HopAI held its inaugural event on Thursday, Nov. 29. The organization, which seeks to connect and expose Hopkins students to artificial intelligence (AI), invited three speakers from different areas of study to describe their work with the diverse technologies.
HopAI’s president, junior Jiali Zhang, emphasized that artificial intelligence covers a wide variety of fields. However, she feels that communication across disciplines is inadequate. The new student group, HopAI, is supposed to improve this.
“This is sort of the goal of HopAI: to bring together Hopkins to talk about AI,” Zhang said. “It’s a very interdisciplinary discussion that needs to happen in order to ensure safe usage.”
Each speaker came from a different field. The first speaker was Dr. Anirudh Sridharan, a doctor at Johns Hopkins HealthCare who worked with University researcher and Computer Science Assistant Professor Suchi Saria to implement her artificial intelligence-driven solution to identifying cases of sepsis. Saria’s program is also used at another community hospital.
Before Saria’s program, Sridharan’s hospital relied upon programs that use hand-coded rules to identify sepsis. However, it was only accurate one in 10 times.
“You’d imagine that both nurses and docs get very frustrated with a system that’s wrong 90 percent of the time,” Sridharan said. “When we applied these alerts to the hospital, within a week everyone just started ignoring them because they weren’t useful.”
Sridharan said that the previous system was so intrusive that it inhibited many doctors’ workflow. He spent much of his presentation discussing how the hospital administration tried to convince its doctors to use his new program, called a targeted real-time early warning score (TREWScore). This required important decisions about how TREWScore was used.
“What we tried to do with TREWScore was to try to make [the notifications] more passive but leverage them in places where clinicians were more likely to see them,” Sridharan said.
Sridharan said that a lot of thinking went into preventing TREWScore from becoming intrusive, like the old system. As a result, usage of TREWScore, even though optional, is above 85 percent in his hospital.
“We worked to make this a lot more easy for clinicians to see who is sick, who is at risk — without having to stop them in their tracks,” Sridharan said. “They can go and do other critical things and then go and see where their patients are along in the sepsis path.”
According to Sridharan, the new program is currently able to correctly identify sepsis 50 percent of the time at his hospital. The ability of the code to learn allows it to improve as time goes on.
Sridharan also explained that each hospital that uses an algorithm trained on their own data to personalize it.
The second speaker, senior Aliza Berger, presented about her interactions with artificial intelligence policy while working in the State department as a cyber policy intern.
“This [presentation] is trying to show that everyone should be thinking about AI, whether it’s nation-states, NGOs or the humanities and STEM,” Berger said.
Berger said that the department suffers from a lack of technological skills.
She also noted how very few nations have artificial intelligence frameworks, and that those that do have created them recently: Canada in 2017, Japan and then Singapore were the first nations to establish plans.
“This isn’t necessarily about laws as it is about a nation’s strategy,” Berger said.
According to Berger, many nation-states’ frameworks focus on different areas. Some care more about the economic usage of artificial intelligence, whereas others focus on its military applications. Other nations worked on plans that include improving government efficiency. The U.S., however, does not have a coherent cyber policy of its own.
“You might be surprised to hear this, but we actually do not have a national strategy,” Berger said. “Shouldn’t this be concerning? China, who we are allegedly in an AI Cold War with, is thinking in the very long term. But here we are, with no strategy. One of the reasons for this is the difference in political ideology.”
Berger stressed that a result of American capitalism is the belief that technological innovation should come from the private sector. Berger indicated that the Chinese government, in contrast, did not have a problem with directing companies to invest in artificial intelligence.
“It’s a very different conversation here,” Berger said. “Whereas in China, with the current political system, the private-public sector relationship is very different. In the United States, we don’t have a political strategy, but we do have a task force.”
According to Berger, the U.S. has by far the greatest number of artificial intelligence specialists in the world. She thinks that to spread this talent, there needs to be a collaborative international framework.
“My questions is: why can’t we do this together?” Berger said. “Why can’t we think about an international framework?”
The final speaker was Chief of the Intelligent Systems Center at the Applied Physics Lab (APL) Ashley J. Llorens. His center focuses on using artificial intelligence to address a wide variety of situations.
“There are so many different and challenging problems [here at APL],” Llorens said. “Working in AI means that you’re actually working at the intersection of many of those problems.”
Llorens says that he broadly thinks of AI as any agent used to pursue complex goals. Artificial intelligence has many layers: He says that machine learning is a subset of AI, then deep learning is a subset of machine learning. He stressed that the field is much larger than most people believe.
“Many people think of a convolutional network [a type of deep learning] as everything that AI has to offer, but it’s actually much broader,” Llorens said. “It involves robotics, theorem-proving and many things that have been around for decades. A lot of recent pursuits in AI are combining the old and the new to create something better.”
Llorens also talked about several technologies that APL has developed or improved using artificial intelligence. The first was Sally, a robot that investigates whether oil spills have occurred. The second was about Johnny, a wounded warrior who, thanks to the APL, could control an artificial limb that is almost as good as an original.
“Your hand and forearm have 27 degrees of freedom, and this has 26. It gets you back most of that,” Llorens said.
Johnny, who was sent home with his artificial limb for a year, has even been learning how to play the piano with his new limb.
The next story was about Jan, a quadriplegic who could control an arm using a brain-computer interface.
“Being a quadriplegic and wanting to feel that capability again, she was willing to be a patient in this experiment. She had a craniotomy, which is having part of your skull removed,” Llorens said.
The sensor on her brain is connected to a device that has a machine learning algorithm on it to infer motor intent from brain signals.
As the very first official event of HopAI, Zhang sees a lot of room to grow.
“My hope for HopAI is that we’ll be able to gather enough resources to support students, undergrads and grad students’ ideas about AI solutions to current world issues,” Zhang said.