Computational technologies that process information, learn, design, and problem solve are poised to transform many aspects of life, including how we discover, educate, remember, make decisions, and even express ourselves. In the 1950s, scientists such as Marvin Minsky and Alan Turing began publishing papers that described intelligent machines. Soon after that, the term “artificial intelligence” (AI) emerged. Today, AI refers to solutions demonstrating new and profound capabilities, sparking optimism for the role this technology might play in reshaping and elevating society.

Artificial intelligence has already taken hold in many industries ranging from education to energy management, law enforcement, medicine, and national defense. Achieving an equitable AI future that benefits communities across society will require broad participation. Institutions of higher education, especially tribal colleges and universities (TCUs), can bring critical support by engaging in this process, helping ensure that AI technologies are developed to address the priorities and protect the interests of the communities they serve. With an ever-growing presence of artificial intelligence technologies, shaping and advancing AI responsibly is not a task for computer scientists and engineers alone. It will require participation from the public and the involvement of experts from the humanities, economics, the social sciences, and many other non-engineering domains.

Societies are dynamic, living systems that continually change and adapt through the exploration and diffusion of new ideas, practices, and technologies. The experience of American Indian tribes since settler-colonist contact has been one of transformation and extreme disruption, including the tragic loss of languages, traditional knowledge, and lifeways. However, all tribes existing today have proven resilient and resourceful, continuing to respond to new existential challenges. AI technologies, when properly designed and managed, can provide tools and resources that strengthen tribal resilience and further the sustainability goals of tribal nations and communities. Without active tribal participation in AI development, these technologies could instead bring new disruptive challenges.

This article contributes to a growing conversation among organizations and individuals engaged in furthering sustainable tribal nations, focusing on potential opportunities and risks of AI development for tribal communities.

HEALTH AND MEDICAL SOLUTIONS

Advancements in artificial intelligence in the health and medical sciences afford opportunities for significant improvements across a broad range of applications, including telemedicine, drug design, medical imaging, personalized medicine, diagnostics, wearables, robotic surgery, and assisted living. The AI platform AlphaFold, for example, allows researchers to improve the quality, speed, and precision of drug development by revealing the protein structures that make up every aspect of our bodies. AI algorithms are also significantly improving the accuracy and speed of analyzing medical images such as x-rays, computed tomography (CT) scans, and magnetic resonance imaging (MRI). AI models are beginning to predict diseases before they manifest by analyzing patterns in historical health data and lifestyle information, and through real-time monitoring utilizing devices such as smartphones, watches, rings, and wristbands. These advancements are improving healthcare and maintenance for many people. As health and medical applications of AI become more effective at addressing health issues experienced by smaller populations, this can drive the resolution of health disparities across issues particularly vital to American Indians.

In the December 2023 issue of the Journal of the American Medical Association, a team of subject-matter experts presented five guiding principles for health AI algorithm development for patients, health professionals, and other stakeholders in the health AI community: 1) promote health and healthcare equity during all phases of the healthcare algorithm lifecycle; 2) ensure healthcare algorithms and their use are transparent and explainable; 3) authentically engage patients and communities during all phases of the healthcare algorithm lifecycle and earn trustworthiness; 4) explicitly identify healthcare algorithmic fairness issues and tradeoffs; 5) establish accountability for equity and fairness in outcomes from healthcare algorithms.

These guidelines call for participation from all minority populations experiencing health disparities in all phases of health AI development and implementation. To further ensure that appropriate solutions are developed, minority populations, in addition to participating in a consultative role, can directly participate in AI research and development through programs such as the Artificial Intelligence/Machine Learning Consortium to Advance Health Equity and Researcher Diversity (AIM-AHEAD), a major initiative funded by the National Institutes of Health.

AIM-AHEAD was established to facilitate partnerships that increase the participation and representation of researchers and communities currently underrepresented in the development and improvement of health AI solutions. The program offers funding for research projects, fellowships, and training so that project partners can gain and apply AI-related skills while participating in the work, or while “on the job.” AIM-AHEAD is positioned to serve TCUs and tribal nations interested in developing AI applications that specifically target American Indian health issues and drive local and regional improvements in health services delivery. The program prioritizes community-based participatory research (CBPR) to ensure that projects are responsive to and, to an extent, driven by health research and service delivery priorities of the populations being served.

