What’s Missing in AI Ethics Literature?

Uncategorized, , , ,

This essay was adapted from the script for a presentation of the same name delivered as part of the Coalition for Networked Information’s Project Briefing Series: Summer 2025.

Over the course of the last few years, I have read a great deal of literature discussing the ethics and latent values in the area of technology and more specifically, artificial intelligence. For the sake of brevity, I will not provide an exhaustive list of what I have read nor is what I have read exhaustive in regard to what’s available, but I will highlight some titles. The issue that I take with most AI ethics literature is that they either discuss AI ethics without ever mentioning frameworks from moral philosophy, or they discuss AI ethics using only one ethical framework. This presentation is a brief discussion of an extensive literature review and concludes with an explanation for where I feel AI ethics discourse should be directed.

Vallor’s (2016) Technology and the Virtues specifies virtue ethics, stemming from Aristotelian traditions, and she does include some discussion of non-Western virtue ethics. Likewise, her book The AI Mirror (2024) also follows a tradition of virtue ethics. Gertz’s (2018) Nihilism and Technology entirely focuses on the concept of nihilism; although this framework comes from existentialism, not moral philosophy, I thought it interesting to include because some scholars consider Friedrich Nietzsche to be a virtue ethicist, and his ubermensch model is portrayed as someone who overcomes universal determinism through self determination by redefining morality. Even the Vatican has offered its views on AI ethics in partnership with Santa Clara University in their book Ethics in the Age of Disruptive Technologies: An Operational Roadmap (2023).

Other books I have read, such as Hepler et al.’s (2024) AI Ethics in Higher Education and Smuha’s (2025) The Cambridge Handbook of the Law, Ethics, and Policy of Artificial Intelligence, say nothing of ethical frameworks or if they do, it is only in passing or merely implied. Typically, the reader is left with a series of chapters that proclaim best practices and unpack problematic characteristics inherent to AI, such as bias, mis/disinformation, stereotyping, harmful language, and infringements on privacy and intellectual property. These sorts of conversations are generally held with a utilitarian framework in mind; however, utilitarianism or any other form of consequentialism is never formally discussed. It is up to the reader to know enough about ethics before entering the conversation. The only book that I have come across that discusses the application of ethics in all its forms to information and technology is Burgess and Knox’s (2019) Foundations of Information Ethics.

I am beginning to notice that this is a pattern in AI ethics literature, and others, it seems, have noticed as well. During a panel presentation at the STRAKER AI ethics workshop held at The University of Alabama on May 10, 2025, one presenter, Dr. Jiaqi Gong from the College of Engineering, mentioned that while worthwhile, AI ethics, as it is now, is unhelpful. Luo (2025, p. 392) reports on a survey of librarians using AI in their workflows and says, “‘The hand-wringing about ethics is warranted but not useful.’ [Generative AI] has become the reality, and a shift in attitude toward AI is unavoidable.” 

Such sensibilities have driven many researchers, ethicists, and consultants toward the notion that it no longer serves us to discuss ethical frameworks, instead grounding discussions of ethics into practice. This, however, reveals problems as well. As opposed to having a discussion about AI ethics, authors offer what I would describe as a code of conduct to be exercised when interacting with AI products and services. They provide a top-down paternalistic essay with every chapter that teaches not ethics but responsibility, mindfulness, and professionalism. These issues, of course, have their time, place, and use in multiple fields integrating AI into their work, not to mention the wisdom they hold for society in general.

The problem, though, with a paternalistic approach is two fold. First, it implies that users can successfully superimpose their own values over the values of the developers who created these AI systems. Second, it overlooks the multiple layers of competing interests that create what ethicists call moral dilemmas and what political scientists call wicked problems (Head & Alford, 2015; McConnell, 2022).

In the realm of AI research, the question of authorship in regard to expressions of ideas is often debated. Is AI responsible for the text, images, audio, and/or video it generates? I use this as an opportunity to relate it to another form of technology on which society seems similarly divided. Do guns kill people, or do people kill people? Talk to any firearm owner, and they will tell you all the reasons why they own a firearm. It is used for self defense, it is used for recreational activity at the range, or maybe it is used for hunting, which for some is recreational and for others is a way to secure food for themselves and their families and friends. These are values that owners impose on their firearms; however, the firearm’s purpose, the teleological value that it is imbued with when made, is to end life as efficiently and effectively as possible. 

A discussion of how we should all use AI, as opposed to a conversation of true AI ethics, suggests that each individual user can compensate for the way these systems are made and function by simply behaving in a responsible manner. AI products and systems, which advocates innocuously call “tools,” were designed with specific purposes in mind. DuPont, for example, developed Teflon for non-stick pans, among many other products, making it easier for cooking and cleaning. Despite the public knowing how to use non-stick pans appropriately, DuPont always knew more than anyone else, including the EPA, about the harms of “forever chemicals,” which have perforated and poisoned American society. While the public may be aware of how to use AI systems thanks to the way companies market them, we will never overcome the fact that AI developers, such as Amazon, Google, Meta, and Microsoft will always know more about these products than we do. As such, these values are hidden from the public, and the only values revealed are the ones that encourage the public to engage with their products.

Furthermore, from a policy perspective, various parties with competing interests want or need AI to operate in different ways, and most of these conversations are in regard to the data used to train AI systems. This sets up a moral dilemma, a situation in which two priorities come into conflict with each other when trying to concurrently observe both. This is most prevalent in the areas of intellectual property and privacy, and how these issues are juxtaposed with diversity in training datasets to account for bias, stereotyping, harmful content, and mis/disinformation. 

For instance, outspoken AI critics want technology developers to respect their privacy. In fact, Amazon, Google, Meta, and Microsoft have been called out for the ways in which they leverage their products to gather consumers’ voice imprints (Henderson, 2023; Taylor, 2025; Zilber, 2024). These voice imprints are most likely used for AI training. The purpose is so that voice interactive AI programs, such as Amazon’s and Google’s AI voice assistants, are programmed to successfully process a wide spectrum of voices and speech patterns, including different kinds of regional accents and speech impediments. A discussion solely grounded in practice, which substitutes responsibility for ethics, assumes that any actions can be justified for the greater good, and indeed, some ethical frameworks can be used to justify this behavior; however, we overlook nuance and meaning by not discussing this scenario within the contexts of other ethical frameworks such as virtue ethics, deontology, social contract theory, and care ethics as well as non-Western frameworks like Ubuntu, Confucianism, and Karmic ethics . 

