Where Creativity Converges with Artificial Intelligence: The Fusion of Arts and Sciences

This article initially featured in the Fall 2024 edition of Arts & Sciences Magazine

Artificial intelligence has recently surged into public awareness, with chatbots like ChatGPT becoming familiar terms and major technology firms investing hundreds of billions of dollars into AI’s potential. Within university disciplines of arts and sciences, including the Krieger School, the influence of AI is also acutely recognized in research, scholarship, and pedagogy.

While significant advancements in AI as a technology predominantly occur at the industrial level, innovative uses are also proliferating in universities. Alan Yuille, whose investigations at Johns Hopkins center on artificial visual processing systems, believes this flourishing emerges naturally from the open, entrepreneurial ethos of academia.

“There are specific abilities that companies possess that we don’t because they have significantly more computing power and data,” Yuille states, a Bloomberg Distinguished Professor of computer science and cognitive science at both the Krieger School and the Whiting School of Engineering. “Conversely, at universities, we are at liberty to engage in our individual pursuits and investigate various ideas.”

Especially in the field of arts and sciences, where the focus is less on immediate profit and more on a human-centered, long-term perspective, Yuille recognizes the advantages of broadly engaging with AI in an academic context. This vast approach could assist in shaping the ongoing development of the technology.

“In arts and sciences,” Yuille notes, “it might be preferable to emphasize substantial, long-term research over studies that will quickly yield smartphone applications.”

Here are a few instances where AI is paving new paths for discovery and understanding at the Krieger School.

Image credit: STEPHANIE DALTON COWAN

Reviving the past

Known today as a property in the board game Monopoly, the Baltimore and Ohio (B&O) Railroad was the nation’s first commercial railway, operating for over 150 years, from 1830 to 1987. Fortunately for historians, the B&O Railroad Museum in Baltimore has maintained a vast collection of insurance documents the company kept for its thousands of personnel, originating in the early 1900s. Unfortunately, these records—a jumble of approximately 16 million handwritten and typed pages—had remained boxed for years due to the impracticality of cataloging them efficiently by human archivists.

Understanding the scale of the obstacle, the B&O Railroad Museum consulted with professors and researchers from Johns Hopkins. Word of the extensive data repository reached Louis Hyman, the Dorothy Ross Professor of History. A labor historian, Hyman teaches a class titled AI and Data Methods in History, which investigates the application of AI technologies as research tools.

Utilizing this knowledge, Hyman and other Johns Hopkins researchers are now collaborating with curators at the B&O Railroad Museum on a project aimed at scanning and digitizing the expansive database. The initiative employs optical character recognition, a form of computer vision, to interpret text in the documents. Researchers subsequently process these texts through large language models to rectify unavoidable computer errors while more significantly extracting and organizing data in a manner that allows for effective searching and aggregation.

“It’s a unique database,” remarks Anna Kresmer, the archivist at the B&O Railroad Museum, “and until now, there hasn’t been a feasible approach to tackle this data outside of AI.”

Developing research prospects

Once made available, the B&O database is expected to unlock a myriad of research opportunities in labor, business, and medical history. Jonathan Goldman, chief curator of the B&O Railroad Museum, anticipates that data-driven approaches will unveil broad trends arising from thorough analyses of the data, such as patterns regarding specific types of injuries or the treatment of certain groups.

“These documents depict the lives of countless individuals,” Hyman concludes. “Patterns can emerge, revealing truths about the past that we perhaps overlooked previously.”

In the upcoming years, Goldman

indicates that there exists a vision to render the digitized files available at public terminals in the museum, empowering guests to perform genealogical inquiries regarding relatives who were employed by the railroad. “We aim to assist certain families in discovering more about their forebears, including some insights into their identities and experiences,” explains Goldman.

Project historians predict that AI techniques will become commonplace for addressing other similarly vast collections of textual materials. “AI possesses an extraordinary endurance for monotony that no human can match,” Hyman remarks. “Employing AI clarifies what is significant about human nature [during research], which isn’t about sitting down and transcribing countless pages—it’s about engaging in thoughtful and critical analysis.”

Image credit: STEPHANIE DALTON COWAN

Developing unexplored materials science

The laboratory of Tyrel M. McQueen focuses on the creation of new materials for societal benefits and to gain understanding of fundamental science. In pursuit of this goal, artificial intelligence is emerging as a valuable instrument to propel discoveries.

For example, Queen served as co-principal investigator on an initiative named MITHRIL—an inventive acronym for Material Invention Through Hypothesis-unbiased, Real-time, Interdisciplinary Learning, inspired by a fictional metal in The Lord of the Rings. The initiative, along with subsequent research derived from it, has utilized AI to quickly investigate a multitude of material combinations. For human researchers, analyzing millions of molecular configurations is unfeasibly time-intensive and inherently limited, as researchers tend to build on established knowledge rather than adventuring into random, undocumented territories.

