Beyond Diverse Datasets: Responsible MIR, Interdisciplinarity, and the Fractured Worlds of Music

2.1 An Ethical Turn?

Given recent advancements and controversies involving AI technology, the ethical implications of integrating such technology into public, private and commercial spheres have become issues of compelling interest to individuals, companies, and governments (Jobin et al., 2019). This has led to the creation of research forums like the ACM Conference on Fairness, Accountability, and Transparency,¹ crowd-sourced initiatives like the AI Incident Database,² the formation of corporate ethics committees, inquiries and reports by government bodies,³ and focus groups of professional global organizations.

The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems⁴ consists of engineers from six continents, and has produced three editions of “Ethically Aligned Design: A Vision for Prioritizing Human Well-being with Autonomous and Intelligent Systems” (IEEE, 2019). The latest edition of this document (EAD1e) argues for the development of autonomous systems guided by eight ethical principles: human rights, well-being, data agency, effectiveness, transparency, accountability, awareness of misuse, and competence. EAD1e supports designing with values that “put human advancement at the core of development of technical systems”, in concert with the recognition that “machines should serve humans and not the other way around … to create autonomous and intelligent technical systems that enhance and extend human well-being and freedom” (IEEE, 2019, p.6). These principles align with several AI guidelines produced around the world (Jobin et al., 2019). The majority of them, however, originate from countries that are economically developed, which incurs the risk of neglecting meaningful local knowledge and value systems from other regions, thus jeopardizing global fairness (Jecker and Nakazawa, 2022). This is acknowledged in EAD1e that emphasizes the need for cross-cultural dialogue so as to encompass a wider range of value systems (IEEE, 2019, pp.49–58).

Ethical frameworks further gain importance in contexts in which the application of AI is not considered to be high-risk, as it will in these cases not be considered by regulatory legal frameworks. As Clancy (2021) argues, since copyright as it stands is not capable of matching AI, an equitable approach to the financial implications of AI can only occur through an ethical response from stakeholders of the music ecosystem. But who would be in the best position to drive such a response? Following the argument by Gold et al. (2022), scholarly conferences can play a central role in defining ethical values of their community’s collective identity and aspiration. As such, the ISMIR community would be in an ideal position to extend its research activities to the pursuit of equitable models for music ecosystems (of different kinds and sizes) in the era of artificial intelligence. However, as analyzed by Holzapfel et al. (2018) and Morreale (2021), the ISMIR community has throughout its history largely ignored ethical implications of MIR technology. In order to perform the needed “ethical turn”, as Morreale further argues, the ISMIR community would need to first conduct an epistemological turn. Morreale’s proposal resonates with the earlier call by Born (2020) to diversify MIR: from expanding the field’s theories of “what music is” (Born (2010, p.232) as cited in Born (2020, p.199) to broadening its “ways of knowing” (Bowker (2018, p.207) as cited in Born (2020, p.200)).

2.2 Ethics, Axiology, and Technology

In this article, we propose that axiology, the philosophical study of value, be taken into account when practicing MIR: what are the core values of our research; what ought to be its impact on its stakeholders? The word “axiology” was introduced into philosophy by Urban (1909). While topics that are now the focus of axiology — including “meaning, characteristics, and classification of value, the nature of evaluation, and the character of value judgments” — have traditionally been attached to the study of ethics, they are increasingly accepted as a special branch of contemporary philosophy (Bunnin and Yu, 2004, p.65). The entwined relation between axiology and ethics is examined by Chang (2001), who defines axiology as a “philosophy of human life” whose purpose is to provide “fundamental codes of action” by inquiring into the “general nature, features and types of value” (p.71). Axiology thus extends beyond the limit of moral values that is the focus of traditional ethics. Or, more colloquially phrased in the context of music, besides ethical considerations our research is guided by, for instance, our musical taste and preferences.

Values, including but not limited to moral and aesthetic values, play a crucial role in the configuration of technology. This is explored by Gonzalez (2015, pp.3–28), who, on the basis that technology is value-laden, advocates the philosophical study of values about technology (or “axiology of technology”). Kroes and Meijers (2016) take a step further in calling for a two-stage “axiological turn” in the philosophy of technology as a continuation of the field’s earlier empirical turn (p.12). Drawing from Verbeek (2010), Kroes and Meijers (2016) argue that philosophers of technology need to participate directly in the development of technology via doing technology ethics, while at the same time moving beyond the scope of ethical values to include other kinds of values (pp.25–26).