For example, employing concepts found in deontology and care ethics would show that gathering consumer voice imprints without their knowledge and consent is impermissible, but with utilitarianism, there is a path toward justifying this behavior. Only after assessing this kind of dataveillance, AI training, and product development using multiple ethical frameworks can we decide which is the higher moral priority. Is it the public’s right to privacy, or is it the public’s right to an AI voice assistant that understands everyone? This is an important process because ethics intends to maximize benefits and minimize harms.

Similarly, the use of copyrighted content for training data has paved the way for the US to take the lead in the global AI race (Sag & Yu, 2025). AI developers have historically claimed fair use, which according to Section 107 of the US Copyright Act, permits the unauthorized use of copyrighted content under specific circumstances, which has so far been upheld in court; see Bartz v. Anthropic and Kadrey v. Meta, which were ruled on this summer in the Northern District of California. Again, there is a moral dilemma facing many courts across the US. Copyright law is heavily influenced by utilitarianism in the sense that copyright laws and court rulings are made under the assumption that these decisions will lead towards maximizing the greatest good or minimizing the greatest harms for the whole of US citizens and government. Which will judges decide is the greater moral imperative? Will it be the rights of copyright holders to determine how their works are used (e.g., duplicated, distributed, etc.), or will it be the rights of AI developers to continue to use copyrighted works for AI training data so that the US may continue to lead the world in AI?

Given that this example has introduced global concerns, this is an excellent opportunity to introduce the concept of wicked problems. A wicked problem is a complex issue with no clear definition or solution, primarily due to layers of uncertainty, shifting values, and competing interests, and solutions to a wicked problem often change the problem by leading to unforeseen consequences. For those familiar with philosophies of modernity and technology, this is very similar to Jacques Ellul’s concept of technique, whereby technical solutions to humanistic problems result in self-defeating and dehumanizing outcomes. 

For example, in the 1960’s, Ghana undertook the Volta River Project. It was thought that damming the Volta River would address perceived poverty in Ghana by encouraging Westernized industrialization, thereby providing a higher standard of living for the public. Despite all intentions, only the urban elite reaped these benefits while nearly eighty-thousand residents were displaced. The project caused a string of public health crises, and failed attempts to address them led to contaminating the drinking water. In the end, employing a cascade of technical solutions resulted in additional problems because the government ignored the local ecology and history. Ghana’s government ended up relinquishing control of the project, turning it over to Western specialists and investors, debasing Ghana’s historical struggle to liberate itself from Western influence (Agbemabiese & Byrne, 2005).

The wicked problem of AI on a global scale is that the US is competing with nations that do not necessarily reciprocate the US’ ideals of human rights and democratic values. AI is an extraordinarily powerful technology that can outcompete human counterparts in work productivity, efficiently persuade and even radicalize people, monitor and analyze human activity at scale, and produce significant novel results in various areas of science and engineering due to its pattern recognition capabilities. It could be very dangerous if a nation that does not prioritize human rights and democratic values were to dominate the AI space. There are, of course, other competing interests relevant to allied and adversarial nations as well as international organizations such as the UN and NATO. However, for the US to continue leading in AI development at its current pace – note that China is not far behind – it seems that it must deprioritize the rights of citizens and residents to protect their privacy and intellectual property. This conclusion has been reached only by way of utilitarianism because that is the framework that US policy is built upon. What answers, then, would we come up with if other frameworks were leveraged? Of those alternatives, which would be viewed as the moral course of action?

This is why AI ethics must be more than just grounded discussion in practice. If we are always grounding discussion in practice, then conversation will always turn into top-down paternalistic decision guidance on responsibility, and that sense of responsibility will inevitably cater to instrumentalism and pragmatic utilitarianism, which ignore competing values and may lead to undesirable outcomes. We need more literature that discusses how to develop, assess, adopt, deploy, and use AI products and systems, and each of these matters can be discussed separately; however, each issue must be analyzed through lenses found in multiple ethical frameworks to ensure that a Western bias toward consequentialism and pragmatism does not shut out other moral concepts. If the US is to lead the world in AI development, then the US is likewise responsible for leading the world in AI ethics. As such, multiple ethical frameworks simultaneously employed to probe the various dimensions of AI are required for the philosophically and politically diverse global population.

Given this diversity, one ethical framework that has emerged and should be further investigated is pluralistic ethics. Summarizing Charles Ess (2006), pluralistic ethics aims to conjoin competing and antithetical value systems among diverse societies and cultures by identifying overlapping values and judgments held by those communities. For example, many cultures have their own versions of the “Golden Rule,” which states, “Do unto others as you would have others do unto you,” and the other versions of this principle state something of similar meaning and effect (Roberts & Black, 2021). In accordance with pluralistic ethics, this means that all societies that adhere to a golden rule would then agree on this issue and find common ground. That, however, is not necessarily the case because pluralistic ethics appears to stop short of considering the value of reciprocity. Just because two opposing societies each adhere to a golden rule does not mean that these societies believe that the terms of their respective golden rules extend to those outside their own societies. 

Pluralistic ethics has been useful in the global governance of other technologically relevant issues such as the Internet and data (Wiesmuller, 2023). What we need now, though, is something that goes a step further. We need a framework of reciprocal ethics. A framework of reciprocal ethics would identify not only superficial commonalities in values but also substantive commonalities in judgments. 

AI developers, consumers, advocates, and regulators all retain competing interests, but wanting useful AI is something they all have in common. The term useful, however, is value-laden, and so, it must be determined what each party means by useful. Within each party’s definition of the term useful, there may be additional value-laden terms, and those terms must be further defined until definitions are made clear in their entirety. All parties involved may need to leverage wisdom from multiple ethical frameworks. Once value-laden terms, like useful, helpful, good, and intuitive are explained to the point where all terminology shares the same meaning to all parties, then all parties can decide on how to proceed with development, assessment, adoption, deployment, and use.

