January 11, 2024: 4 pm Bonn & Prague time | 3 pm London time

Ethical Artificial Intelligence for Cultural Good - Challenges and potentials of AI in preserving and advancing cultural diversity

by Desiree Custers

Main takeaways:

  • The term “AI” has many different definitions, depending on the context in which it is used. Even experts are going back and forth on how to best define AI.
  • Cultural context should not be ignored during AI model development, as it can play a role in the way the model learns and, ultimately, the way the model is used in real-life applications.
  • As the field of AI is still relatively new, it is crucial to understand how AI models are shaped by their cultural context and what the implications are for preserving cultural diversity.
  • When collecting data for developing AI, caution should be exercised as to ethically gather information on people, especially marginalised groups. 
  • AI should not be seen as only a threat, as it also offers many avenues for exploring culture in new and exciting ways. 
  • AI can provide a tool that produces cultural meaning, explores cultural connections, preserves religious and cultural heritage, and incorporates cultural diversity.

Artificial intelligence (AI) has become an indispensable aspect of our daily lives. But for all the positive developments it offers, there are also negative aspects when it comes to AI’s inclusion of cultural diversity. The third session of the Maqha Conversations series brought together an esteemed panel of guest speakers from Europe, West Asia, and North Africa (the EWANA region) who each presented their experiences and expertise on the cultural implications of AI, both in terms of challenges and potentials in the session titled: “Challenges and potentials of Artificial Intelligence in preserving and advancing cultural diversity”.

The session was held online on January 11, 2024, and featured the following speakers: Yasmine Boudiaf, a researcher and creative technologist, Mohammed Babikar, a data-analyst, traveller, ex-Comedian, and blogger, and Sophie Decher, a linguist and researcher working on natural language processing (NLP) applications. The session was moderated by Huda Azar, an analytics advisor and was attended by AI practitioners, interested participants, and people working in the field of relations between Europe, West Asia, and North Africa.

@Sophie Decher | Presentation at the third Maqha Conversation

What is AI? – a broad field of practice

The Maqha Conversation started with the speakers and participants exploring different definitions of AI. Before long, it became clear that finding one definition of the phenomenon is particularly difficult. The term “AI” may refer, for example, to the field of study, an instance of software used for a particular task, or a particular entity (“an AI”) that has “intelligent” capabilities. Many researchers in the field of AI work with AI models. Once they have been shown many examples of a certain type of task, these statistical models are able to “learn” to accurately complete new similar tasks themselves.

Large Language Models:  the importance of context  

Perhaps the most well-known modern AI models are those that have been trained on language data. Language models are commonly used in natural language processing (NLP) applications — the field of our first speaker, Sophie Decher– where a user inputs a query in natural language to generate a result. ChatGPT, for example, is based on a large language model (LLM), meaning it has been trained on massive amounts of language data. As such, it is important to take a linguistic perspective when analyzing the way these models work. 

In linguistics, language can be analyzed on different levels, starting with the smallest grammatical components of an utterance and working all the way up to pragmatic elements (contextual and/or cultural information that is often not explicitly stated). These levels of linguistic analysis are equally relevant for AI language models. Current research aims to find out how AI models analyze language and which linguistic levels of analysis are involved. Grammatical aspects of language are much easier for models to learn than pragmatic knowledge. It is very important to carefully craft the training data in order to prevent bias towards a particular cultural or linguistic context in the resulting model. Knowledge of how an AI model has been trained can help us understand why it reaches certain decisions in particular contexts and how to best make use of that model.

AI and the Quran challenge: a tool to cultivate curiosity and understanding

The second speaker, Mohammed Babiker, spoke about how AI could play a role in the Quran challenge, meaning the attempts to replicate the miraculous quality, the I’jaz (inimitability), of the Quran. This inimitability is considered to be the main proof of the prophethood of Muhammad ﷺ and Islam as a divine religion revealed by God. Attempts at the Quran challenge have taken place over the centuries, with one of the biggest examples being Musaylimah. 

Texts that attempt to match the Quran, as Mohammed explained, meet certain criteria, such as having a rhyming style (القافية), containing a call to a new message and being free from linguistic errors (فصيح). AI is seen as a massive turning point as it lowers the barriers for meeting these criteria. An example of the AI generated Surat Corona was given, which called for the prevention of Covid-19. 