Not unlike ethics, axiology has a cultural dimension, too. Examining Confucianism⁵ in nineteenth-century East Asia, Chen (2018) asks what “axiologies” predominated in this historically and culturally specific context. Via the Confucian idea of “benevolence and love” (renai 仁愛), Chen illustrates how this concept is not only a political one, but also carries with it “ethical and axiological import” (p.103). Scholars have also characterized the “New Confucian Movement”, a modern transformation of Confucian humanism, as a philosophical discourse that heavily “depends on axiological themes and traits” (Berthrong, 2008, p.423). The two case studies presented in this article demonstrate this dynamic interplay between value, ethics, and culture: in the former case, a number of philosophical traditions practiced in Asia and their associated values are considered for a re-interpretation of existing ethical principles for A/IS; in the latter, the value systems that help define ITM are taken seriously in order to conduct an auto-critique over AI’s intervention in the tradition.

2.3 A Cultural Turn?

An increase of attention in MIR toward aspects of the cultural contexts in which music takes place would be a strong indicator of a cultural turn. One way to collect evidence for such a cultural turn is to investigate how research presented at annual ISMIR conferences has engaged with the field of ethnomusicology, a discipline that studies music in (or as) culture (definition based on Rice, 2014, pp.1–10) and that is undergoing its own process of identity solidification (Amico, 2020). It should be noted, however, that while our following analysis focuses on the ISMIR conference, there have been initiatives beyond ISMIR that are relevant to this discussion. An example is the ERC-funded CompMusic project (2011–2017),⁶ which provides MIR datasets of and tools for Hindustani, Carnatic, Turkish-makam, Arab-Andalusian, and Beijing Opera traditions. In 2017, a workshop titled “Computational Ethnomusicology: Methodologies for a New Field” was held at the Lorentz Center, marking another step forward in reducing the semantic gap between MIR and ethnomusicology.⁷ On the other hand, ethnomusicology is only one of the many disciplines that MIR would increasingly engage with in a cultural turn, with other disciplines being — to name only a few — science and technology studies, anthropology, critical theory and cultural studies, philosophy of technology, music theory, or sociology. Hence, the following investigation presents evidence from a limited perspective focused on the intersections between MIR and ethnomusicology, without intending to assign an exclusive role to ethnomusicology in a cultural turn of MIR.

In the context of ISMIR, the relation between MIR and ethnomusicology seems to have been subject to discussion within the conference’s organizing committees. In 2011, the topic of “applications to non-Western music” was listed, to be replaced in the years 2012–2016 by the topic “computational ethnomusicology” (CE). In 2017, this topic disappeared from the list, and the calls since 2017 have listed computational musicology and computational music theory as topics related to the musicologies. In 2021, CE re-entered the list of ISMIR topics, and in 2022 seven ISMIR papers belonged to this category, an increased presence that may have been encouraged by the special calls for papers on cultural and social diversity in MIR in the years 2021 and 2022.

To obtain a better understanding of how discourses in MIR engage with ethnomusicology, the published body of papers of 16 years of the ISMIR conference was analyzed. After a short period of intense growth, the number of accepted papers at ISMIR between 2004 and 2018 remained largely constant, with about 110 papers in the mean (std: 11). Using the ISMIR paper explorer (Low et al., 2019), we searched all ISMIR papers in this “stable” period (2004–2018) for occurrences of the term “ethnomusicology”, and manually searched all papers from the 2019 conference (which is not contained in the explorer). We obtained a total of 78 papers that mention the term. Figure 1 depicts the percentage of the papers published in the proceedings that mention the term in each year. Despite fluctuations from year to year that would be speculative to interpret, there is a trend of increasing presence of the term “ethnomusicology” in ISMIR papers until 2009. After that, about 5% of published ISMIR papers each year mention the term, which illustrates a continuing awareness of the field in MIR that is consistent with numbers reported about a decade ago (Cornelis et al., 2010, p.1011).

Figure 1

Whereas a detailed analysis of how the 78 papers relate to the discipline of ethnomusicology is beyond the scope of this article, some qualitative insights may shed light on the nature of references to the term as it occurs in the papers. To this end, we selected a subset of papers from two periods: the first period contains the years 2008 and 2009, which marks the beginning of a continuous phase of heightened interest in ethnomusicology as depicted in Figure 1; the second period contains the years 2018 and 2019, in order to identify potential shifts in discourse over the range of 10 years. In the first period, 14 papers mentioned the term, whereas the second period contains 11 papers. A thematic analysis (Braun and Clarke, 2006) was conducted with these 25 papers by Holzapfel to investigate in which context the term “ethnomusicology” is used. In this analysis, all sentences including the term were extracted and assigned to non-predefined codes, which were in turn grouped into themes related to the type of interaction between the fields.