I don’t know exactly what that looks like from an administrative perspective; however, we do have reciprocity models in the US to consider such as concealed weapons permits and credentials for teachers, electricians, realtors, architects, and insurance claim adjusters. Recognition of these various certifications differ from state to state. Driver’s licenses, on the other hand, despite being issued by individual states, are reciprocal in every US state and territory and even in some foreign countries. We can, therefore, ask, “What criteria of values does each governing body prioritize before recognizing another governing body’s issued credential?” From further understanding systems of reciprocity, we can begin to build onto pluralistic ethics by establishing an ethics of reciprocity.

References

Agbemabiese, L., & Byrne, J. (2005). Commodification of Ghana’s Volta River: An Example of Ellul’s Autonomy of Technique. Bulletin of Science, Technology & Society, 25 (1), 17-25. https://doi.org/10.1177/0270467604273821. 

Burgess, J. T. F., & Knox, E. J. M. (2019). Foundations of Information Ethics. American Library Association.

Ess, C. (2006). Ethical Pluralism and Global Information Ethics. Ethics and Information Technology, 8, 215-226. https://doi.org/10.1007/s10676-006-9113-3. 

Flahaux, J. R., Green, B. P., & Skeet, A. G. (2023). Ethics in the Age of Disruptive Technologies: An Operational Roadmap. Markkula Center for Applied Ethics.

Gertz, N. (2018). Nihilism and Technology. Rowman and Littlefield International.

Head, B. W., & Alford, J. (2013). Wicked Problems: Implications for Public Policy and Management. Administration & Society, 47 (6), 711-739. https://doi.org/10.1177/0095399713481601. 

Henderson, J. G. (2023). FTC and DOJ Charge Amazon with Violating Children’s Privacy Law by Keeping Kids’ Alexa Voice Recordings Forever and Undermining Parents’ Deletion Requests. Federal Trade Commission. Retrieved July 10, 2025, from https://www.ftc.gov/news-events/news/press-releases/2023/05/ftc-doj-charge-amazon-violating-childrens-privacy-law-keeping-kids-alexa-voice-recordings-forever. 

Luo, L. (2025). Use of Generative AI in Aiding Daily Professional Tasks: A Survey of Librarians’ Experiences. Library Trends, 73 (3), 381-398. https://doi.org/10.1353/lib.2025.a961200. 

McConnell, T. (2022). Moral Dilemmas. Stanford Encyclopedia of Philosophy. Retrieved July 10, 2025, from https://plato.stanford.edu/entries/moral-dilemmas.

Roberts, C., & Black, J. (2022). Doing Ethics in Media: Theories and Practical Applications. Routledge.

Sag, M., & Yu, P. K. (2025). The Globalization of Copyright Exceptions for AI Training. Emory Law Journal, 74, 1-47. https://doi.org/10.2139/ssrn.4976393. 

Smuha, N. A. (2025). The Cambridge Handbook of the Law, Ethics, and Policy of Artificial Intelligence. Cambridge University Press.

Taylor, J. (2025). NSW Education Department Caught Unaware after Microsoft Teams began Collecting Students’ Biometric Data. The Guardian. Retrieved May 22, 2025, from https://www.theguardian.com/australia-news/2025/may/19/nsw-education-department-caught-unaware-after-microsoft-teams-began-collecting-students-biometric-data. 

Vallor, S. (2024). The AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking. Oxford University Press.

Vallor, S. (2016). Technology and the Virtues: A Philosophical Guide to a Future Worth Wanting. Oxford University Press.

Wiesmuller, S. (2023). The Relational Governance of Artificial Intelligence: Forms and Interactions. Springer.


Zilber, A. (2024). Marketing Firm Admits Using Your Own Phone to Listen in on Your Conversations. New York Post. Retrieved July 10, 2025, from https://nypost.com/2024/09/03/business/marketing-firm-spies-on-you-through-your-phones-microphone-report/.

UX and Web Technologies for the Music Librarian

Uncategorized, , , , , , ,

At the 2024 Music Library Association (MLA) Conference, three-quarters of the MLA Web Team (Web Managers Karen Berry McCool and Kerry Carwile Masteller and Web Committee Co-Chair Laura Jacyna) gave a lightning talk presentation about various web platforms we’re using as we revamp the MLA website.

MLA’s web presence will be migrated in the near future from being hosted mostly on YourMembership (YM) to mostly or entirely on WordPress (WP). This is not an easy process, as it will include not just moving information, but also redesigning. The process will be very time-consuming, and requires a lot of decision-making that we need to do based on high-quality information. We’ll be using various tools as we make the transition; of course, YM and WP are two of these tools, but we’re going to focus on the newer-to-us, more supportive technologies.

First, there’s Figma, which calls itself “The Collaborative Interface Design Tool.” The possibilities with Figma are really endless; it can be used like Canva to design literally all kinds of things, but with even more granular options. It can be used like PhotoShop, but in a cloud-based, collaborative platform. There’s a whiteboard feature, where you can work with a team to draw, type, or create Post-it-style notes. According to their website, you can use it for anything from brainstorming or mind-mapping to strategic planning, concept mapping, or diagramming. So far, the Web Team has used it to do some wire-framing; in other words, we’ve recreated the menu structure of the website visually in Figma so that we can move things around, make comments and changes, and lay everything out so we can envision how to move forward. The screenshot below demonstrates how designers can collaborate when prototyping a design.

Figma's web interface.

Next is Jamboard (see screenshot below this paragraph), which many of you may have used before. We’ll be brief with this one, since we recently learned that Google is sunsetting it at the end of the year… It’s basically the whiteboard feature from Figma, where you can collaboratively arrange Post-its, text, and pictures on big rectangles. We’re using it as one option for card sorting (more on that later).

A Google Jamboard canvas, set up for a card sort activity based on a library website. A row of blue square notes identify categories, and a pile of yellow square notes show menu items to be sorted.

