As artificial intelligence continues to infiltrate everyday life, IU’s researchers are focused on developing these technologies, while working to ensure their research is safe and ethical. The Luddy Center for Artificial Intelligence is set to open this month, providing researchers a place to focus on the intersection of robotics, complex networks, health and social media.
Kay Connelly, Luddy School’s associate dean of research, studies proactive health and AI technologies that can help the terminally ill and older people as they age, specifically wearable devices. She said proactive health is like “Fitbit before Fitbit.”
Her work is key in one of IU’s Grand Challenges, and it tackles how to get the right treatment to patients at the right time. To research how to treat gestational diabetes, women are given wearable devices to track their heart rate, sleep and physical activity. The hope is to prevent development of the disease and stop it from becoming Type 2 diabetes.
Gestational diabetes diagnoses come during a woman’s pregnancy, according to mayoclinic.org, and the data collected from these wearable devices allows researchers to detect which women are at higher risk to develop gestational diabetes.
Former Luddy School dean Raj Acharya’s work in AI also influences the medical community, especially as variants of COVID-19 continue to arise. Acharya currently has a National Science Foundation grant dedicated to reconstructing the genome of viruses using DNA and RNA sequencing, which is essential to the vaccine development process.
He takes short, cut-up strands of DNA and RNA, then uses the information they contain to solve questions about the full strands, which scientists are not able to fully read.
He said viruses want to fool the immune system — this is why viruses mutate — so AI and machine-learning techniques are used to reconstruct viruses to understand the structure and order of characters which make them up.
He said the interaction between our immune system and viruses is a type of game theory. When a virus enters our body, our immune system learns and stores a sequence of the virus so it can fight it, but recombination of its character makeup allows the virus to trick our immune system.
While Acharya uses sequencing for developments in the medical field, AI can also help researchers understand the evolutionary process called phylogenetic analysis, which is used to construct evolutionary trees showing which animals evolved from another.
IU computer scientist David Crandall’s expertise is computer vision, the part of AI that tries to get computers to recognize things and make decisions based on what they see. Self-driving cars use this kind of research because they need to recognize things like stop signs and be able to make the decision to stop.
This process uses machine learning, which trains computers to recognize things based on a large data set. Machine learning is an important development because in the past it would take months or even years to program specific commands, but now machine learning can expedite the process by having the computer teach itself, Crandall said.
Crandall said studying how children learn has been influential to this type of research. In the same way children begin to understand things by observing them over and over, so do the programmed machines.
This is also how programs like Alexa learn to respond to a person’s command. Amazon engineers have given the computers powering Alexa thousands of collected audio clips and taught the algorithm how to recognize certain commands, Crandall said. These are examples of the black box model, which is a system used to study the inputs something is given and what outputs it gives back.
Beth Plale, a professor and executive director of the Pervasive Technology Institute at IU, does research in data provenance, which is essentially data auditing. Her job is to study algorithms and their similarities to unconscious human bias, which can be encoded because of structural issues in our society, she said.
Plale said studying the outcomes of AI is important because if their work can be misused, people can use machine-learning algorithms to teach robots to be bigoted and do bad things. She brought up Microsoft’s conversational bot Tay, which was pulled from Twitter because just hours into its launch users had taught it to be racist and use inflammatory political speech.
AI needs to be accountable, she said, and her work is focused on looking into issues that arise and mitigating them, like unconscious biases in algorithms or people’s ability to manipulate outcomes. She says while trust in science is still high, transparency in the work researchers are doing is critical. Researchers need to jump on the important conversations early so they don’t become politicized.
“There’s only so much a researcher — an academic researcher — there’s only so much we can do or say because there are forces that squelch our voices,” Plale said.
As Michael McRobbie’s tenure as Indiana University’s president was coming to an end this June, one of his final moves was to dedicate the Luddy Center for Artificial Intelligence, a new “state of the art” facility for advanced AI research and machine learning.
The $35 million center is part of a $60 million gift from alumnus Fred Luddy, who is the namesake of IU’s school of informatics, computing and engineering. Construction began in February 2020 and it is set to open this month.
According to a news release from IU, the center’s initial focus will be robotics, complex networks, health and social media.
“The explosion worldwide of the uses and applications of AI, building on decades of steady research progress, made this the perfect time for IU to establish a major holistic initiative in artificial intelligence,” McRobbie said.
