Can AI bridge disciplinary boundaries?
Is AI the solution to interdisciplinarity’s problems?
This is the second post in a new living literature review about interdisciplinarity. You can read more about the project here.
“Human researchers can barely scratch the surface of the potential of interdisciplinary transfer due to its cognitive difficulty and the very large number of combinations of similar societal problems. If AI transfer learning could be applied to IDR, it could open up vast possibilities for learning about societal problems…” – Baum, 2020
The artificial intelligence (AI) revolution brings both promise and perils across many domains, and its potential to influence the boundaries between disciplines is no different. As discussed in a previous post, is interdisciplinarity on the rise?, there are loud calls among researchers and practitioners to bridge the boundaries between sectors and disciplines through multidisciplinarity, transdisciplinarity, and particularly, “interdisciplinarity,” with significant societal and institutional buy-in. Among areas of hope raised by interdisciplinarity’s proponents, many anticipate that large language models, innovations in data processing tools, and other AI-facilitated technologies will help reduce the barriers that maintain disciplinary and sectoral siloes, unlocking potential for societal progress and innovation.
Is AI the solution to interdisciplinarity’s problems? Can technology, especially tools that help facilitate data knowledge accumulation, sharing, and access, serve as the key to bridging the boundaries that have otherwise proven so persistent and sticky in modern research and practice?
AI is in many ways fertile testing ground for efforts to bridge disciplines. Scholars increasingly call for multiple disciplinary perspectives to inform AI tool development. They often call for inputs across the hard sciences, social sciences, and humanities to shape AI tool development, AI safety and related issues. In particular, scholars and practitioners increasingly have brought together engineers with those in social science and ethics to monitor and inform AI tools for bias and discrimination.
The Center for AI Safety highlights this call for inputs from multiple disciplines, writing on its website, “We pursue conceptual research that examines AI safety from a multidisciplinary perspective, incorporating insights from safety engineering, complex systems, international relations, philosophy, and other fields…”
AI may also offer promise itself for facilitating cross-disciplinary research and practice through various processes– not only by serving as a site to apply multiple disciplines, but also by itself providing tools for researchers to use to break down disciplinary barriers. By helping bring together large volumes of data and information and expanding access to that information and speed with which it may be process, large language models and other tools offer potential to help researchers access and process more information, more quickly, which may facilitate greater opportunities to engage across disciplines.
The AI revolution also comes with risks that could harden disciplinary distinctions and expand siloes. AI’s basis in large data and quantitative models and the methods used to process information could build up the access to data and related methodologies at the expense of human-led qualitative reasoning, leading to fears that AI could challenge threaten the demise of qualitative research, or at least would harden the distinctions between particular fields where AI tools will have different effects.
Will AI help bridge these disciplinary boundaries, either by applying AI tools to research, by applying interdisciplinary research to AI questions, or both? Can AI facilitate what scholars and practitioners call for by way of expanded networked knowledge communities, in place of ongoing sectoral siloes?
This post explores the state of knowledge on the intersection of AI and interdisciplinarity, highlighting the status of insights around promising practices and potential pitfalls.
The importance of interdisciplinarity for AI
Interdisciplinary collaborations relating to AI might improve research quality in a variety of ways. Some posit that disciplinary collaborations between AI specialists and other disciplines can help foster innovation and help solve societal challenges.
A study by Abbonato et al (2024) explored the question of whether interdisciplinary collaborations that involve AI and medicine are particularly impactful, noting the rise of studies starting in 2021 in which scholars “quickly embraced the idea of adopting AI techniques to tackle the challenges presented by COVID-19” (the authors cite DeGrave, Janizek, & Lee, 2021, Khan, Mehran et al, 2021; and Roberts, Driggs et al, 2021 as part of this hype). Abbonato et al’s study reviewed the impact of around 15,000 interdisciplinary papers bringing together AI researchers and medical researchers covering issues related to COVID-19, noting a rise in collaborations between AI researchers and medical researchers with the rise of the pandemic in 2021 (Figure 1). Their paper seeks to explore whether interdisciplinary research teams were particularly impactful. They model the question using two primary measures of research impact– the number of citations received by an interdisciplinary publication and the Altmetric attention scores of these publications – and one measure of “interdisciplinary spread” based on the team’s coding of the disciplinary composition of research teams.