Padlet (see screenshot below) is pretty similar, although you can set it up in different ways so that it’s more like a whiteboard or a row of columns with cards underneath. Each item added to Padlet can be commented on by people with or without an account — Using someone’s Padlet without an account makes for anonymous comments. They seem to be limiting what you can do with a free account, but for the moment it can still be useful for collaborating and arranging ideas for free.

A Padlet canvas, set up for a card sort activity based on a library website. A row of tabs identifies categories, and a column of cards shows menu items to be sorted.

MLA is in the process of activating a paid subscription to UXtweak, where, like Figma, the possibilities are almost endless (though for user experience, or UX, testing, not necessarily design). There are options for preference tests, mobile testing, tree testing, surveys, and prototype testing, among lots of other things, in the free version alone. You’re limited to how many of each of the tests you can perform with the free version, which is why MLA is going to pay for a subscription for a limited time. I will once again mention card sorting, a research methodology that the three of us and Web Committee Co-Chair Woody Colahan presented on at the conference. This screenshot depicts a card sort in UXtweak in action:

A UXTweak online card sort session in progress, showing a user test to organize a list of fruits into user-created categories.

Finally, we’ve utilized something called XML-Sitemaps. This tool is what we’re going to utilize before we start user testing in earnest. According to their website, two of the other purposes of the site is to create something that will allow search engines to provide better results about your website, and to allow website visitors to navigate more easily. The MLA Web Team plans to use the tool to compile a list of all the pages on our website, displaying all of the pages that are publicly available at the time of the scan. Believe it or not, this is not something that YM is able to provide for us, despite being the service that hosts our website. By creating this list, we’ll be able to perform a full content audit, where we’ll figure out what we have, what’s redundant, what’s outdated, what can be deleted, and what we should definitely move forward onto the next version of the website. The screenshots below show two different ways that XML-Sitemaps can display your website information:

A spreadsheet of website URLs.
A hyperlinked hierarchical list of webpage titles.

There are some other tools that the Web Team is considering using, as well as others that we simply didn’t have time to present on. UX testing and website migration and redesign are complex tasks, so don’t go it alone! And the beauty of these tools is how versatile they are; these tools can be used for various research methodologies, (library) instruction, and even just for fun. Best of luck on all your projects!

The Future of Music Libraries: AI Integration and the Evolving Role of Digital Platforms

Uncategorized

by Nurhak Tuncer-Bayramli

Introduction

As a music librarian at Elizabeth City State University, I find myself at the forefront of a rapidly evolving landscape where digital music scores and AI are transforming the way we approach music curation and distribution. My roles as a digital repository coordinator and music cataloger, coupled with my doctoral studies in Educational Leadership and Management with a focus on instructional technology, have given me a unique perspective on these changes. However, it’s crucial to recognize this evolving landscape; while they offer significant benefits, they pose challenges for traditional collection development and should supplement, not replace, our existing collections.

Advocacy for Digital Music Score Platforms

Digital music platforms like Nkoda and MuseScore are reshaping access to music scores and compositions. Nkoda offers a vast repository of scores, valuable for accessing a wide variety of works, including those that are less commonly available. Libraries and academic departments increasingly provide subscription access to platforms like Nkoda, easing the financial burden on students and educators.

MuseScore stands out with its social networking capabilities, allowing users to engage through features like liking, commenting, and sharing scores. This interactive aspect builds a community among composers and musicians, promoting collaboration and feedback. MuseScore’s role as a communal and creative platform makes it a unique resource for composers and students looking to expand their musical network and exposure. MuseScore offers a more interactive experience, allowing composers to create and share music freely. David MacDonald, a self-publishing composer, provided an insightful review of the Nkoda music subscription service. He highlights Nkoda’s extensive library as its most significant asset, offering an array of scores and parts from a broad spectrum of composers and publishers. However, he notes some limitations, such as the need for more comprehensive coverage and issues with image quality in some scores. MacDonald also mentions the app’s functionality for uploading personal PDF files but points out the limitations in their usage and integration within the app​​. (his post from “scoring notes” in 2018. https://www.scoringnotes.com/reviews/nkoda-review/#)

Compared to traditional composition software like Finale and Sibelius, MuseScore offers a distinct experience. While Finale and Sibelius are renowned for their comprehensive professional tools and capabilities, ideal for intricate and detailed score writing, MuseScore emphasizes user accessibility and community engagement. Unlike the more professional orientation of Finale and Sibelius, MuseScore’s social network features cater to a broader range of users, from hobbyists to budding composers. This makes MuseScore not just a tool for score creation but also a platform for sharing, learning, and connecting with other music enthusiasts.

These platforms democratize music distribution, making it more accessible to a broader audience, which is a significant step forward for music education and appreciation. For self-publishing composers, these platforms offer unprecedented visibility and access to markets that were previously unreachable. They can now share their works with a global audience, receive immediate feedback, and collaborate with other musicians and composers worldwide. 

While platforms like MuseScore and Nkoda significantly enhance access to music scores and foster community engagement, they still can not replace the actual archival function of digital repositories in libraries. Library repositories offer more secure and permanent archiving capabilities, ensuring the long-term preservation of works. This is particularly important considering the potential risks associated with composer accounts on commercial platforms — for instance if a self-publishing composer loses access to their platform account or decides to delete it. In such scenarios, the permanence and stability offered by library repositories become invaluable, maintaining access to compositions regardless of changes in the digital platform’s status or individual user accounts.

The Dark Side: Challenges in Collection Development

While digital platforms are a boon to accessibility and distribution, they present significant challenges in collection development for music libraries. One of the primary concerns is the over-reliance on these platforms, potentially leading to the neglect of physical collections. Physical scores and manuscripts have an intrinsic value and historical significance that digital copies cannot replicate. The tactile experience of handling a physical score, the nuances in annotations, and the historical context they carry are irreplaceable.

Moreover, digital platforms often operate on subscription models, which raises concerns about long-term access and ownership. Libraries may have to continually invest in subscriptions to ensure ongoing access, unlike physical collections, where a one-time purchase guarantees perpetual access. This dependence on external platforms for access to scores can be a precarious position for libraries, especially in the face of budget cuts or changes in the platform’s policies.