In addition to AIM-AHEAD, TCUs and tribal nations can participate in programs such as Jetstream and ACCESS, each funded by the National Science Foundation, to open broader access to scientific tools, datasets, and infrastructure. The software, data, and library carpentry programs, supported through a volunteer-driven organization operating since 1998, also provide data-science resources and workshops designed for non-technical students, educators, and researchers. Each program offers opportunities to engage in developing data and AI-related skills.

EDUCATION FOR STUDENTS AND PROFESSIONALS

Emerging artificial intelligence teaching and education tools can incorporate an individual’s cultural, cognitive, and emotional profiles, providing students and teachers with ways of presenting, connecting, and interacting with content that is responsive to a student’s learning preferences, cognitive style, and interests. By personalizing learner journeys based on students’ strengths, weaknesses, and other relevant characteristics, AI can help students stay engaged while optimizing for targeted educational outcomes. These solutions can employ motivational triggers and other forms of feedback to encourage engagement and steer students toward adopting specific problem-solving strategies and practices. Beyond enhancing traditional teaching practices, these platforms can provide adaptive upskilling experiences that reflect the changing needs of professions and market spaces or guide learners toward addressing the workforce development needs and opportunities of a region.

AI systems that manage and improve student performance are poised to become foundational to modern education and can help address the capacity challenges educational institutions often face. Teachers who struggle to respond to the needs of each student in a large class can benefit from systems that support not only individual learners but entire classes by actively facilitating collaborative learning. Such collaborative learning experiences are active, social, and can be either formally structured or more informal and student driven. They offer a wide range of benefits, encouraging higher-level and innovative thinking, shared responsibility, common goals, collective effort, and exposure to diverse perspectives. AI-facilitated collaborative learning tools can be used to organize and support study groups that span a range of skills, interests, and perspectives, helping to optimize for the collective learning benefit of the group. In theory, AI systems could facilitate group exploration of a problem space through cycles of divergent and convergent thinking that foster group-based innovation in a problem-based learning context. This can be particularly valuable when student projects focus on specific local challenges and opportunities, such as those involving climate change. Perhaps most compelling, AI tribal education platforms can be designed to facilitate the acquisition and use of traditional knowledge practices, perspectives, and approaches to problem-solving while enabling student learning across standardized curricula. Teachers and education researchers, working closely with tribal cultural experts and traditional knowledge holders, can develop and implement these AI-supported models of Indigenous education to help tribal education systems meet state and national education requirements while achieving tribal cultural education goals.

LANGUAGE PRESERVATION AND REPRESENTATION

Many artificial intelligence platforms available today can engage in written and spoken dialogue. Advancements in the underlying approaches that support this capability, such as machine and deep learning, have even prompted researchers to explore the use of AI to unlock the languages of intelligent animal species through new scientific projects including Project CETI, established in 2020 to explore and eventually interpret the songs and other vocalizations of whales.

Large language models (LLMs) can also serve as powerful tools for language revitalization, education, and preservation. LLMs create representations of language by evaluating content made available when the model is developed. This process is called training. For languages, material from highly proficient speakers improves the process. When they are trained appropriately, models achieve high levels of fluency in terms of grammar, word choice, and intonation. LLMs represent the potential to create platforms that can write and speak in Indigenous languages and perform real-time translations.

Communities could deploy Indigenous-speaking AI platforms to teach and promote the use of languages through interactive practice. Learners could engage in sustained conversations, navigating between language translations in real time. This capability would require developing sufficient language data to effectively train and develop such a platform. Today’s most advanced LLMs, including ChatGPT from OpenAI, are exposed to very limited Indigenous language data during development, without which they cannot represent tribal languages well. Onri Jay Benally, a Navajo doctoral researcher at the University of Minnesota and fluent Navajo speaker, recently explored ChatGPT’s ability to generate Navajo translations. After working with the platform for just a short time, Benally determined that its translation capability is “quite muddy.” She added, “It certainly has access to the characters, though. It seems to be able to say ‘hi’ correctly and a few things like that.”

