Some Ohio State experts said they believe Artificial Intelligence can have positive impacts on STEM fields. Credit: Nevaeh Repp | Lantern File Photo
Algorithms are the backbone of everyday life. From social media feeds and streaming services to advertising and the business world, these computer instructions have become more sophisticated at collecting and calculating data based on user engagement.
Overall, people support customized feeds, but new research published in an experimental psychology journal led by Giwon Bahg, an Ohio State doctoral student, suggests that content learned from personalized algorithms can have a negative effect on how a user views reality.?
“In this experiment, we really wanted to focus on the effect of learning, rather than just evolving any prior belief or prior attitude or opinion on any certain topics,” Bahg said.
Brandon Turner, a co-author and professor in cognitive psychology and neuroscience, said people learn through using web browsers to look up information.
“We see people trying to acquire information, trying to genuinely learn about the world all the time. These days, that’s pretty much done through Googling things,” Turner said. “You send in some kind of query and you get some information that is curated by some kind of algorithm and what we wanted to know is, what are the effects of that?”
Bahg said the study used a “crystal-like aliens” scenario where participants would be presented with new fictional information to eliminate any previously-held beliefs.
“It is really hard to distinguish what is coming from your prior attitude, prior belief, or what is coming from a purely learning process and how that learning process interacts with an algorithm, so we really wanted to distinguish between the two,” Bahg said.
The study asked 346 online participants to identify different categories of aliens without knowing how many different categories there were. Among them, six features varied in slightly different ways from changes in color, to the orientation of shapes on the bodies. Features would be hidden behind grey boxes that participants could reveal throughout the course of the experiment.
One group of participants were given a choice to reveal all of the features as they wished, while in another group, participants were given a choice of features they could reveal, and from there an algorithm would repeatedly suggest the same feature based on the participants’ input, while also giving them the option to reveal all of the other features before moving on.
The results showed that participants who were given the choice by the algorithm sampled less of the overall features despite having the option to do so.?
When participants were tested and asked to identify the different aliens based on what they had learned, those who had been fed the information with the personalized algorithm answered incorrectly more often than those who had not, but, importantly, Bahg said that their confidence in their answers was higher when they answered incorrectly compared to when they answered correctly.
“You are not pursuing learning but still end up learning and remembering something based on this biased interaction, and you become confident about what you don’t understand,” Bahg said.
The researchers conclude that personalized algorithms can limit the “amount and diversity of information that learners access,” which can have real-world implications, as they said it can be hard to fix the issue of algorithmically-driven selective attention.
“Searching through Google and clicking the first thing that shows up, that’s not research, right?” Turner said. “That’s not really studying a problem and trying to understand the depth of it.”
Turner said that algorithms are sensitive to how information is distributed based on what keeps a user engaged for as long as possible and to “give you results that make you feel good and happy and entertained.”
“If the whole purpose of the algorithm is not tuned toward learning, but for profit and information consumption, then you might be not only exposed to really biased information, you end up learning something biased from them,” Bahg said.
The full text of the research can be found on the American Psychological Association website.