Supplementing, Not Replacing, Traditional Collections

In today’s music libraries, the integration of digital music platforms is becoming increasingly prevalent, reflecting a significant shift in access to music scores. While the enduring preference for traditional print music among many musicians and educators remains due to its tangible presence, historical richness, and the unique experience it offers, the use of digital scores is on the rise, more so now than at any previous time. This trend towards digitalization should be viewed as a complement to, rather than a replacement of, the existing traditional collections in music libraries. Such an integrated approach facilitates a comprehensive and diverse collection, harmonizing the classic allure and physicality of print scores with the modern convenience and extensive range that digital platforms provide. This balanced strategy caters to a wide array of preferences and needs, ensuring that music libraries continue to serve as versatile and dynamic resources in the evolving landscape of music education and performance.

The Role of AI in Music Libraries and Education

Integrating AI in music libraries and education is a rapidly evolving field with immense potential. In music education, AI-driven tools can provide personalized learning experiences and adaptive feedback and even assist in composing and arranging music. For self-publishing composers, AI can offer insights into music trends and audience preferences and enhance their compositions with sophisticated algorithms. However, this integration has its challenges. AI algorithms, while efficient, may need a more nuanced understanding of cultural contexts and the subjective elements inherent in music. There’s also a risk of AI perpetuating biases present in its training data, potentially leading to a narrow representation of music styles and cultures. Libraries must, therefore, be cautious, ensuring that AI tools are used ethically and in ways that enrich rather than diminish the diversity of their collections.

While not fully realized yet, the prospective integration of AI in music libraries and educational platforms like Nkoda and MuseScore holds significant potential for the future. This development could revolutionize these platforms by introducing advanced personalization and analytical capabilities. For music libraries, this means transitioning towards more AI-driven tools that could offer enhanced user experiences, from personalized score recommendations to insights into evolving music trends.

In the future, we can envision a scenario where AI assists in understanding and predicting broader musical patterns and preferences. This evolution could enable music libraries to become more adaptive and responsive to the needs of their users, offering a more tailored and insightful experience. However, this forward-looking integration must be approached with a careful understanding of its implications, ensuring that AI’s use in music libraries and platforms like Nkoda and MuseScore enriches the diversity and cultural richness of the collections they offer.

Conclusion

As we navigate the integration of digital platforms and AI in music libraries, embracing these technologies is essential while preserving the essence of traditional collections. By striking a balance with human interaction, we ensure our libraries remain dynamic resources that honor the past while embracing the future. In doing so, we continue to support the evolving needs of composers, educators, students, and music enthusiasts in an increasingly digital world.

Creating and Editing Closed Captions

Uncategorized

Closed captions are an important step in making video (and in some cases audio) materials more accessible to larger groups of people.  In addition to being a pivotal part of the video-watching experience for viewers who are Deaf or hard of hearing, closed captions are also a manner of universal design – Even viewers who can hear benefit from captions when they are watching in a setting where they can’t use sound, or when the audio is indistinguishable.  In this day and age, there are many options available for closed caption creation and editing, both manually and automatically with platforms that utilize artificial intelligence (AI).  This post will delve into some options that I’ve personally explored, organized here from least expensive to most expensive.

Before diving in, it should be mentioned that closed caption files typically come in one of two file formats: SRT (SubRip subTitle or .srt) and VTT (Video Text Tracks or .vtt).  SRT is essentially just basic text with time markers, while VTT does allow for personalization and metadata.  When creating captions, it can be very helpful to have a transcript of what was said if it’s available, but in lieu of that, listening to the video or using an auto-caption platform (like those below) can be even more efficient.

Notepad

At this point, most operating systems have some sort of a native note-taking application available for those who need to create low-tech captions without much of a budget (Notepad on PC, TextEdit on Mac, Google Docs on Android, etc.).  Both SRT and VTT files can be edited in this way, as well as basic TXT/.txt files.  Writing captions from scratch on these applications isn’t the most straightforward thing, so if possible it may be most useful to download automatic captions from one of the below options and then open them with a Notepad app.  This will provide you with a template that includes timing for the captions in the form of timestamp ranges, which is crucial for captions to display correctly with a video.

YouTube

                YouTube, available for free and as a native app for Android and other operating systems, provides automatic captioning for newer videos, and allows you to create captions from scratch for all videos.  Once these captions are created, they can be downloaded as SRT or VTT files, or SBV, which is YouTube’s native caption format.  Caption files in these formats can also be uploaded to YouTube.  The user interface is quite user-friendly, and includes automatic syncing features, a “Pause while typing” feature, and keyboard shortcuts.  Additionally, YouTube allows video titles, captions, and metadata to be translated into foreign languages.  While the title and metadata must be translated by hand, the captions can be automatically translated by Google Translate.

Vimeo

            Vimeo is available for free, with additional features available with subscriptions at varying tiers.  Like YouTube, Vimeo offers automatic captioning, as well as caption file upload and download.  Only VTT is supported, but Vimeo does cooperate with Rev to allow uploaders to pay $1.50 per minute of content for accurate captions in English or other languages.

Amara

            Amara is a crowd-sourced caption editing site that allows caption editing through Vimeo and YouTube, as well as with video uploads in MP4, WebM, OGG, and MP3 formats.  Upon uploading (or linking via URL to Vimeo or YouTube), videos are added to “Amara Public,” which is a “workspace…designed for collective creation and use for public videos by all Amara users.”  In other words, once a video is in Amara, it’s theoretically available for others to help caption, though I have never personally had anyone assist with my caption editing.  Alternatively, $12 or more per month per user can be spent for a private Amara workspace that is not publicly accessible.  Once a video is uploaded/linked, a caption file can be uploaded or created from scratch, so no auto-caption options are natively available.  For this reason, Amara is really best for caption editing/accuracy rather than creation, but its user interface is very robust and flexible.  It’s an especially great platform for creating foreign language captions, and works with SBV, SRT, TXT, and VTT file formats, as well as DFXP and SSA.  Once captions are completed in Amara, they must be downloaded and then uploaded back to the original platform (i.e. Vimeo or YouTube), as needed.  The Music Library Association (MLA) has used Amara for conference session recording caption editing in the past, though the MLA Web Team eventually decided that more volunteer hours were needed than we had available, so we transitioned away toward 3Play (below).