The four main themes that emerge were (1) computational ethnomusicology, (2) MIR informing ethnomusicology, (3) ethnomusicology informing MIR, and (4) synergies and collaborations. The first theme, computational ethnomusicology, is represented by occurrences in 4 papers (exclusively in the first period), and in these the term CE is used to situate the papers within this new area of research, which refers to the “design, development and usage of computer tools that have the potential to assist in ethnomusicological research” (Tzanetakis et al., 2007). The second thematic group is related to the first in this motivation, but does not employ the term CE. A total of 9 papers out of the total of 25 involve statements that emphasize the relevance of presented results for research in ethnomusicology. The third and largest thematic group is ethnomusicology informing MIR, with 17 papers using the term in such a context. Within this theme, three sub-themes are identified. The first (12 papers) uses references to ethnomusicological research to contextualize the research presented in the paper. The second sub-theme contains references to collections of recordings and field recording practice (6 papers). It should be pointed out that the emphasis here lies on field recording, the material outcome of field work, rather than the overall practice of ethnography. The third sub-theme finds expression in only 1 of the examined 25 papers, by employing ethnography to obtain insights into the listening behavior of a specific group of users (Cunningham and Nichols, 2009). The final theme, synergies and collaborations, finds expression in only 3 out of the 25 papers. In these, either the intention to collaborate with ethnomusicologists is mentioned (Wright et al., 2008), or ethnomusicologists are included as a user group in the study of the paper (Holzapfel and Benetos, 2019).

This analysis reveals a relationship between MIR and ethnomusicology that is far from the “agonistic interdisciplinarity” Born (2020) proposes. While MIR has incorporated outcomes of ethnomusicological studies when they could support or “serve” its discourse,⁸ synergies and collaborations are rare. An exception is the paper by Finkensiep et al. (2019), which includes an ethnomusicologist as co-author, and combines data-analytic approaches with ethnomusicological expertise to gain insights into the working of a specific musical idiom. More often, MIR scholars have focused on the “power” of computational methods to “contribute to” (ethno)musicology and to “upgrade” the humanistic discipline from a “data-poor field” into a “data-rich field” capable of providing “greater power for hypothesis testing” (Volk et al. (2011), citing Huron (1999)). Implied here is a hierarchical relationship between the physical and natural sciences and the qualitative humanities and social sciences, a hard-to-avoid problem that Cook (2005) addressed in his invited talk at ISMIR. It is also telling that in Gómez et al. (2013)’s introduction of a special issue of the Journal of New Music Research devoted to CE, the authors acknowledge that despite the issue’s coverage of “a broad area” of topics, “ethnomusicology core-problems” are “barely represented” (Gómez et al., 2013, p.111).

We thus argue that while MIR has made several significant attempts to diversify itself, to arrive at a more profound “cultural turn”, the field needs to open itself to non-hierarchical partnerships with humanistic and social scientific music scholarship, including but not limited to contemporary (ethno)musicology and music theory. Such dialogues could lead all involved disciplines toward critically rethinking the dialectics between scientific objectivity and humanistic subjectivity, between universality and particularity, between coherence and fragmentality, and between the modern and “the postmodern condition” in the study of (musical) culture (Lyotard, 1984). In light of the increased popularization of “digital humanities” (DH), it seems timely to ask: is the growth of CE and the reinvigoration of a “new comparative musicology” signaling a “groundbreaking, forward-looking, wonderfully experimental” moment as suggested by Tilley (2018, p.971), or is it a worrisome trend — considering ethnomusicology’s disciplinary origin as a turn away from comparative musicology and reductive quantitative approaches — that may risk taking the cultural study of music “backwards into the future” (Born, 2020, p.198)?

3.1 Non-Western Approaches to Technology

Discussions of technology have historically been driven by Western thought rooted in Plato and Aristotle. In an attempt to challenge the assumption of a universality of Western science and technology, Dusek (2006) in his overview of the philosophy of technology devotes two chapters to those who have historically been excluded from mainstream accounts of the nature of technology. This includes a discussion on Feminist philosophy of technology (Dusek, 2006, pp.136–155) and on contributions of non-Western knowledge systems to the ongoing scientific and technological development (Dusek, 2006, pp.156–175). While we are aware of the significance of these gendered, feminist, and intersectional interventions also exemplified in the works of Noble (2018) and D’Ignazio and Klein (2020), our focus for this article lands on the latter, that is, the value of considering non-Western philosophical perspectives when assessing responsible (music) AI.