ChatGPT fails to indicate a lack of proficiency with the Navajo language. If prompted, GPT-4o—the conversational interface to GPT—will attempt to speak exclusively in Navajo without offering a caveat. In the case of these and other modern LLM solutions, in instances where a platform has not been sufficiently developed to represent a language, there is no assurance that the intonations demonstrated are phonetically accurate or that the words used are appropriate.

Figure 1. Midjourney AI-generated image from the prompt “render Tuba City, Arizona, in 2010” (above), compared to an actual photo of Tuba City (below).

According to OpenAI, ChatGPT currently supports more than 50 languages, none of which was developed by members of a North American tribe. Nevertheless, the technological approach used to generate LLM solutions can fully support the development of new platforms capable of accurately representing tribal languages. To advance toward Native-speaking LLMs, tribal nations collaborating with TCUs could organize data-gathering projects under data sovereignty protocols, using audio or video recording capabilities of smartphones and computers to capture the data necessary to train a model. At the same time, large language models can also be trained to preserve stories, conversational language, and expressions of local knowledge. The data resources developed could inform other AI applications to advance tribal language and cultural priorities, as suggested earlier.

KNOWLEDGE AND KNOWLEDGE PRACTICES

Modern AI platforms can offer a means of preserving and providing meaningful access to vast knowledge that tribes have accumulated and continue to retain, including through daily life practices, rituals and ceremonies, historical narratives, spiritual understandings, ethical and behavioral norms, and observations of natural phenomena.

Native scholars, cultural experts, and traditional knowledge holders, as well as writers, poets, and diarists, have been keepers and caretakers of tribal culture, maintaining and sharing traditional and culturally relevant understandings and continuing traditional practices in the face of continuous pressure to assimilate. Much of what has not been retained by individual tribal members due to historical factors associated with colonization has, for a number of tribes, been preserved and documented in print, audio, image, and video formats, as well as through collections of physical artifacts. While the methodologies and intentions of the linguists, anthropologists, historians, photographers, and journalists who have documented Native cultures were predominantly colonialist, the texts and other materials they generated that have been preserved represent available resources for those interested in making certain elements of cultural knowledge more broadly available and relevant to members of individual tribes today.

The traditional and cultural knowledge of living tribal members as well as the materials created by tribal members—including autobiographies, diaries, fictional works, poetry, research studies, and materials developed and preserved by non-tribal members since settler-colonist contact—represent the entirety of what is currently available to inform a tribal cultural memory AI system. A traditional knowledge AI system could function as much more than an encyclopedic collection of cultural knowledge. In addition to being able to retrieve specific stories, songs, and prayers, the platform could also carry out deep analyses of this material, identifying connections and relationships across knowledge and belief systems, practices, and ethical norms. AI approaches could identify significant relationships across individual instances of cultural expression. For example, they could identify relationships among motifs that may not be immediately obvious even to members and practitioners, in turn not just preserving but deepening cultural knowledge. Such a platform could then provide recommendations to tribal decision-makers dealing with new challenges, helping to inform solutions that arguably carry the authority of the ancestors.

AI platforms can deliver learning experiences in tribal history, culture, and practices. Michael Running Wolf (Northern Cheyenne/Lakota/Blackfeet) is exploring Indigenous language and culture reclamation by combining artificial intelligence with technologies such as augmented and virtual reality. He envisions an immersive environment where users can engage conversationally with Native speakers in culturally appropriate contexts. The technology needed to create powerful and immersive AI-driven learning environments that build both language skills and cultural knowledge is already available.

BIAS IN MODERN PLATFORMS

While industry giants such as Microsoft and Google are investing billions to advance AI, today’s most popular platforms often demonstrate bias, a result of the datasets and techniques used to develop them. Certainly, the primary responsibility for preventing these unacceptable outcomes rests with the organizations and individuals creating these platforms. Tribal data repositories are needed to inform the development of tribally driven AI solutions. With thoughtful consideration of which elements to share, opening up data from these repositories would also improve tribal representation across AI solutions that others develop.