Camtasia

            TechSmith’s Camtasia is a video editing software suite that has a caption editing/adding feature in addition to countless others for video creation.  The 2023 edition costs $299.99 for a perpetual license (with discounts for educational and governmental organizations), so it is not a worthwhile investment purely for caption editing, but for beginning-to-end video creation and editing it’s quite a bargain.  It has had an auto-caption feature for years, which is honestly one of the most accurate I’ve seen (at least in terms of offline, AI caption software).  Camtasia imports and exports captions in SAMI or SRT files.

3Play

            For those with a larger volume of videos (as well as the funds and desire to have very accurate captions), 3Play is one of the very best captioning resources on the market.  In addition to closed captions, they provide live captioning, audio description, subtitling, and translation services.  Videos can be shared with 3Play from over twenty video-sharing platforms including YouTube, Facebook, Vimeo, and Panopto, and costs depend on how quickly accurate captions are requested (starting at $2.95 per minute of content for express captions in English, or even cheaper for captions requested 10 or more days out).  3Play prefers to start from scratch, so uploading captions for editing is neither necessary nor possible.  For increased accuracy, 3Play requests that any unusual words, names, or acronyms mentioned in the video are provided by the customer, and words that the [human] captioner was unsure about after review are flagged for the customer to correct.  While there is a bit of a learning curve with their user interface, once it’s learned the process is incredibly efficient, and even at the cheapest/slowest rate, 3Play often delivers early.  Their customer support team takes great care in what they do, and they come highly recommended by other experts in the field of professional video editing.

Conclusion

While investing in accurate closed captions can be time consuming and possibly expensive, it’s a very important step that should be taken to make your videos available to a larger audience.  In this post, I only discussed caption creation/editing post-recording/-event; live captioning is an entirely different beast, but one that MLA hopes to explore in greater detail soon.  I also did not address transcripts for audio-only recordings, though that tends to be more straightforward since a special file format isn’t usually necessary for timing’s sake.  In general, captions are easiest to edit when they’re initially created using software because of the timing element, but the very best captions still need a human touch for the highest accuracy.  Happy captioning!

Gamified Instruction with Twine

Uncategorized

Twine is a text-based, interactive fiction platform created in 2009 by Baltimore-based writer, game designer, and web developer Chris Klimas. Twine runs on Windows, Mac OS, or Linux, and is also available as a web app. It can be downloaded from twinery.org.

Interactive fiction means you, the reader, make choices about the direction of the narrative. Twine is a free, open-source program that allows helps you author a story whose twists and turns are determined by your reader choosing which links to click on in every panel of text. Easy to use, Twine requires no experience with coding. The Twine Cookbook shows you first how to create a simple, basic story, then how to add features and complexity. Twine is designed to work with text, but it is easy to add graphics audio and to customize the look and feel with CSS.

Authoring a narrative in Twine, you will see a graphic display that shows text passages as discrete elements and maps the relationships between them based on the links they contain. When you are ready, you can export your story as a small, stand-alone HTML file that can be opened in a web browser.

Part of the structure of the project Marion Sparkle and the Mansion of Fire, opened in Twine.
(source: University of York)

The splash screen of Marion Sparkle and the Mansion of Fire.
(source: University of York)

Game designers embraced Twine right from the beginning. The same capabilities that make Twine ideal for authoring interactive fiction also make it a powerful tool for creating text-based games. The simplicity of using Twine makes it especially appealing to individual game authors, the kind who create prosocial games for reasons other than commercial ones. The most influential is one you have probably never heard of, Depression Quest. This is a journey through the life of a person with clinical depression. Not very fun, really, but it made its own kind of history when it touched off the “Gamergate” harassment campaign against women in the video game industry in 2014. (Was that really less than ten years ago?) Another critically acclaimed title developed with Twine is Howling Dogs, a meditation on trauma and escape.

Thousands of interactive fiction works have been created with Twine; You can browse many of them on the Interactive Fiction Database and on itch.io.

The simplicity of use that makes Twine easy for an individual fiction or game author to work with – no need for a team of developers! – makes it a great platform to develop a gamified instruction activity. Because Twine games are published to HTML files, they can be distributed to students directly or published to interactive fiction repositories like IFDB. This flexibility makes them especially useful for asynchronous instruction. Here is an example of a gamified library instruction activity that guides players through a tour of library collections, subscriptions, services, and policies at the University of Denver: Aliens in the AAC.

Aliens in the AAC, a game that guides players through a tour of University of Denver’s
collections, subscriptions, services, and policies.

Gamified instruction allows you to introduce concepts in a fun context that bypasses the anxiety and resistance students sometimes have toward formal instruction. Twine is a simple, easy-to use game authoring platform optimized for a busy instructor without a background in coding. Give it a try!

Open Research Practices

Uncategorized,

As I prepare to attend the 2023 Music Librarian Association Conference in St. Louis, I find myself excitedly browsing through the draft schedule, planning how to juggle my time in order to hear from as many of my music research colleagues as possible.Conferences such as this one, which pull from many different communities, offer opportunity to expand creativity by exploring ideas with people outside local communication channels (Rogers, 2003).  I am thankful for the opportunity to connect with others and learn about ideas or practices they have implemented that I might perceive as new, and consider how I might adopt those innovations as I seek to identify and solve problems in my own scholarly and creative work.

The Emerging Technologies and Services Committee (ETSC) Tech Hub will provide opportunities for attendees to explore a variety of innovations of potential use in identifying and solving problems. In response to MLA community feedback, this session will build on past years’ Tech Hub presentations and is planned to include topics and platforms facilitating data sonification for beginners, open research practices, music therapy and evidence synthesis, and digital music score platforms, among others. 

One of the platforms which will be provided for participant exploration is Pressbooks, an online platform intended to support the creation, adaptation and sharing of content. Attendees will have a hands-on opportunity to experiment with the platform via the OpenOKState program, the Oklahoma State University Libraries’ initiative supporting integration of open practices into research, teaching and learning at Oklahoma State University. A quick search of the term ‘music’ in the Pressbooks Directory returns at least 74 books whose authors have intentionally created and licensed them for use and customization by other scholars and instructors. During the ETSC Pressbooks session, participants will learn how they might adapt, customize, or even create similar resources of their own.