In partnership with the Johns Hopkins Applied Physics Laboratory, McQueen and colleagues from the Krieger School recently employed MITHRIL in a comprehensive quest for new superconductors. When subjected to very low temperatures, these distinct materials transmit electricity without any energy expenditure. While superconductors find numerous applications today, such as in maglev trains and MRI machinery, researchers have long sought high-temperature superconductors that could transform electricity transmission, energy storage, and other fields. “We aspire to discover superconductors that operate at room temperature,” McQueen, a professor in the Departments of Chemistry and Physics and Astronomy, notes.

Algorithms propelling progress

The researchers fed MITHRIL data regarding a few hundred known superconducting materials, training its machine learning algorithms to predict whether other substances would exhibit the sought-after property. The model performed remarkably well, “rediscovering” four recognized superconductors and uncovering a fifth novel one. “Swiftly, we identified materials that are known superconductors but were not part of the training dataset,” McQueen shares. “We also discovered a new superconductor that has never been created before.”

The scientists moved forward to synthesize and examine this newly proposed superconductor, an alloy composed of zirconium, indium, and nickel that starts to function as a superconductor at approximately 9 Kelvin, or -443° Fahrenheit. While not revolutionary at such low temperatures, the finding still illustrated AI’s potential to drive progress in materials science research.

In a related effort, McQueen is also employing AI to enhance the identification of truly novel material concepts to increase the likelihood of discovering previously unimagined substances that might be groundbreaking. AI models typically provide two outputs—what the model believes is the correct answer and the level of confidence the model has in that answer’s accuracy. McQueen asserts that the most valuable results are those where the model has the least certainty about a material’s specific qualities.

“That’s the material you should create, because that’s the area of chemical space that hasn’t been delved into for that property yet,” McQueen explains. “You apply these methods to confer significantly better insights than any human could achieve.”

Image credit: STEPHANIE DALTON COWAN

Educating with AI

One of the most significant effects of AI is observed in higher education, where student utilization of chatbots continues to spark intense discussion. Numerous individuals are concerned that chatbots might encourage widespread dishonesty along with poor academic practices, given their tendency to provide incorrect information. Conversely, others are optimistic that these AI tools could exemplify a form of information technology that, if employed correctly, boosts the learning experience.

In order to assess optimal AI practices in student education, sociologists at the Krieger School are committed to examining how undergraduate students engage with chatbots. “AI necessitates a novel kind of information literacy,” states Mike Reese, head of the Krieger School’s Center for Teaching Excellence and Innovation and an associate teaching professor in the Department of Sociology. “It’s crucial that we grasp how students are utilizing these tools.”

Investigating student-AI engagement

To this purpose, Lingxin Hao, a professor in public policy within the Department of Sociology, alongside Reese and colleagues from the computer science department in the School of Engineering, is spearheading innovative research on campus involving undergraduates. A significant study was carried out over three weeks last fall. The researchers conducted an experiment with participants in PILOT, a peer-led tutoring initiative where groups of approximately six to ten students meet weekly to collaboratively work on problem sets.

Initially, the researchers developed course-related questions using a Johns Hopkins software-based quiz instrument for students to respond to. A number of student participants were randomly assigned access to ChatGPT, the widely-used online chatbot, and were further separated into groups that could either inquire ChatGPT a single question or multiple questions over rounds of interaction. All student prompts directed to the AI and the received responses were meticulously documented, so the entire interaction could be analyzed. “We were able to observe how humans and ChatGPT truly connect,” says Hao.

In total, nearly 700 students participated, providing the researchers with a comprehensive dataset to scrutinize student behavior and performance with ChatGPT. Encouragingly, the researchers noted that some students—particularly those who indicated familiarity with chatbots—demonstrated proficiency in utilizing the tools as aids; not merely as means to jump to potentially erroneous conclusions, “but as resources for how to approach a problem and learn,” asserts Hao. “The objective is to comprehend the social science underpinnings of human-AI interaction through the gathered student-ChatGPT discussions.”

Establishing guidelines

The researchers are building upon these and additional findings to formulate guidelines for teaching with AI. The study aims to assist instructors in effectively and fairly integrating the emerging technology into their classrooms. Doing so can not only propel educational objectives but also prepare students with a foundational knowledge for their future careers.

“It is essential for us as faculty to determine how AI tools will be utilized in our field, whether within academia or in industry, and subsequently introduce them into our classroom so that students can be better equipped to use AI in their future professions,” explains Reese.

Image credit: STEPHANIE DALTON COWAN

Improving social care

For over two decades, through her research in impoverished urban communities in Chile, Clara Han has directly witnessed the difficulties of delivering social care services to at-risk populations, particularly children. Social service personnel frequently lack adequate resources and time to devote to each case—this issue is magnified by insufficient data sharing among institutional entities such ashealth systems and welfare organizations.

