Research professor explains how bias factors into computational learning

  • 0
  • 2 min to read
Brain-to-brain communication

When President Donald Trump made his presidential announcement speech and stated that Mexicans are “bringing drugs. They’re bringing crime. They’re rapists,” how did that affect our personal prejudice against Mexicans? 

Dr. David Amodio, a professor of psychology at New York University and the University of Amsterdam, opened his Nov. 6 lecture with this question.

“Many of us dismissed it as just mere political rhetoric, just words, not something we actually believe,” Amodio said of Trump’s statements. “But nonetheless, just by hearing it, this kind of information still gets inside our heads.”

Implicit prejudice, also called implicit bias, is the set of stereotypes people hold that unconsciously affect their actions. Amodio first studied how people learn socially, specifically through instrumental learning, to help him study how we internalize these implicit prejudices.

Instrumental learning is when you complete an action, get feedback, and it’s encoded in your brain based on its reward value; it also operates unconsciously. This is repeated, and you learn through the combination of these interactions.

For example, after playing a few rounds on a slot machine, you may expect it to give you a certain amount of money, Amodio explained. When on the next round it instead gives you more money, you update your expectations for a higher payoff.

Amodio connects human instrumental learning to a computational model using reinforcement learning. In the same way a person tries to learn which slot machine will give them the most reward, a computer can do the same thing using reinforcement learning.

After discovering that we could test how we learn about humans in the same way we learn about slot machines, Amodio began to alter these parameters.

“When you use models like this you could start playing with the parameters of the models to test [a] specific hypothesis,” Amodio said.

In one study, participants decide between two avatars on their screen and pick the one that will give them the most money. Before the task starts, however, the instructions provide a stereotype about the different types of avatars. Group A is described with positive traits such as being trustworthy, honest, and generous, while group B is described with negative traits such as being poor, hostile, untrustworthy, and criminal.

The results showed that participants consistently trusted group A with their money over B, even though both groups had equal probability of sharing their money with the participant. 

Further, when another group of participants was asked to watch the choices of the previous participants, they also trusted group A without ever reading the stereotypes provided to the first group.

Compared to a standard learning model with no biases, a computational model that accounts for both biased learning and biased priors better predicts the results of the study.

“This is the signature of prejudice, where you’re making decisions based on some group representation and not from the experience that you’re having with the individuals,” Amodio said.

Amoido shared his prediction of the biological causes of implicit bias. A message that Trump shares gets encoded into our Anterior Temporal Lobe (ATL). The ATL helps with our memory and knowledge of the world, and we have no control over it. 

Another part of the brain is the striatum, which handles instrumental learning and which interacts with the ventromedial prefrontal cortex (vmPFC) which handles our expectations. These two systems are in charge of the reward reinforcement. The vmPFC has a large nerve track connected to the ATL, meaning the information stored in the ATL could influence our decision making.

Dr. Amodio shared that his team is currently working on testing real-world stereotypes, like black and white in America or white Dutch and Moroccan in the Netherlands.

“If there’s influence of this message on expected value, that could bias the expectations that we have when we go to interact with a person from another group,” Amodio said. “This is just one first step in this program of work but this is starting to demonstrate a mechanism for the internalization of implicit group knowledge and implicit choice preferences.”

Reach contributing writer Kaya Bramble at science@dailyuw.com. Twitter: @KayaBramble

Like what you’re reading? Support high-quality student journalism by donating here.

(0) comments

Welcome to the discussion.

Keep it Clean. Please avoid obscene, vulgar, lewd, racist or sexually-oriented language.
PLEASE TURN OFF YOUR CAPS LOCK.
Don't Threaten. Threats of harming another person will not be tolerated.
Be Truthful. Don't knowingly lie about anyone or anything.
Be Nice. No racism, sexism or any sort of -ism that is degrading to another person.
Be Proactive. Use the 'Report' link on each comment to let us know of abusive posts.
Share with Us. We'd love to hear eyewitness accounts, the history behind an article.