The Pressbooks session presentation is supported in part by a grant from the Institute of Museum and Library Services in support of a project using open research practices to explore open educational resources (OER) and lifelong learning. The goal of the project is to develop a replicable, reliable method to assess the efficacy of OER on lifelong learning competencies. Anticipated project deliverables include a toolkit applicable to multiple contexts which faculty can easily implement to measure the efficacy of OER on developing lifelong learning competencies in their own courses. A second deliverable will be an openly available book on research methodology focused on librarians conducting research. The entire project, including its final deliverables, has been intentionally implemented using open research practices.

An understanding of open research practices begins with an operational definition of research, itself. The Open Lifelong Learning project has defined research as a systematic investigation whose goal is identifying and/or solving problems. The term systematic refers to the intentionality with which the investigation is planned and implemented, and the goal leaves room for continued curiosity as well as provision of solutions as acceptable outcomes. ‘Open’ refers to transparent processes and practices through which the project is strengthened by the input of others’ expertise and experiences. 

One aspect of open processes has to do with the point at which the research is shared with others. Rather than waiting until the research has been completed and the project deliverables finalized to share their work, the Open Lifelong Learning team has presented at scholarly conferences throughout the research process. The intent of this transparency has been to seed ideas for a wide range of research projects, as well as to invite the unique expertise of other scholars. For example, a close study of empirical research into OER surfaces an emphasis on quantitative research investigating the impact of OER on outcomes such as DFW rates or grades. While these findings are useful, the field will benefit from research using a broader range of methodologies to explore a variety of outcomes. Another challenge has to do with the dispositions, skills, and subject matter understanding of individual researchers. As the Open Lifelong Learning team opened their work for input from others at scholarly conferences, questions were surfaced and answered by scholars and experts outside disciplines represented by the researchers. The outcome of this democratized approach to the scholarly conversation is a survey instrument which has been strengthened through intradisciplinary interrogation. It will also be interesting to note to what extent interest in the final project is influenced by others’ interaction with the process overall. 

The Pressbooks platform helps facilitate the open research process implemented by the Open Lifelong Learning team. Since both the process and the product embed ideas of contextual customization, the usability and discoverability of Pressbooks made use of the platform a logical choice. While open research practices can certainly take place independent of the Pressbooks platform, we hope those who are curious about its potential are able to come try it out during the MLA 2023 TechHub session.

Data Sonification for Beginners

Uncategorized

Data Sonification

What if you could make data more engaging? Imagine a data presentation that could elicit an emotional response from your audience. Data that can talk to you. Even sing to you. This is the world of data sonification.

We are all familiar with data visualization, the realm of techniques that translate data into visual images. These images allow us to grasp data patterns quickly and easily. We learn to produce and consume simple visualizations – pie charts, bar charts, line graphs – as early as elementary school. These are so ubiquitous, we rarely notice them.

Data sonification is analogous to data visualization, but instead of perceptualizing data in the visual realm, it perceptualizes data in the sonic realm. Sonification has a reputation as a cutting-edge, experimental practice, and in many ways it is just that. But it has also been around longer than many of us realize. David Worrall, in his 2019 book, Sonification Design, describes how the Egyptian Pharaoh audited granary accounts by having independently-prepared ledgers read aloud before him, and listening for discrepancies. (In fact, the very word, “audit,” comes from the Latin word meaning “to hear.”)

Another newer, but still retro, manifestation of data sonification should be familiar from cold-war era science fiction movies, or maybe old episodes of Mission Impossible: the sound of a Geiger counter, an electronic instrument that measures ionizing radiation. Hand-held Geiger counters characteristically produce audible clicks in response to ionizing events, to optimize their usefulness when the user’s attention must be focused somewhere other than reading a meter visually.

Modern attempts at computer-assisted data sonification began to gather speed in the early 1990s. A typical study is Scaletti and Craig’s 1991 paper, “Using Sound to Extract Meaning from Complex Data,” which explored the possibilities of parameter-mapped sonification using technology available at the time. The International Community for Audio Display (ICAD) was founded in 1992 and has held a conference most years since then. The Sound and Music Computing Conference and the Interactive Sonification Workshop both started in 2004. Sonification research is now regularly published in engineering journals, psychology journals, music journals, and small handful of specialty interdisciplinary publications like the Journal of Multimodal User Interfaces.

Most data sonification projects fall into one of three categories: audification, parameter-mapped sonification, or model-based sonification. Of these, audification is the simplest; it involves shifting a data stream into the audible realm by using it to produce sound directly, often dramatically speeding it up or slowing it down in the process. This has often been applied in seismology to allow researchers to listen to earthquakes, such as this sonification of the 2011 Tohoku Earthquake in Japan. It has also been applied to astronomical data, notably by NASA, and also in the work of Wanda Diaz Merced.

Model-based sonification is a much more subtle process. Here the technique is to take a basic sound and modify particular aspects of it according to data values. The sound must first be represented by a mathematical model, as is done routinely with musical instruments for computer music applications. Then, different parts of the model are made to interact with values representing different data variables. The resulting transformations of the model yield a new sound, which reflects the influence of the data values. Think of the sound of a bell being rung. The sound of the bell depends on various qualities: its size, its thickness, the ratio of length to width, how much it flares, what kind of metal it is made of. Vary any one of these, and the sound is altered. This is how model-based sonification works, except that a mathematical model of a bell is subjected to variation, rather than an actual bell. (No bells were harmed in the course of this research!) These four sounds use this kind of process to sonify distributions of neurons in artificial networks: id=1 cluster, id=3 cluster, id=5 cluster, id=6 cluster. (The examples are from Chapter 16 of Thomas Hermann’s The Sonification Handbook, and can be found along with others here.

Parameter-mapped sonification lies somewhere in the middle between these two approaches on the continuum of sophistication. In parameter-mapped sonification, individual sound parameters such as pitch, loudness, duration, or timbre are mapped to values in a dataset. This is the most accessible approach to sonification for most people; the easiest to grasp intuitively, and the easiest to experiment with. It works particularly well for single-variable, time-series data.