Using AI for the Quran challenge, as Mohammed argued, provides a unique insight into AI’s potential. It is also relevant for those interested in Arabic and its poetic quality and it can provide a tool to explore the meaning of the Quran and cultivate curiosity for its linguistic value. However, in line with the points made by the first speaker, more specialised AI is needed to accurately combine message and meaning to meet the task of the Quran challenge. There is no specialised Arabic language AI model, and developing such a tool will need Arabic-language specialists. 

@Mohammed Babiker | Presentation Qur´an and AI

Ethical AI: preserving cultural heritage & making cultural connections 

The third speaker, Yasmine Boudiaf, presented her projects that use ethical AI to preserve cultural heritage and create cultural connections. The first An Algerian Techno-Ritual, uses machine learning to document facial tattoos of Algerian women, a cultural practice, as Yasmine described, that is at risk of being erased from collective memory. The second project, Mediterranean Hand Gestures, explores the language of hand gestures that is shared across the mediterranean region, another cultural phenomenon that risks dying out. This project links recorded hand gestures to language and intonation through machine learning. 

When implementing both projects, Yasmine is careful not to reproduce the colonial gaze, namely, not to exoticize Arab women and not to collect biometric data in an extractive manner. The projects take a community centred approach, empowering local people while archiving their cultural heritage. Here AI becomes a tool that can produce meaning. 

A third project, the AI Justice Matrix, expands on the idea of responsible use of AI by mapping ethical AI and the discourses and mechanics of how knowledge is formed and collected through AI. Using satire and ridicule, the Matrix reflects on policy making around and regulation of AI,  which is very unclear and complicated. The project sheds light on the background on which AI is formed, without necessarily providing answers. 

@Yasmine Boudiaf | AI As Artistic Tool

 AI and cultivating cultural diversity – recommendations:

Throughout the session, the dangers of AI were also discussed, such as the use of AI in the context of war (for example autonomous weapon systems), the AI arms race, and the collection of biometric and other data from individuals. Nevertheless, the Maqha Conversation concluded on the note that if used ethically, AI can also be a tool to produce cultural meaning, explore cultural connections, preserve religious and cultural heritage, and incorporate cultural diversity. 

To this end, the following recommendations were made: 

  • Policy-development on AI is behind, and should be developed to ethically deal with the data collected through AI as well as the protection of its users;
  • More clarity should be provided on the way AI is designed and used, specifically on the topic of cultural biases;
  • More research needs to be done to understand how AI models are shaped by their cultural context and what the implications are for preserving cultural diversity;
  • Culturally diverse AI models should be developed that include a variety of languages and cultural contexts. In the case of Arabic, this could be done with the help of language and cultural experts. 

Additional reading:

Kate Crawford – Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence (2022)

Speakers

Yasmine Boudiaf

Yasmine Boudiaf is a researcher and creative technologist. She is a fellow of the Royal Society of Arts and the Ada Lovelace Institute, researching anti-colonial ethics for artificial intelligence. She was named as one of ‘100 Brilliant Women in AI Ethics 2022’. She is currently collaborating with the CSNI at South Bank University and The Photographer’s Gallery on an Alan Turing Institute funded project on visual cultures and computer vision and has recently published a paper on ethnic disparities in higher education in England, which was featured in the Times Higher Education. She has researched and taught at universities in the UK and Sweden. Her creative and consultancy projects are listed on her website: https://yasmine-boudiaf.com/.

Sophie Decher

Sophie Decher is a linguist and researcher working on natural language processing (NLP) applications. She holds a B.A. in Anthropology from American University in Washington, D.C., U.S.A and an M.A. in Applied Linguistics from the University of Bonn in Bonn, Germany. Her interests include text classification, intercultural communication, and pragmatic competence.

Mohammed Babiker

Mohammed Babiker, originally from Sudan, is a data-analyst, traveller, ex-Comedian, and blogger who was raised in the UK and Saudi Arabia, and is currently based in Prague, Czechia. As a full time Data Analyst, he naturally has a lot of free time which he likes to spend getting into niche hobbies. Last year, he got into the online debate scene, which he had a university background in as he  was heavily involved in debate contests. Mohammed believes debates are a great way to learn at a deeper level and challenge your own ideas and beliefs. His favourite topics to debate are Religion and Politics. You can find Mohammed on the discord server Modern-Day Debate or follow him on twitter/ his blog.

Host

Huda Azar, analytics advisor