image title

The role of AI in higher education

June 22, 2023

ASU faculty members share their thoughts on how AI can shape education today and in the future

In recent decades, the advent of machine learning and neural networks — among other key advancements — to build artificial intelligence systems that mirror human capabilities has dramatically shifted many landscapes, ushering in a new era of AI characterized by adaptability and learning capabilities.

Today, AI’s fingerprints can be found everywhere — from voice assistants in our homes to advanced research tools in our labs. Education is no exception. In the not-so-distant future, AI-driven personalized learning platforms are set to reshape the learning experience, using personal data to tailor content based on each student's needs. 

RELATEDThe journey to learn ChatGPT

We talked with three faculty members at ASU to gain insight into how AI is currently shaping education across the university and beyond.  

Speaking from both their unique and shared experiences in academia, three central themes emerged, along with specific insights to advance our understanding of AI tools in the classroom.

1. Guiding AI transformation with digital literacy

The promise of generative AI in higher education is compelling. It can bring abstract ideas to life through visual aids, enhancing teaching and learning. It can handle routine tasks, freeing educators to focus more on teaching and students' individual needs. 

“I think it’s a marvelously exciting time to be a student, an educator, a researcher or an artist because we have to fundamentally look at our basic assumptions and ideas about how we interact with each other and our environment,” says Lance Gharavi, a professor in the School of Music, Dance and Theatre, part of the Herberger Institute for Design and Arts.

The challenge, Gharavi points out, is ensuring that we actively guide this transformative process.

“We have to ask ourselves why we’ve been doing things a certain way for decades, and maybe rethink our approach — and that is both exhilarating and anxiety producing.”

But where do we draw the line?

“There’s a problem with the limiting principle — a rule or guideline that tells us when to stop or how far to go with something,” Gharavi says. “On the one hand, we have something as simple as artificial intelligence like spell check, and on the other, we have something as sophisticated as ChatGPT — it’s a complicated conversation. Because if we’re not acknowledging the messiness, we’re not being honest.”

All three faculty members underscored the importance of teaching digital literacy alongside the integration of AI tools. Digital literacy serves as a compass of sorts, guiding users to ensure that AI tools do not inadvertently strengthen biases or become channels for malpractice and propaganda.

“From the very beginning, propaganda has been a part of our nation's fabric," says Retha Hill, director of the New Media Innovation and Entrepreneur Lab at the Walter Cronkite School of Journalism and Mass Communication"But today, we're seeing persuasive, AI-generated visuals accompanying targeted messages.

“AI can fabricate news articles filled with plausible-sounding narratives and fabricated quotes, and people tend to believe them. Now more than ever, understanding and discerning the credibility of content is more crucial than ever.”

2. A closer look at embedded bias 

The faculty members also emphasized the need for diverse input, continued development and regular audits to ensure AI tools are not perpetuating biases — but it’s not that simple. Intelligent learning machines, like people, are inherently complex.

“What we see now in generative AI is an offshoot of a whole generation of researchers trying to solve what is called the object recognition problem, where the goal was to take a picture, detect objects and give it a name and category,” says Pavan Turaga, director and professor in the School of Arts, Media and Engineering.

AI technology today is the result of many researchers aiming to teach computers to recognize and categorize content, Turaga says, but issues occur when this technology is used to generate images of people, especially when those people fall into uncommon categories.

To showcase this, Turaga asked the generative AI system DALL-E, which uses AI technology to generate high-quality digital images from text prompts, to create an image of a “non-binary math teacher in a classroom.” Turaga, who uses they/them pronouns, said it produced an image that is not accurate or fair, likely because the AI’s learning examples for this category are limited and may not represent the true diversity of the category.

“Human beings always spill over into buckets, like race is not cleanly defined into six buckets — there are mixed-race people; gender is not two buckets or even three buckets — it's considered a spectrum,” Turaga said. “So human attributes often defy the notion that things can be in nice little buckets. So how do you get past that? That's the big question.”

3. Fostering opportunity and the future of AI  

So, just how much is generative AI set to reshape the landscape of higher education? And more specifically, how do we navigate AI transformations with integrity?

In the realm of content creation, Turaga sees AI not as a looming threat but as a tool for helping their students to stay competitive. 

“Content creation is going to be impacted in a different way than say a writing program,” Turaga says. “Our students are actively addressing AI tools, because we have to be at the cutting edge of content creation — no matter what the tool is — and right now it is AI.”

Hill says there is a need for transparency in the deployment of AI tools, especially regarding attribution. 

“It’s important we lead with transparency and we can accomplish that in part with proper attribution,” she said. “In the projects where we have used generative AI, we say that it was created through MidJourney, or Stable Diffusion, or whatever the tool is — so we're upfront about our attribution and sourcing.” 

But all three faculty members acknowledged AI’s potential to enhance teaching and learning, as long as we keep an eye on the ethical implications. 