Figure 2. Midjourney AI-generated image from the prompt “render Page, Arizona, in 2010”

For now, bias remains a fixture of many modern AI platforms, impacting the information, images, and other content they generate. As AI continues to spread into vital spaces of infrastructure and service, this bias can lead to severe consequences for those excluded or inappropriately represented.

A study published in 2019 revealed that a widely used healthcare AI algorithm, which was designed to identify patients with complex needs for enrollment in care-management programs, systematically underestimated the needs of African American patients compared to white patients with similar profiles. The algorithm had been trained using healthcare costs as a proxy for health needs, and African American patients typically incurred lower healthcare costs due to systemic barriers in accessing healthcare.

AI facial recognition systems from companies like IBM, Microsoft, and Amazon were shown to have higher error rates for darker-skinned individuals and were less effective when evaluating women. The system had been trained using primarily images of white male faces. Darker-skinned women were misidentified up to 35% of the time, while lighter-skinned men registered error rates of less than 1%. Such bias could motivate wrongful arrests and lead to heightened surveillance of minority communities.

In real estate, a 2021 study found that mortgage approval AI systems used by some lenders in the U.S. were biased against Native American, African American, Hispanic, and other nonwhite applicants. The algorithms were more likely to reject applications of individuals from these groups than applications from white individuals with similar financial profiles. This made it more difficult for minority families to achieve homeownership.

In addition to these instances of bias in special-purpose solutions, modern general-purpose AI knowledge platforms sometimes misrepresent important interests when prompted to provide specific information about tribal nations. When asked to describe the political structure of the Tohono O’odham Nation, ChatGPT only describes the original constitution established in 1937 and fails to mention the active constitution adopted on January 18, 1986. These misrepresentations may reveal underlying and unintentional mechanisms of bias. AI platforms like ChatGPT are designed to optimize for the amount of computational resources they utilize when generating a response because these resources are expensive. Companies that deploy AI are incentivized to ensure that results are generated as efficiently as possible for a user. These systems also employ a variety of techniques, such as inference, to interpret a request and rapidly locate relevant information so that a response can be generated quickly, again minimizing the amount of resources needed. By failing to represent the current governing constitution of the Tohono O’odham Nation, ChatGPT may be reflecting cost-saving behavior. In other words, the platform may have settled for locating the historical constitution and stopped short of determining whether representing that alone was an adequate response. The system achieved both the task of providing the user with a result and managing resource utilization, although it decidedly and unacceptably failed to adequately represent the current Tohono O’odham political structure. In addition to text and voice content, today’s AI platforms can also generate images, and newer platforms are emerging that can produce on-demand video content. While AI prompt-to-video generation is at an early stage, the potential for this technology to reshape learning, entertainment, and other industries seems profound.

Figure 3. Midjourney AI-generated image from the prompt “a modern Native American city”

As AI video generation continues to develop, it is not clear how well content generated by these platforms will represent tribal interests. Bias is apparent in results produced by modern AI image generation platforms. For example, when the popular image-generation platform Midjourney is prompted to represent Tuba City, Arizona, on the Navajo reservation, it portrays what appears to be a deserted shantytown with a broken-down vehicle at the center of the image (see Figure 1). The AI-generated depiction lacks contemporary housing or other modern features; the image better represents a community from several decades earlier. There are also many poverty-related elements, including a rusted roof, broken-down fencing, and dirt roads.

An AI-rendered photo of Tuba City juxtaposes dramatically with one of the off-reservation city of Page, Arizona, which is only 72 miles away with a similar population (see Figure 2). The differences between the images of Tuba City and Page suggest that Midjourney displays a bias toward tribal communities as non-contemporary and impoverished. This bias appears to be confirmed in alternative images. When prompted to render “a modern Native American city,” the platform produces a set of four images that fail to represent contemporary Native American life. Instead, the images appear to perpetuate idealized visions of Indigenous civilizations (see Figure 3).

