By Stephy Mahoney | Reporter
Baylor clinical professor of management information systems and business analytics, Alison Alford has created an AI project called Daughtering101 to analyze data using AI to code behaviors and emotions in a specific research group.
They have 10 interviewers conducting the research split evenly between them. The goal of their project is to assess how women feel about their roles as daughters throughout their lifetime.
“This is a random sample across the United States,” Alford said. “They could be Muslim, Mormon or even atheist.”
Although they are only conducting 300 interviews to actually test, Alford said they have had over 1,000 people apply to be assessed.
“The question we are asking everyone is, ‘What are some things you do well as a daughter?’” Alford said.
The data is collected, put into a spreadsheet, given to AI and then analyzed to train the Generative Pre-trained Transformer to think about the data in the way they want it to be analyzed, given the topic of research.
Alford said that they have made remarkable progress with the help of data collection and analyzing tool NVivo. NVivo is a tech software that is able to lump data together and analyze it all in a clean way. This is helpful when moving data from place to place.
“We are trying to teach the GPT to analyze virtues,” Alford said. “What we are trying to analyze is what virtues people associate with the role, and are there more virtues associated when you are also religious or spiritual?”
This is a way to let AI find patterns and use sentiment analysis. The AI can code the tone or temperature of the response.
“It’s also a great way to double-check our human inclinations,” Alford said.
Alford described how it can be hard to remain unbiased when analyzing responses of a subject group she is part of, being a daughter herself.
“I want to read the research with the eye of an insider, but I can also teach AI to read the research as an outsider and that might help reveal different patterns I can’t see,” Alford said.
Alford’s project manager and recent Baylor graduate Luke Stipanovic has been working alongside Alford since May to analyze data for Daughtering101.
“The most time-consuming part of the project has been getting the interviewers trained on procedures and data collection,” Stipanovic said.
According to Stipanovic, each interview is anywhere from 30 to 60 minutes long. As they get to know the interviewees, there are long transcripts produced during this time.
“It can be really tedious to hand-code all of these transcripts,” Stipanovic said. “Having tools like AI code and depict the language for you is really helpful.”
They have trained the GPT to section each tone of response into a subcategory. Stipanovic uses an example: if a woman was raised by a single mom, her responses would be recorded and sectioned out with other women of single mothers for cross-analysis.
“We are getting into the next stage of the project as we analyze the trends and data in the upcoming weeks,” Stipanovic said.