In the context of AI ethics, a minority of authors have highlighted the importance of diversifying perspectives (Jobin et al., 2019; Hagerty and Rubinov, 2019; Fjeld et al., 2020; Clancy, 2021; Jecker and Nakazawa, 2022). Goffi (2021) argues against global AI governance that imposes an arbitrary hegemonic set of Western-based ethical norms upon the majority of Humanity, in the worst case as a guise to promote narrow commercial interests. The Montreal AI Ethics Institute, an organization working to democratize AI ethics literacy, has recently curated a directory of experts in non-Western AI ethics.¹⁰ Notably, Hui (2016), in exploring the possibilities of constructing a philosophy of technology that is “properly Chinese” (p.7), argues for the necessity of considering not one universal technology, but multiple “cosmotechnics” both philosophically and historically. The concept of “cosmotechnics”, which the author further develops in Hui (2021) and which informs this article, is defined as “the unification between the cosmic order and the moral order through technical activities” (Hui, 2016, pp.19–20).

There have been a number of studies that draw a direct connection between technology and specific Asian philosophical traditions. Scholars have, for one, turned to Confucianism. Writing on “ethical pluralism”, Ess (2006) juxtaposes contemporary Western ethics with Confucian thought to emphasize the importance of embracing both “pluralistic structures of connection” and “irreducible differences” between intellectual traditions (p.215). Kirk et al. (2020) explore how the Chinese government builds on select Confucian notions to inform the nation’s approach to technological governance, noting the “stickiness” of central Confucian values (such as hierarchy, family, and social order¹¹) in several East Asian societies despite divergence of political ideology. Advocating a “multicultural turn” in technology ethics, Wong and Wang (2021) develop what they call “Confucian ethics of technology”.

Buddhism also provides new opportunities for ethical (what is right?) and axiological (what is good and valuable?) reflections regarding technology. Rambelli (2018) explores the presence of machines in the Japanese Buddhist tradition, e.g., robotic monks and priests. In today’s Japan, one can see the Buddhist robot priest “Mindar” delivering sermons inside the 400-year-old Kodaiji temple (Nair, 2019). Hongladarom (2020) makes a significant attempt to bridge the ancient Buddhist tradition with technology (AI) ethics, as the author argues for a Buddhism-inspired standard of ethical perfection, namely “machine enlightenment” (p.7).

When studying the technology-friendly nature of Japanese society, scholars often turn to Shintoism as a source of understanding the nation’s anthropomorphic view of technology. In Shinto beliefs, there is no categorical distinction between humans, animals, and inanimate objects, as the religion attributes spirits, or kami 神 to all forms of existence. Juxtaposing this “Shinto-infused techno-animism” with actor-network theory,¹² Jensen and Blok (2013) posit that Shinto cosmic views offer a vantage point for interpreting the contributions of non-humans to “collective life”, and for studying the entanglements of politics, ecology, science, and cosmos in contemporary Japanese society (p.84).

While scholars have held that technologies are in essence antithetical to the concept of “self-so” (ziran 自然)¹³ promoted in Daoist thought¹⁴ (Allen, 2010), others (Needham, 1981) have uncovered the long history of Daoism engaging with technology. A study of Nelson (2014) reveals how Daoist ideas have influenced early twentieth-century German thinkers (Buber and Heidegger) and their views on technological rationality and modernity. This further motivates critical engagement with the ethics and axiology of technology that traverses cultural and philosophical boundaries.

3.2 Toward an Ethically Aligned Design for Music AI in Asia

We will now examine three ethical principles listed in EAD1e (IEEE, 2019), namely “human rights”, “well-being”, and “awareness of misuse”, and investigate how their meanings may shift (or not) when taken out of familiar, Western philosophical and cultural contexts, within and beyond music.