“Frontline personnel are consistently faced with the necessity of making choices regarding resource distribution and determining the priority of one case over another, yet there’s simply not enough to meet every demand,” remarks Han, a faculty member in the Department of Anthropology.

Consequently, the coordination of care frequently falls short of optimal standards, resulting in individuals requiring assistance falling through the gaps, often with dire outcomes. Han illustrates this with the case of a child with a health issue who enters foster care without her doctor being notified by the agency. Since her newly appointed social worker lacks access to her prior health information, the child ultimately misses doses of a vital medication.

Han is keen on examining both the regulatory frameworks that influence data access as well as the consequences of current applications of predictive algorithms in social services. To this goal, alongside pediatricians at Johns Hopkins University School of Medicine and other colleagues across the institution, Han is constructing a research framework aimed at examining the ethical and legal implications of AI-driven decision support constructs for healthcare coordination in Baltimore City.

Assessing AI’s boundaries

Such decision support methods have become progressively prevalent in immigration and criminal justice realms. It is increasingly evident that predictive algorithms can negatively affect decision-making dynamics by dramatically “amplifying” biases present in social services. “There’s what is termed ‘analog harm,’ where individuals remain in a state of uncertainty,” Han elaborates.

Grasping the regulatory landscape is essential to enhancing the integration of data from various sources, Han states. Simultaneously, there is a growing acknowledgment of the limitations of metrics when navigating intricate social environments. “AI is capable of inflicting harm, just as any decision-making process can lead to damage,” asserts Han. Thus, Han is intrigued by how AI elements, as part of the decision-making framework, address the legitimacy of various stakeholders’ assertions, even if those assertions are conflicting. In light of these varied assertions, any resource allocation model must incorporate a mechanism for human appeal.

In summary, Han envisions significant potential for cross-disciplinary research on AI, in which the knowledge of frontline providers and their patients is actively integrated into the possibilities provided by algorithmic decision-making. “We aspire to advance our research framework,” Han states, “because it could yield tangible advantages for individuals in Baltimore and elsewhere, where the technological and digital landscape may vary significantly, affecting data-sharing regulations.”

Image credit: STEPHANIE DALTON COWAN

The past informs the future

Henry Farrell aims to comprehend AI’s prospective effects on the international order, especially as large language models (LLMs) become deeply rooted in everyday life.

Farrell, the SNF Agora Institute Professor of International Relations at the School of Advanced International Studies (SAIS), reminisces about his startling initial experience with a rudimentary, earlier version of an LLM several years ago. While the model facilitated a Dungeons & Dragons-like online roleplaying game, it occasionally lost awareness of the user’s character and produced odd, even inappropriate responses and assertions. Nevertheless, the underlying power and novelty of the technology deeply impressed Farrell. “It was undeniably very flawed,” observes Farrell, “but it was also something that was genuinely unprecedented in the world.”

Analyzing social information

Farrell has subsequently examined LLMs within the realms of technology and political science. With the performance of LLMs having vastly improved over the last few years, alongside widespread acceptance, Farrell and his team are currently investigating how LLMs and machine learning more generally serve as tools for processing social information.

This ongoing evaluation is guided by comparisons to two historical, drastically transformative social technologies: market systems and bureaucracies. Both frameworks, extensive in reach and intricacy, assimilate colossal amounts of information—similar to the often-Internet-wide training datasets of LLMs.

In the market system, all harvested information is distilled into supply-and-demand-driven pricing for products and services, whereas bureaucracy condenses the totality of societal activities into categories for governmental supervision. LLMs execute a comparable process by scrutinizing human knowledge repositories to deliver succinct responses to user inquiries.

In their vastness, all three technologies surpass human comprehension and control, engendering both dread and reverence, Farrell states. “Much of the anxiety and optimism associated with AI mirror the concerns and expectations we have had about markets and bureaucracies,” asserts Farrell. Some of those anxieties have become reality, evident in significant job losses as entire sectors relocate or become redundant due to market dynamics, or when governmental agencies fail to effectively serve their constituents. However, numerous hopes have been realized as well, from the wide availability of food and consumer goods to successful government initiatives.

Thus, to gain insights into AI’s implications today and in the future, Farrell and his co-workers are extracting lessons from history. “We can derive some understanding by examining these earlier social technologies and the types of disruptions they caused,” says Farrell. “This at least helps us begin to pose the right questions, potentially offering us a clearer sense of what lies ahead.”

This page was generated programmatically; to read the article in its original context, you can follow the link below:
https://hub.jhu.edu/2025/01/07/arts-and-sciences-meets-ai/
and if you wish to remove this article from our site, please contact us