Low Barrier to Entry

A number of easy-to-use tools have been developed by sonification researchers to allow you to develop parameter-mapped sonifications on your own. One of these is TwoTone, developed by Sonify, Inc., in partnership with Google. Twotone is available as a free web app with an intuitive user interface. It comes with a library of existing datasets for you to play around with, or you can upload a spreadsheet file of your own to sonify. TwoTone will map your data onto midi pitches, according to ranges and constraints you can specify. In addition to sonification, it shows a real-time animated graph indicating what part of the data you are listening to at any particular moment, making it a multi-modal tool for experiencing data. You can download your sonification as an MP3 file, but to capture the visualization you need to use a screen recorder.

Another free web app for data sonification is offered by Music Algorithms, developed by Jonathan N. Middleton of Eastern Washington University. Music Algorithms does not offer a visualization to go along with its sonification, but it does offer duration as a parameter for sonification, which TwoTone does not. Where TwoTone comes pre-loaded with sample datasets to play with, Music Algorithms offers mathematical series such as Pi, Fibonacci, or a DNA sequence, in addition to a “custom” function that allows you to input your own data. You can download your finished sonification as a MIDI file.

Much of the leading work in data sonification happens in the Sonification Lab at Georgia Tech. Their Sonification Sandbox was one of the first tools to allow public users to create their own sonifications. First released publicly in 2007, it is a free program you can download and install on your own computer. However the program is written in Java, and the creators have not kept it current with Java version updates. The last version (still available), is from 2014, and includes modifications to support Java 7. The most recent Java version is Java 19, released in Sept. 2022, and Sonification Sandbox works poorly with it. To get the best results with Sonification Sandbox, use a dedicated system (or a virtual installation within another system) running Java 7.

That doesn’t mean Georgia Tech has been sitting on its hands. Highcharts Sonification Studio, released in 2021, is a fully-updated web-based sonification platform, developed in partnership between GT and the data visualization software developer Highcharts. Users can upload a CSV file, choose data and sonification parameters, and produce a MIDI-based sonic rendering of their data.

Medium Barrier to Entry

Anyone who has spent much time around electronic composition is probably familiar with a visual object-oriented programming environment called Max, originally developed by Miller Puckette at IRCAM in the 1980s. Although not developed with data sonification in mind, this is one of Max’s capabilities. Max offers great flexibility, but it comes with a correspondingly steep learning curve. Fortunately, it is known for its great documentation, tutorials, and a user community not shy about posting instructional videos. If you are interested in using Max for sonification, tutorial 18 is the one you will be shooting for. Start at the beginning and take the tutorials one by one, and when you get to tutorial 18, you will learn how to use Max to convert spreadsheet data to sound and animated graphing.

Max is a little pricey, at least compared to a free web app; you can expect to pay around $400 for a license, or $250 with an academic discount. For the more adventurous, there is a free, open-source alternative called Pure Data. Pure Data (or PD), also developed by Puckette, is a completely separate and independent tool, but is designed to do the things Max does, using an interface similar to that of Max. The big difference is in the documentation: PD’s documentation is mostly community-developed, so it isn’t always as beginner-friendly as the documentation in Max. However, if you are patient, you can learn to do the same things in PD that you can do in Max. Besides being free, PD also has the advantage that it is available in a version for Linux, as well as for MacOS and Windows. (Max is available for Windows and Mac only.)

Sonification for Librarians

So what might you do once you get your hands on these tools? Good question! Here are a few sonifications I have created using the humblest data at my disposal: the log of the gate counter in my music library at University of Denver. Using TwoTone, I created a sonification (and recorded the animated graph) of patron visits to the library in FY 2018. Play the video, and watch for the small orange line moving from the left through the rows of blue lines. You will notice that higher pitch correlates to higher values in the sonified data.

The top row is a repeating sequence of seven values indicating weeks in the year; it is placed in the lowest range. The next row is the mornings, with the noon hour included; it is in a higher range. It begins with a period of lower values representing reduced library traffic in the middle and late summer, then jumps dramatically when school begins in the fall. The next row is the afternoon/evening reading; it is in an even higher range. You may notice that after school begins, the gaps in this row between groups of values seen in the summer disappear. This is because during school breaks we do not open on the weekend, while during school terms, we have weekend hours in the afternoon. (But not in the morning – note the contrast with the row above.) University of Denver is on the quarter system; you can easily identify Fall, Winter, and Spring Quarters, separated by a long Winter Break and a much shorter Spring Break. The last row is the night shift, and it is placed in the highest range of all; I have also further distinguished it by representing it in two-note arpeggios. You can see that we are open nights only during school terms.

The same data was differently, for this sonification. Here, the daily totals were used instead of individual shifts, but Fall, Winter, and Spring Quarters were mapped against each other as a comparison.

Here is another sonification of the quarter-against quarter data, this one created using Music Algorithms. Data values are again mapped to pitch. The time values in this sonification are manipulated in such a way that they grow longer in a repeating cycle of seven values, with Monday as the shortest and Sunday as the longest. This allows us to identify the weekend data as the two longest time values at the end of each series.

A sonification of the same data using Sonification Sandbox is here. Like the other tools, this one has its own look and sound. The recording is a little slow; it takes 60 seconds to hear all of it. You might want to try speeding it up on YouTube; alternatively, this recording completes the process in ten seconds.

Here is a sonification created with Max. In this case, three years’ gate-count data were superimposed on each other, with each displayed in a different-colored graph, and all three sonified simultaneously.

Listening to these sonifications, you are likely to experience them asthetically, or even emotionally, in a way that rarely happens when we inspect a visual chart or graph. This is possibly the most unusual aspect of data sonification: the immediacy and urgency of sound make vision seems cool and analytical by comparison. Unlike our eyes, we cannot “close” our ears; nor can we “listen away” from one stimulus as we can look away in order to focus on another. This difference in quality between aural and visual perception helps explain why so many sonification tools are designed for multimodal presentation – hearing and vision are designed to reinforce each other.

Try some of these tools and see what you can create with data sonification. If you come up with something interesting, post a comment here!