During McRobbie’s tenure he also launched the Grand Challenges program, a $300 million investment into solving some of Indiana’s most pressing issues. These challenges include research into precision medicine, environmental change and addiction crises.
While artificial intelligence research can lead to breakthroughs that make people’s lives easier, IU’s faculty also is concerned about what consequences it will have and what conversations need to be had as the field develops.
Both Plale and Crandall point to automated military drone strikes as an example of ethical problems society will have to grapple with. This discussion is particularly relevant in Crandall’s field of study, because computer vision would be used in such drones, and he said AI is notoriously bad at understanding things out of context.
“I think they struggle in certain circumstances. For example we know that they struggle often when they encounter situations that they’ve never encountered before,” Crandall said.
For example, a sticker on a stop sign may cause an autonomous car to not recognize it or act accordingly. The stakes are even higher when considering autonomous drones having the power to kill people without human intervention, and Plale said the ethical implications of whether military drones should have the power to do so is a problem society will have to face.
She tied this in with the conversation of the moral considerations a sentient AI robot would be given if researchers are able to develop one. The moral considerations are broad questions that AI researchers should contribute to the discussion of, she said, but ultimately society will be responsible for coming up with the answers.
However, Connelly believes IU and universities in general are a good place for ethical discussions because there are diverse voices at the table and the structure of funding and research is different compared with a corporation.
“You find that people who are advocating for the ethics often get pushed aside, or pushed out completely because it doesn’t fit in that profit model,” she said.
Corporations also tend to ignore the populations Connelly’s research is focused on because there’s less profit incentive in low-income communities. Her research is specifically targeted to making sure people with increased risk of health disparities get access.
In addition to the less-profit-based research IU is able to engage in, Connelly believes the university setting is perfect because diverse voices will be heard and be able to point out when research is short-sighted or developments don’t take into account their communities.
She mentioned problems with facial recognition not working for people of color in some cases as a potential way more diverse voices can shape the future of AI. She believes IU does a good job mentoring students from all backgrounds, including first-generation or students with a non-English primary language, because IU’s apprenticeship-style research helps students learn the “nitty gritty” of the field.
“They are bringing their perspectives to the table, so students are critical,” Connelly said.
In terms of research resources, IU is well situated thanks to top-notch researchers in all fields of AI and having medical professionals and experts within other schools at IU, Acharya said. He said with IU’s Grand Challenges there is a lot of cross pollination between the Luddy School and IU Health faculty in Indianapolis.
In the three ways of looking at AI — algorithmic or computer science, cognitive science and hardware — IU has great people in all three, Acharya said. Plale also pointed out IU’s well positioned infrastructure, including Big Red 200, a supercomputer designed to support research in artificial intelligence, machine learning and data analytics, which was installed in January 2020.
Plale noted Crandall’s research in the cognitive science department is a collaborative effort with the psychology department, where studies are looking at how babies learn, which allows the cognitive science department of the Luddy School to apply those methods to machine learning development.
“I think what it suggests is that the collaborative activity that goes on in the academic setting can infuse new thinking,” Plale said.
Another collaborative effort Connelly brought up was IU’s Observatory on Social Media, a joint project between the Luddy School, The Media School and Network Science Institute. This project unites data scientists and journalists to study the role of media and technology in society.
Connelly worries about manipulation of entire segments of the population, which she believes algorithms on social media can propagate. She sees it as a direct threat to democracy. She said people fall into silos of information and when that happens you’re not exposed to a breadth of information and alternative viewpoints.
In 2017, former FBI agent and cybersecurity expert Clint Watts testified before Senate Intelligence Committee about the role Russian bots played in the 2016 presidential election. He said Russians used “armies” of Twitter bots to spread misinformation about the election, and in 2017 he’d already been tracking this kind of activity for more than three years.
Connelly said everyone is susceptible to this regardless of their views, and she thinks social networks can enforce seeking approval and conforming to whichever way you lean on issues. When you start to sway on issues, those silos are an effective way to pull you back in.
Artificial intelligence has been a part of science fiction media since at least the 19th century, and in the same way science fiction has had different portrayals of how AI will become part of our lives, IU’s researchers have different attitudes toward the depiction of AI in literature and film.
Crandall, for example, rolls his eyes at a lot of the depictions of AI in the media because it’s very dramatic, especially concerning things like an AI ignoring its programming and thinking for itself.