“We find ourselves in a time that is both exhilarating and frightening. Change is scary and it's exciting,” Gharavi says. “I think the potential to radically transform higher education is really there, and we have to make sure that we are working vigorously to steer how that change happens.”

Written by Kevin Pirehpour, Enterprise Technology.

Top image generated using the generative AI platform MidJourney. Prompt: “Educators envisioning the future of AI-enhanced universities and society, with tech developments, suspended in air, swirling around." 

image title

ASU study sheds new light on inconclusive gun evidence

June 22, 2023

Inconclusive results on cartridge-case comparisons can be predictive of innocence

When bullets are fired from a criminal's gun, ejected cartridge cases can scatter around the shooter. And unless that person is fastidious enough to gather those cases, they serve as a valuable tool in solving a crime.

Microscopic markings on ejected cartridge cases that match the gun used in a crime can usually provide enough conclusive evidence to land a suspect in the slammer. 

But what happens if the results are inconclusive? 

In court cases, inconclusive decisions on cartridge-case comparisons can be a contributing factor in putting innocent people behind bars, and the lack of rigorous research in the field of forensic ballistics can keep them there. 

A new study by Arizona State University researchers may help change that.

The study, which was published in the Proceedings of the National Academy of Sciences, found that 85% of cartridge cases that were judged inconclusive by forensic firearm examiners were actually fired by two different guns. In an actual crime scene investigation, that would mean that the cartridge cases did not match the gun in question. 

“The study was the first to closely examine the usefulness of inconclusive decisions in helping to clear the innocent of wrongdoing,” said Max Guyll, one of the researchers behind the study who is an associate professor of psychology in ASU’s new School of Interdisciplinary Forensics.

“What we found was that inconclusive decisions were predictive of, or related to, cartridges that are fired from different guns,” Guyll said.

Portrait of ASU professor

Stephanie Madon

According to Stephanie Madon, the principal investigator on the study and a psychology professor in the School of Interdisciplinary Forensics, some prior studies either ignored inconclusive decisions when examining firearm identification or inappropriately treated inconclusive decisions as correct decisions.

“What made our study unique is that we focused on the exculpatory probative value of inconclusive decisions,” Madon said. 

Coming to an inconclusive decision

When a bullet is fired from a firearm, it creates dents, tiny grooves or tool markings that make up a fingerprint of sorts on the cartridge case. If the markings on the case match those on a gun, it is considered an “identification.”

In some cases, when forensic examiners cannot definitely match a cartridge case and the gun being investigated for the crime, forensic examiners will use the term “inconclusive.”

According to Guyll, there are many reasons for an inconclusive result.

“They (the cartridge cases) may come from different guns or they may come from the same gun, but marked differently because a suspect tampered with the weapon before the police confiscated it,” he said. 

Cartridge cases could also hold slightly different amounts of gunpowder, or changes may have occurred in the gun due to wear and tear and debris before police took possession of it, Guyll said.

The study found that examiners rendered an inconclusive decision more than one-fifth of the time. Examiners also tended to disproportionately apply “inconclusive” to different-source cartridge cases over same-source cases — 35% percent of the cartridge cases that had been fired by different guns were judged inconclusive while only 6.5% of cartridge cases that had been fired from the same gun were judged inconclusive.

But, back in the courtroom, "inconclusive terminology might not communicate to jurors that inconclusive decisions are more likely exculpatory than inculpatory," Madon said.

“You don’t want to be an innocent guy sitting in a courtroom when results come back inconclusive,” Guyll said.

Portrait of ASU professor

Max Guyll

The aim of the study

According to the researchers, the study was born out of a National Academy of Sciences report released in 2009 that lamented the lack of rigorous research in forensic science to support its use in criminal cases (except in the case of DNA).  

“The message of those reports was that they couldn’t say which techniques were valid and which were not because they had not yet been rigorously tested,” Madon said.

So the study, which was funded by the National Science Foundation, provided much-needed research on the validity of cartridge-case evidence. It showed that when not accounting for inconclusive cases, the examiners accurately said that the cartridge cases matched or didn't match more than 99% of the time.

As part of their study, researchers gathered cartridge cases from ammunition coming from two different types of guns and then provided each examiner with eight sets of four cartridge cases to examine. Of the four cartridge cases in each set, three came from the same gun. The fourth may or may not have been fired from the same gun. Participants were tasked with deciding if the cartridges came from the same gun.

The examiners used microscopes to compare the tool markings on cartridge cases.

The results of the study were based on 1,811 decisions made by 228 trained firearms examiners from across the country. The experts analyzed cartridge cases left by 7,244 bullets fired from 28 guns. 

The researchers hope that their findings will help prosecutors decide which cases to pursue and encourage defense attorneys to use inconclusive results as a possible basis for post-conviction actions.

“This is new information,” Madon said. “New evidence that wrongly convicted people can now (be used) for their post-conviction arguments.”

Top photo from iStock

Dolores Tropiano

Reporter , ASU News