Figure 1
Their findings muted optimism that AI was the key to research impact. They found that these interdisciplinary collaborations between medical professionals and AI specialists largely resulted in publications with “low visibility” and “low impact.” They posit that interdisciplinary research teams that engage AI researchers with other disciplines face a variety of challenges when applying AI techniques to other disciplines, challenges that include:
· Poor data quality
· Lack of global standards and database interoperability (the ability to translate formatted data across multiple platforms)
· The inability of algorithms to work without sufficient knowledge domain
· Overly exacting computational, architectural, and infrastructural requirements
· Legal and ethical opacity associated with privacy and intellectual property
These challenges may be mounted if advancements are made, for example, to synergize data standards improve data quality, in turn improving the potential to apply AI tools to a variety of disciplines. Beyond this, the authors caution against viewing disciplinary homes of authors as a measure of the potential of AI collaboration impact, and rather, suggest that the quality and depth of interdisciplinary knowledge alongside diverse research methods and framings may ultimately drive research impact. They write, “[w]hat appears to ensure the impact of a publication is, above all else, is the interdisciplinarity of the knowledge mobilized via its references, that is the actual epistemological diversity of the research conducted by a team” (p. 10). Their study suggests, in sum, while interdisciplinary research involving AI holds promise for advancing impactful research, blunt efforts to bring AI researchers together with other disciplines will not alone advance research impact and could even have adverse effects.
If one contends that interdisciplinarity is critical for AI-related research, what are some relevant best-practices for doing so?
Another study exploring the intersections of interdisciplinarity, AI, and research quality from Bisconti et al (2022) probed how to maximize interdisciplinarity’s benefits for AI research. Considering the importance of various disciplinary inputs to the development of the European Union’s Artificial Intelligence Act the authors, a diverse disciplinary group of eleven researchers, begin by asserting interdisciplinarity’s importance for AI policy. They write, that work on AI “requires an interdisciplinary approach to balance the technical issues and forecast biases, paying attention to the weight that these measures have on policy decisions” (p. 1444). Assuming the imperative of ensuring interdisciplinary inputs for AI, they probe potential good practices for applying interdisciplinary research inputs to AI governance, noting various challenges to applying interdisciplinarity in practice. These include divergences they noted in their research between different conceptions of AI among researchers spanning different fields. They call for researchers to build consensus around an agreed “methodology to formalize an interdisciplinary process aimed at increasing the synergies of ICWR research groups dealing with AI systems with urgent social implications” (p. 1451).
“Without a process theory that enables effective group communication and understanding, interdisciplinarity is not sufficient to ensure that the solutions and implementations of AI systems are addressed in the best manner.” – Bisconti et al, 2022
Their proposed methodology (Figure 2 below) focuses on bridge-building through common narrative and semantic exercises that build a common rhetoric and framework for research. Their method for interdisciplinarity in AI research based on this process is as follows:
· Stage 1: Facilitate the communication of different understandings on the same issue, and work in interdisciplinary teams to define hypotheses.
· Stage 2: Set up a common lexicon.
· Stage 3: Narrow down the possible applications and implications of the AI system in a real-world scenario.
· Stage 4: Conduct follow up assessments, with developed metrics on how significantly “interdisciplinary” goals were researched, to inform further iteration.
Figure 2
The importance of AI for interdisciplinarity
Just as AI-related research may benefit from interdisciplinarity, interdisciplinarity may, too, benefit from AI tools. A paper from Baum (2020) explored the question: in what ways can artificial intelligence assist with interdisciplinary research for addressing complex societal problems and advancing the social good? Baum contends that interdisciplinarity is critical for work on many modern societal challenges, writing, “[p]roblems such as environmental protection, public health, and emerging technology governance do not fit neatly within traditional academic disciplines and therefore require an interdisciplinary approach” (p.545). Baum then notes a number of challenges to the application of interdisciplinarity in practice, including: Institutional divides that discourage disciplinary engagement and collaboration; Large literatures making access to vast amounts of external knowledge unrealistic; Peer review systems that are a barrier to bridging specialism; Challenges in transferring knowledge from one domain to another.
Introducing the concept of “Artificial interdisciplinarity (A-ID)” – “Artificial intelligence that performs interdisciplinary research or supports other agents in the performance of interdisciplinary research,” he probes the areas where AI may help advance interdisciplinarity already given the current state of technology. He finds the following potential areas where AI may facilitate improved interdisciplinarity:
· Search engines: AI can help challenge disciplinary divides by providing tools to explore vast literatures in other disciplines.
· Recommendation engines: AI can provide recommendations customized for individual user profiles, as currently exists through Google Scholar and Elsevier.
· Automated content analysis: Applying machine learning to analyze literature content.