For example, recent movies like “Ex Machina” and “Her” explore the potential relationships between humans and sentient forms of AI. “Ex Machina” (2015) ponders the idea of the creation surpassing the creator and whether an AI can escape the black box, and “Her” (2013) explores whether humans can fall in love with an AI.
“The state of robotics is that just doing something like folding a piece of clothing is like, beyond what any robot is able to do right now in any reasonable amount of time,” he said.
While the more dramatic depictions of AI are a bit laughable to him, Crandall recognizes the more subtle ways AI can control parts of people’s lives, such as using Facebook to influence elections. He said it seems innocent, but it can become something that controls what people see. While this is much more subtle, Crandall said conversations are needed about these negative impacts and where AI goes from here.
Plale did not want to speculate too much about media representation, but she is also skeptical about some of the sci-fi portrayals of AI. She said from what she’s seen, AI in the media has been given much more sentience than what exists today, and while robots in factories are very well developed, the ability to get them to process things and react the way a human brain does is limited.
Acharya is a bit more optimistic about the role of sci-fi in the development of AI, and thinks in many ways the genre helps researchers be more imaginative.
“I think in a way that the media people might be ahead,” he said. “The science fiction is probably ahead because the scientists are constrained by what can happen today. Media people can imagine.”
As artificial intelligence continues to carve out a role in our lives, IU’s researchers are determined to make the future of AI a positive influence on the lives of Hoosiers. To make sure they usher in technology as a social good, Plale believes the conversations on ethics and the research being conducted need to move forward simultaneously.
Acharya brought up deepfakes, which use a form of artificial intelligence called deep learning to make realistic-looking images of fake events, as an example of AI being abused by people with bad intentions. In September 2019, AI firm Deeptrace found that 96 percent of deepfake videos were pornographic, and Boston University law professor Danielle Citron said “Deepfake technology is being weaponized against women” because it can fuel things such as hate porn.
Still, Acharya is optimistic about the role AI will play in the lives of people moving forward. He said continued AI research in the medical field will be important in the lives of retired people, as it will allow more of them to live more fully as they age and develop medical problems. Human and machine becoming one is the ultimate goal, he said, in the sense that research into artificial limbs and organs can extend and have a positive impact on people’s lives.
Also, hypothesis generation, a new type of AI application which speeds up the process of developing hypotheses and comes up with new ones researchers might have overlooked, helps make research more efficient, Acharya said. Overall, his outlook on the future of AI is positive.
For research as a whole, one issue Crandall sees is some researchers’ eagerness to present things without full context because they want their work to seem like a breakthrough. It’s not anyone’s fault, he said, and it might not even be a bad thing because it gets people talking about the work, but the hype about some developments in AI aren’t always centered in reality.
“The reality is much more boring,” he said.
People have also blown the hype around AI’s impact on the job market out of proportion, Crandall said, and he compared it with the same type of hype around the internet being a job killer when it came out. Forbes reported that in a Gallup poll from 2018, 73 percent of Americans believed AI would be a net job destroyer, but only 23 percent of them were worried about it, partially because they didn’t believe their jobs would be affected.
“It’s hard for me now to really predict what really the impact will be of AI,” Crandall said. “Except that like, I think it really is something that we have under our control.”
He brought up self-driving trucks as an example of something that could push people out of certain jobs, but he’s curious what effect that will have on the job market as a whole. Automating 99.9 percent of what goes into driving is easy, he said, but in the 0.1 percent of cases where unexpected things happen, AI is not good at anticipating and reacting the way a human would.
He said he’s comfortable with the current situation of trusting the people around him to make responsible and rational decisions while he’s driving.
“Somehow we’ve grown accustomed and we’ve grown comfortable with this situation, which when you think about it that way is quite terrifying,” he said. “Every time you get on the road you have to trust everyone else who’s driving to do the right thing.”
He thinks on one hand AI is good because you can program it to make decisions, but problems arise because society has to grapple with what decisions it should make. What or who an automated car should focus on protecting when a crash can't be avoided altogether is one example of the discussions Crandall says need to be had.
“I think the tricky part is we’re going to have to decide, ‘What is the right thing for it to do?’” he said.
©2021 the Herald-Times (Bloomington, Ind.). Distributed by Tribune Content Agency, LLC.