When explicitly prompted to “represent artificial intelligence benefitting tribal nations,” Midjourney produced idealized environmental images or renderings of individual faces, each overlaid with abstract technological imagery (see Figure 4). These portrayals of bias represent the datasets and calibrations used to develop these AI solutions. Platforms such as Midjourney are not inherently predisposed to generating appropriate or inappropriate content. While we can recognize bias in ChatGPT, Midjourney, and other modern AI systems, alternative platforms developed using community-driven and responsible innovation practices could perform much differently.

Krystal Tsosie (Navajo) is an assistant professor at the School of Life Sciences and associate director for the Biodiversity Knowledge Integration Center at Arizona State University. Her work centers around technology, ethics, and Indigenous interests. As an activist for Indigenous data sovereignty, she points to the need for tribal community members to get involved, explaining, “Artificial intelligence has the potential to reify discrimination against Indigenous peoples and facilitate exploitation of our languages, cultures, and livelihoods through data. At the same time, under the governance and stewardship of Indigenous peoples, it can also represent a new way to increase our visibility in data, enable data protections, and provide moves towards Indigenous-led, data-based economies. Data about Indigenous peoples should be wielded by Indigenous peoples.”

INDIGENOUS PROTOCOLS FOR AI

In 2019, the Indigenous Protocol and Artificial Intelligence (IPAI) working group held two workshops. Participants from many tribal communities joined from across Aotearoa, Australia, North America, and the Pacific. Attendees responded to the project’s core question: “[From an Indigenous perspective] what should our relationship with AI be?” These discussions informed the development of a position paper on Indigenous protocol and artificial intelligence published in 2020.

Figure 4. AI-generated image from the prompt “represent artificial intelligence benefitting tribal nations”

This working group developed the following guidelines for Indigenous-centered AI design: 1) locality—AI systems should be designed in partnership with specific Indigenous communities to ensure the systems are capable of responding to and helping care for that community; 2) relationality and reciprocity—AI systems should be designed to understand how humans and non-humans are related to and interdependent on each other; 3) responsibility, relevance, and accountability—AI systems developed by, with, or for Indigenous communities should be responsible to those communities, provide relevant support, and be accountable to those communities first and foremost; 4) development of governance guidelines from Indigenous protocols—protocol is developed out of ontological, epistemological, and customary configurations of knowledge grounded in locality, relationality, and responsibility. Indigenous protocol should provide the foundation for developing governance frameworks that guide the use, role, and rights of AI entities in society; 5) recognition of the cultural nature of all computational technology—all technical systems are cultural and social systems. AI system designers need to be aware of their own cultural frameworks, socially dominant concepts, and normative ideals; be wary of the biases that come with them; and develop strategies for accommodating other cultural and social frameworks; 6) application of ethical design to the extended stack—culture forms the foundation of the technology development ecosystem, or “stack.” Every component of the AI system hardware and software stack should be considered in the ethical evaluation of the system. The core ethic should be that of do-no-harm; 7) respect and support data sovereignty—Indigenous communities must control how their data is solicited, collected, analyzed, and operationalized. All AI systems should be designed to respect and support data sovereignty.

The American Indian Higher Education Consortium’s (AIHEC) Indigenous Evaluation Framework (IEF) offers a similar set of core ideas within the realm of education. Developed by Dr. Joan LaFrance and Richard Nichols in collaboration with AIHEC, and supported through funding from the National Science Foundation, the IEF project was established to address the need for an evaluation methodology robust enough to accommodate and value different ways of knowing within Indigenous epistemologies, build ownership and a sense of community among Indigenous educators and evaluators, and effectively contribute to developing high-quality and sustainable science and mathematics education programs. The IEF provides an Indigenous perspective on education programs that can be applied to AI systems intended to support tribal goals more broadly.