He then identifies “Immediate-term” A-ID systems, those that go beyond current capabilities in the near future. These include:
· Interpretation: Using AI tools to interpret research publications through LLMs.
· Translation: Helping researchers translate knowledge across epistemic divides, including between languages (e.g. English to Spanish) as well as conceptual divides.
· Transfer: AI can help transfer knowledge gained in one domain to another with similar features.
Finally, Baum suggests that “Long-term A-ID” with distant future progress through artificial general intelligence (AGI) may significantly improve the potential for interdisciplinarity, though projections are inconclusive, and a variety of ethical dilemmas also emerge.
Ongoing Challenges
While there appears to be significant consensus around the importance of interdisciplinarity for AI research, studies variously explore the practical challenges for doing so. Kusters et al, 2020 in their paper raise the question: How can an interdisciplinary approach towards AI benefit from and contribute to the AI revolution? They call for viewing the relationship between interdisciplinarity and AI as a “two way street,” recognizing that AI should be applied to a variety of fields, and that AI should be informed by a variety of fields. Citing a number of examples, most prominently Frontier Development Lab, they insist that efforts to bring multiple disciplines together productively through AI research efforts is “not utopic,” and already achievable. They call for ramping up similar efforts.
Among challenges, the authors call out reforming educational systems to better meet the needs of this “two-way street” will be critical. They write, “Of course, combining AI with other fields is not without challenges. Like any time when fields synergize, barriers in communication arise, due to differences in terminologies, methods, cultures, and interests. How to bridge such gaps remains an open question, but having a solid education in both machine learning and the field of interest is clearly imperative.”
A question remains in how to scale these efforts that aim to bring AI and interdisciplinarity together. To what degree might all disciplines require some grounding in topics related to machine learning? And, if they must, how might this be institutionalized within educational systems at scale?
Final thoughts
Scholars continue to debate the merits of disciplinary boundaries, but overall, calls remain loud to bridge them. Is AI the solution? AI appears to offer potential towards a “two-way street” model. In terms of applying interdisciplinarity to AI research, the field has proven fertile testing ground for applying multiple disciplines to inform and shape its development and regulation. AI may also itself help facilitate the bridging of other disciplines by offering new tools for processing and accessing large amounts of information.
AI will likely facilitate revolutions in the ways that scholars can access and process information. As AI improves the speed and scale with which scholars can interact with a diverse set of inputs and information, the once restrictive barriers of time and resources for processing information that once made working across disciplines too onerous could reduce and weaken, presenting new opportunities. Yet, as scholarship reviewed in this entry has noted, new challenges emerge, with AI applications not yet enabling coherent and synergistic processes that fully bridge disciplinary boundaries, and the potential requirement for education systems to advance alongside AI development. With the rapid pace of technological advancement in AI, further study of the multiple ways that AI intersects with interdisciplinarity will require further documentation and scrutiny.
Articles Cited
Abbonato, Diletta, Stefano Bianchini, Floriana Gargiulo, and Tommaso Venturini. “Interdisciplinary Research in Artificial Intelligence: Lessons from COVID-19.” Quantitative Science Studies. https://doi.org/10.1162/qss_a_00329.
Baum, Seth D. “Artificial Interdisciplinarity: Artificial Intelligence for Research on Complex Societal Problems.” Philosophy & Technology 34 (2021): 45–63. https://doi.org/10.1007/s13347-020-00416-5.
Bisconti, Piercosma, and Davide Orsitto et al. “Maximizing Team Synergy in AI-Related Interdisciplinary Groups: An Interdisciplinary-by-Design Iterative Methodology.” AI & Society 38 (2023): 1443–1452.
DeGrave, A.J., Janizek, J.D. & Lee, SI. AI for radiographic COVID-19 detection selects shortcuts over signal. Nat Mach Intell 3, 610–619 (2021). https://doi.org/10.1038/s42256-021-00338-7
Khan, Muzammil, Muhammad Taqi Mehran et al. “Applications of Artificial Intelligence in COVID-19 Pandemic: A Comprehensive Review.” Expert Systems with Applications 185 (December 15, 2021): 115695.
Kusters, Remy, and Dusan Misevic et al. “Interdisciplinary Research in Artificial Intelligence: Challenges and Opportunities.” Frontiers in Big Data 3 (2020). https://doi.org/10.3389/fdata.2020.577974.
Roberts, M., Driggs, D., Thorpe, M. et al. “Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans.” Nat Mach Intell 3, 199–217 (2021). https://doi.org/10.1038/s42256-021-00307-0.