Summerfield: I was fortunate to study a variety of different topics during my graduate and postdoctoral training, including visual perception and attention, memory, and the “executive” functions that humans use to perform high-level tasks. Decision-making is a ubiquitous topic in experimental psychology and neuroscience, because all behaviors – from riding a bicycle to buying a house – involve decisions. The topic thus brings together my diverse interests in the cognitive and neural sciences.
I also like studying decision-making because decades of research have provided a wealth of computational models that describe how decisions might be made. This allows researchers to formulate and test very clear predictions, by comparing human data to that of computational simulations. This makes it easy to do hypothesis-driven research, which I think we as cognitive neuroscientists should all be doing more of.
CNS: Can you explain the “perceptual classification tasks” that your lab studies?
Summerfield: In my lab, we mostly study perceptual categorization judgments, that is how humans classify features and objects into two or more classes. We focus mainly on the visual domain. For example, a foraging animal might have to decide whether a fruit is good or bad to eat on the basis of its shape and color. Humans are primates, and the primate visual system is very highly developed, so some perceptual decisions that can be challenging for machine vision systems, such as deciding whether a face is male or female, are trivially easy for us.
CNS: How do you test this in the lab?
Summerfield: In the lab, to make perceptual decisions challenging, we usually create stimuli that are blurry, indistinct, or composed of variable or contradictory features, and ask humans to classify them by pressing buttons. Participants in our experiments learn from feedback to which category they belong. By varying the parameters of stimulation, we are able to carefully control the level of “noise” or variability in the stimulus, and thus the difficulty of the category judgment. This allows us to measure how well people are performing relative to a simulated observer that makes the best possible judgments given how discriminable the categories are. These “ideal” observers are not always correct. For example, if good apples and rotten apples looked exactly alike, then even a perfect observer would choose rotten apples at least 50% of the time.
CNS: How do you find that type of decision-making compares to making economic decisions?
Summerfield: This is a very good question, and one that we do not have complete answers to yet. One traditional perspective from the literature is that humans tend to be very good at perceptual classification judgments (e.g. distinguishing a male from a female face), but often make biased or suboptimal choices in the economic domain (e.g. buying the wrong car, or marrying the wrong person). However, the two types of decision have been studied in very different ways, and it is not clear whether we are really better at one class of decision than the other, and if so, then why. Our working hypothesis is that the two types of decision depend on shared neural mechanisms and computational principles. But evidence for this is lacking as yet. We and others are currently working on this question.
CNS: How does expectation shape our decision-making?
Summerfield: We are interested not just in whether people make mistakes, but in the sort of mistakes they make. For example, some of our experiments have shown that humans exhibit systematic biases to respond one way or another. Many of these biases have to do with the expectations that humans have about what will occur in the task. If these expectations are wrong, then decisions will be biased. For example, we have shown that people performing sensory categorization tasks give more credence to information that is consistent with an evolving belief about the category from which the stimuli are drawn.
Although these decision biases are suboptimal from a formal perspective, often they are quite sensible. For example, if the brain’s processing power is limited, it makes sense to focus computational resources on those examples that are most expected, because these will often be hardest to tell apart.
For example, imagine that you are marking exam scripts to decide whether students pass or fail a difficult exam. Student marks, like much of the information our brains have evolved to process, are distributed in a a bell-shaped fashion, with most falling in a bulge in the middle (expected) and fewer falling at the extremes (unexpected). Thus, while some students will clearly pass (one hopes) and others will clearly fail (unfortunately), but the majority will be somewhere in between close to the pass/fail boundary. An “ideal” observer would allocate equal resources to all scripts, but an “efficient” observer might focus resources – e.g. marking time and effort – marking on those that are “expected,” close to the boundary between pass and fail.
Some of our work has shown that the brain seems to have evolved this sort of strategy when making perceptual decisions. However, this is not a new idea – in fact, it was first proposed in the 1950s, but has come back to prominence only relatively recently.
CNS: Does your research yet offer any suggestions for how people should go about making better decisions?
Summerfield: We hope that the research we are conducting will help identify ways in which humans can make better decisions. For example, returning to the distinction between “perceptual” and “economic” decisions, the quality of these decisions is often probed in very different sorts of tasks. Recent work has suggested that when financial decisions are presented in a format similar to a standard perceptual classification task, some of the suboptimal biases typical of economic decisions are less pronounced. We are working on ways to exploit this to help improve human decision-making.
-Lisa M.P. Munoz
Summerfield will give his award lecture on Sunday, March 29, 2015, 1:30 –2:30 pm, in the Grand Ballroom A in the Hyatt Regency San Francisco.