Beginning in 2003, LaFrance and Nichols conducted an extensive consultation process, engaging cultural experts, American Indian educators, evaluators, parents, and other stakeholders in focus group discussions across tribal communities. Through this broadly inclusive process, LaFrance and Nichols identified four core values for Indigenous evaluation: 1) respect for Indigenous knowledge creation—Indigenous knowledge practices are increasingly being recognized by the scientific community as making important contributions to understanding across disciplines in which traditional knowledge is most relevant, such as biology, ecology, health, and history; 2) emphasis on local context in the implementation and evaluation of evidence-based practices—Indigenous learning is primarily place-based, therefore education programs should, to the extent possible, reflect this; 3) incorporation of cultural components in learning models and evaluation— education programs serving Indigenous students must include local, culturally based outcomes and learning activities; 4) need for community engagement during every phase of education program implementation, from planning to evaluation—education programs must be grounded in and reflective of the community; engagement of the community is essential.

The IEF core values and the IPAI development guidelines underscore the importance of community engagement in any initiative involving Indigenous people. AI systems are cultural artifacts that are designed for and reflective of the cultural understandings of their creators, which, for prevailing systems today, can often be characterized as biased. Ensuring that AI for Indigenous or American Indian use truly reflects Indigenous values and supports Indigenous goals will require the participation and contributions from the communities for which these systems are being designed.

TCU NEXT STEPS AND SUGGESTIONS

The first essential step toward developing tribal AI solutions involves developing and managing tribal data. Success in the emerging world of AI is predicated on access to high-quality data. A first step will be identifying current data repositories that contain resources relevant to local interests, including health, environmental, and public safety. Then, through data-resource mapping, any gaps relative to the development of specific AI applications can be identified before moving on to collect and prepare additional resources needed.

In addition to facing the challenge of collecting and managing tribal data necessary for AI development, tribal nations need to protect sensitive data by applying data sovereignty protocols. The Sovereign Data Network (SDN), a project funded by NSF and led by Southwestern Indian Polytechnic Institute, is a network of locally managed databases that support the CARE principles (Collective benefit, Authority to control, Responsibility, Ethics) of Indigenous data governance developed by the Global Indigenous Data Alliance. The core data management platform can be customized to work with all data types. Tribes and TCUs can review the SDN model as they establish data management infrastructure to support local AI developments. Programs such as IndigiData also serve as effective models for participant training and incorporating tribal data safely for advanced data science.

Building on the work of the Indigenous Protocol and Artificial Intelligence working group and the IEF development team, AIHEC and the TCUs, in partnership with Arizona State University and others involved in AI research, can organize an invitation-only workshop on tribal AI. Invitees could include tribal leaders, educators, cultural experts, and elders, as well as AI researchers and developers, social scientists, and linguists. The workshop’s purpose would be to sketch out a framework for developing and applying AI systems to support tribal sustainability interests across economics, education, culture, and health. A secondary goal could be to organize an Indigenous AI working group to develop proposals and secure funding for AI projects consistent with tribal goals.

AI systems can be designed to support Indigenous understanding, perspective, and values through enhanced learning, problem-solving, and decision-making, providing important and possibly transformative solutions for current and future generations of American Indian leaders, educators, scientists, resource managers, entrepreneurs, and all those who are committed to sovereign tribal nations and people. We encourage everyone involved in the Tribal College Movement to find ways to get involved in making Indigenous AI a reality.

Sean Dudley serves as the chief research information officer and associate vice president for research technology at Arizona State University; Al Kuslikis is an independent consultant who focuses on tribal and TCU-relevant projects, including Indigenous digital humanities, STEM education, climate resilience, and artificial intelligence.