“Human Rights”, the first principle in EAD1e (IEEE, 2019), underlies most major guidelines for AI ethics. What the term signifies, however, can shift with contexts (An-Na’im, 2010). A fundamental question one ought to ask here is “what it means to be human”. The notion of personhood in Confucian thought, for instance, is characterized as inherently relational, developmental, and virtue-based (Alford, 2010; Wong, 2021), leaving space for the concept to be extended from human to non-human actors. According to AI ethicist Pak-Hang Wong, based on the “role-based” Confucian ethics, one can attribute personhood to non-human beings as long as they “play ethically relevant roles and duties as humans” (Cassauwers, 2019). Wong’s comment resonates with what we find from examining the Japanese Society for Artificial Intelligence (JSAI) Ethical Guidelines, which includes a clause stating that once AI abides by all policies described therein, it can then become a “member” or “quasi-member” of society (JSAI Ethics Committee, 2017, p.42). While we certainly cannot reduce this aspect of JSAI to a result of Confucian influences (despite the major role Confucianism plays in shaping Japan’s history), it is nevertheless an example of how fundamental concepts often assumed to be universal may take on new meanings when they travel.

With this greater flexibility in defining personhood come opportunities to think beyond the narrowly defined “human”, to move from humanism to posthumanism, and ultimately to expand this first ethical principle of “Human Rights” in EAD1e so that it cares for the rights of both human and non-human actors (including the environment). Writing on the political ecology of music, Devine (2019) writes that the carbon footprint of the music industry did not decrease in the age of streaming. Eco-musicological considerations — as we continue to operate under the framework of music “ecosystems” — thus become important with the advance of large, energy-consuming neural networks (Strubell et al., 2019) and, specifically, AI-generated music increasingly marked by “overproduction” and “uninspired excess” (Mersch, 2022, p.65). It is here that we may find value in certain perspectives of Laozi — the ancient Daoist thinker who promotes subtle “action” through “inaction” (wuwei 無為), who considers the “great note” (dayin 大音) as the one that “sounds faint” (xisheng 希聲), and who teaches that “only by relying on what is not there, do we have use of the room” (Van Norden and Ivanhoe, 2005). These notions can become sources of inspiration when considering “music of the environment” and the recovery of “positive silence” (Schafer, 2006), the reduction rather than (over-)production of sound, and the complementary forces between sounding (the Daoist yang 陽) and non-sounding (the Daoist yin 陰)¹⁵ in this era of algorithmic explosion.

Meanwhile, Mozi — after whom the philosophical school of Mohism is named — strongly condemns wasteful performances of music which, according to his utilitarian calculus, will interfere with the fair distribution of resources in society and with the maintenance of “good order” (zhi 治) (Van Norden and Ivanhoe, 2005). This Mohist stance is highly relevant when it comes to evaluating AI systems that can generate billions of tunes (see Section 4) and flood our already overloaded info- and sound-scape with algorithmically-generated musical “spam”. According to boomy.com, the users of Boomy — an AI-powered “instant” music generator — have generated around 13.32% of the world’s recorded music as of April 17, 2023.

While it is beyond the scope of this work to explicate the above Daoist and Mohist thoughts, we hope the point has been made that a quick mental exercise of philosophical re-orientation may open up new perspectives for thinking critically about responsible music AI. As attempts are being made by researchers using neural-network soundscapes to protect natural environments (Yirka, 2020; Learn, 2019), it is the responsibility of every actor within the AI music ecosystem to consider the impact of such tools over the health of our collective soundscape, and what human and posthuman rights may consist of in this changing context.

“Well-being”, the second principle in EAD1e (IEEE, 2019, p.4), asks that A/IS creators “adopt increased human well-being as a primary success criterion for development”. Different cultures and communities, once again, may have varying views over how technologies can best serve mankind and its well-being. For one, the tendency in Confucianism-based societies to blur the boundary between the self and the community, the individual and the collective, and the private and the public has a profound impact on how well-being is understood and how technologies relate to it. According to Hagerty and Rubinov (2019), AI-driven surveillance technologies do not generate as much controversy in Singapore and China because state surveillance is perceived as an “acceptable exchange for security and stability” (p.16). Hung (2021) also notes that due to a prioritization of societal “harmony” (hexie 和諧) over individual well-being, “collectively mediating technologies” implemented without full, individual consent are more readily accepted in Confucianism-based societies. This is reflected in the Beijing AI principles that explicitly raise “Harmony and Cooperation” as one of the central pillars of responsible AI governance (Beijing Academy of Artificial Intelligence, 2019).

Bringing in a Confucian perspective also enriches the ways in which one can conceptualize human-technology relations, and how such relations can contribute to human flourishing. Reflecting on a Confucian “ritual technicity”, Wang (2021) brings forth the ritual dimensions of artifacts that transcend their sheer practicality and examines how, in “performing” (rather than merely “using”) technologies, humans are able to moralize themselves with artifacts (p.21). Because of this intimate techno-human relationship implied in Confucian theories of self-cultivation, once embodied and ritualized, technologies become integral to human pursuits of growth, wellness, and harmony with the world.