REFERENCES

Andreas, J., Beguš, G., Bronstein, M.M., Diamant, R., Delaney, D., Gero, S., et al. (2021). Cetacean Translation Initiative: A Roadmap to Deciphering the Communication of Sperm Whales. ArXiv. Retrieved from https://doi.org/10.48550/arXiv.2104.08614

Boerner, T.J., Deems, S., Furlani, T.R., Knuth, S.L., & Towns, J. (2023). ACCESS: Advancing Innovation: NSF’s Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support. Practice and Experience in Advanced Research Computing, 173–176. https://doi.org/10.1145/3569951.3597559

Buolamwini, J. (2019, January 24). Amazon Is Pushing Facial Technology that a Study Says Could Be Biased. MIT Media Lab. Retrieved from https://www. media.mit.edu/articles/amazon-is-pushing-facial-technology-that-a-studysays-could-be-biased/

Canavera, L. (2023). Blending Indigenous Knowledge and Artificial Intelligence to Enable Adaptation. WWF Arctic. Retrieved from https://www.arcticwwf.org/the-circle/stories/blending-indigenous-knowledge-and-artificial- intelligence-to-enable-adaptation/

Carroll, S.R., Garba, I., Figueroa-Rodríguez, O.L., Holbrook, J., Lovett, R., Materechera, S., et al. (2020). The CARE Principles for Indigenous Data Governance. Data Science Journal, 19(1). Retrieved from https://doi.org/10.5334/dsj-2020-043

Chin, M.H., Afsar-Manesh, N., Bierman, A.S., Chang, C., Colón-Rodríguez, C.J., Dullabh, P., et al. (2023). Guiding Principles to Address the Impact of Algorithm Bias on Racial and Ethnic Disparities in Health and Health Care. JAMA Network Open, 6(12). Retrieved from https://doi.org/10.1001/jamanetworkopen.2023.45050

Dastin, J. (2018, October 11). Insight—Amazon Scraps Secret AI Recruiting Tool that Showed Bias Against Women. Reuters. Retrieved from https://www. reuters.com/article/world/insight-amazon-scraps-secret-ai-recruiting-toolthat- showed-bias-against-women-idUSKCN1MK0AG/

LaFrance, J., & Nichols, R. (2008). Reframing Evaluation: Defining an Indigenous Evaluation Framework. Canadian Journal of Program Evaluation, 23(2), 13–31. Retrieved from https://doi.org/10.3138/cjpe.23.003

Lewis, J., Abdilla, A., Arista, N., Baker, K., Benesiinaabandan, S., Brown, M., et al. (2020). Indigenous Protocol and Artificial Intelligence Position Paper. Retrieved from https://www.semanticscholar.org/paper/Indigenous-Protocol-and-Artificial-Intelligence-LewisAbdilla/08e11a534e62f89a962f 409c9570575463eb2833

Lewis, J.E., Arista, N., Pechawis, A., & Kite, S. (2018, July 16). Making Kin with the Machines. Journal of Design and Science. Retrieved from https://doi.org/10.21428/bfafd97b

Martinez, E., & Kirchner, L. (2021, August 25). The Secret Bias Hidden in Mortgage-Approval Algorithms. Associated Press. Retrieved from https://apnews.com/article/lifestyle-technology-business-race-and-ethnicity-mortgages-2d3d40d5751f933a88c1e17063657586

Mattu, J.A., Jeff Larson, L., & Kirchner, S. (2016, March 23). Machine Bias. ProPublica. Retrieved from https://www.propublica.org/article/machinebias-risk-assessments-in-criminal-sentencing

Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations. Science, 366(6464), 447–453. Retrieved from https://doi.org/10.1126/science.aax2342

Indigitization. (n.d.). Tools for Digitizing and Sustaining Indigenous Knowledge. Retrieved from https://www.indigitization.ca/

OpenAI. (2023). ChatGPT (4.0 version) [Large language model]. Retrieved from https://chat.openai.com/chat

Running Wolf, M. (2019, March 18). What Does the Future Look Like for AI? Indigenous AI. Retrieved from https://www.indigenous-ai.net/what-does-the-future-look-like-for-ai-4/

Scager, K., Boonstra, J., Peeters, T., Vulperhorst, J., & Wiegant, F. (2016). Collaborative Learning in Higher Education: Evoking Positive Interdependence. CBE Life Sciences Education, 15(4). Retrieved from https://doi.org/10.1187/cbe.16-07-0219

Will, J. (2024, February 26). Could AI Help Save Indigenous Languages? Dropbox. Retrieved from https://blog.dropbox.com/topics/workculture/could-ai-help-save-indigenous-languages