While EAD1e does not explicitly address music AI, included is a subsection titled “Affective Computing” that discusses issues related to emotion-like control in both humans and AI systems. Considering the role of music as a culturally dependent regulator of emotions, one could propose that any AI music system be considered as a form of “affective computing” and follow the guidelines detailed in this subsection. To build an ethical application of AI to music that could foster well-being, it would be productive for researchers and developers to understand what makes a certain way of organizing sound culturally appropriate and aesthetically pleasing in a given music ecosystem. In Confucian ontology of music, for instance, not all “sounds” (sheng 聲) are considered melodious “tones” (yin 音), and not all organized “tones” (yin 音) carry the potential of contributing to human flourishing — a required condition for “music” (yue 樂) (Huang, 2019, pp.51–52).

Meanwhile, the tendency of many East Asian communities to align “cutting-edge technologies” (AI) with agendas of traditional culture preservation provides an insight on how one may ethically deploy music AI in these societies to maximize well-being. On the website of Aichi’s World Expo (cited in Šabanović, 2014, p.360), one can find claims indicating that “conservation should replace mass production and consumption”. Writing on Japan’s assembly of science, technology, and culture, Šabanović (2014) explores the ways in which Japan legitimizes its adoption of new technologies through strategic association with traditional practices and cultural continuity: it is recorded that robots are used to preserve aizu bandaisan, a Japanese folk dance, as there are no human inheritors to carry them out (p.350). Meanwhile, in Beijing, the director of conceptual theater show 2047 Apologue (andy-Robot, 2020) creatively fuses AI and robotics with Chinese folk music to shed light on larger themes such as culture preservation and environmental crises.

Technologies, in these cases, are perceived as culturally situated artifacts. Traditions, on the other hand, are continuously renegotiated and redefined to include emerging technological devices and practices. Moving forward, similar programs may be initiated with the design of AI systems that could work to revitalize traditional repertoire. When Huang spoke with Huo (2022), deputy director of the Divine Music Administration (DMA) of the Temple of Heaven in Beijing, where work is ongoing to recover court music of the Qing dynasty (1636–1912), Huo shared that DMA would much appreciate an AI artist-collaborator, as the institute’s current funding policy makes any new hire of human composers virtually impossible.

“Awareness of misuse”, the third principle in EAD1e that we will examine, emphasizes the need to minimize the risks of potential misuse of AI. Similar to how the concepts of “human rights” and “well-being” are often more fluid than one may assume, what constitutes misuse can similarly be context-specific. According to Hongladarom (2020), if Buddhist values and perspectives are taken into account, AI and its manufacturers must consider the interest of others before those of their own, carrying in mind the goal of relieving “all others” — as much as they possibly can — of suffering (pp.5–6). Thus, AI devices should strive to achieve both ethical and technical excellence, and any design that causes suffering would be considered a case of misuse. This Buddhist vision of AI ethics is in line with Floridi (2008)’s definition of information ethics (IE) as an ethics that seeks to free all being — defined here as “the existence and flourishing of all entities and their global environment” — of entropy, a state “more fundamental than suffering” (p.12). Besides Buddhist values, the Confucian notion of “harmony” (he 和) is also a productive concept when conceptualizing “misuse” and “harm”, as it implies the possibility of harmonizing diverging interests among a wide range of stakeholders without erasing their irreducible differences.

When developers make the effort to understand and respect indigenous value systems (the axiological dimension), potential misuses can be avoided. According to TikTok/ByteDance research scientist Lamtharn (Hanoi) Hantrakul (2021), when designing Tone Transfer — a web app that uses Google’s machine learning AI to realize timbre conversion between different sound sources — he was met with the challenge of having to balance the interests of developers, target users, and those of local music practitioners who might oppose having the unique timbre of their traditional musical instruments taken out of context. In an interview with Huang, Hantrakul (2021) described the team’s decision to not include guqin, the ancient Chinese zither, in Tone Transfer so as not to misrepresent the instrument in front of an audience with limited knowledge about its sound, history, and aesthetics. Hantrakul then contrasted Tone Transfer with his work on Sounds of India, an AI-powered app that transforms sounds into specific Indian music instruments. The developers of Sounds of India, Hantrakul shared, were more confident to engage classical Indian instruments in this case as they knew the app would be directed toward communities already familiar with their sounds and usage, thus reducing the risks of misrepresentation and cultural appropriation.