This article is adapted from a series of presentations from 2022 to 2026 that I have given on AI and machine bias. It includes evidence from Māori working in the tech sector, TUANZ (2026) interviews with students, and the latest workforce data. Together, they reinforce that bias against Māori is built into AI from conception to deployment.
Artificial Intelligence (AI) is already making decisions that affect Māori lives: who is flagged by a risk model, whose face a camera recognises, which patients are prioritised and which applications are approved.
These systems are often described as objective, yet they aren’t. AI learns from data that people collected, and it follows rules that people wrote. When the people, the data and the rules leave Māori out, the bias that results is predictable.
This article sets out where machine bias comes from, the main types of bias Māori should understand, and some practical solutions grounded in Te Tiriti o Waitangi and kaupapa Māori.
Technology builders
Bias starts with who is at the table. Internationally, the technology sector is still dominated by a narrow group of people. In the United States, women held 27.5% of computer and mathematical jobs in 2025 (U.S. Bureau of Labor Statistics [BLS], 2026), and about half of the women who enter tech leave the industry by the age of 35 (Accenture, cited in CIO, 2025). For women of colour the gap is even wider. Black women hold 3% of US tech jobs and Latinx women hold 1% (National Academies of Sciences, Engineering, and Medicine, 2021).
Race also shapes who gets hired. In 2025, US computer and mathematical workers were 62.2% White, 25.2% Asian, 9.5% Black and 8.7% Hispanic or Latino (BLS, 2026).
It also shapes who gets promoted. An analysis of the San Francisco Bay Area tech sector found that in 2015, white women were 17% above parity for executive roles, while Black men were 41% below and Asian men 38% below. In other words, white women were substantially more likely to become executives than minority men or women (Gee & Peck, 2017).
An investigation by Reveal from The Center for Investigative Reporting found that in 2016, ten large Silicon Valley technology companies did not employ a single Black woman, three had no Black employees at all, and six had no female executives (Reveal, 2018).
Aotearoa New Zealand is no different. The most recent Digital Skills Aotearoa data puts Māori at about 4.8% of the technology workforce, even though Māori are around 17% of the population (NZTech, 2024).
Two years earlier the figure was about 4%. At that time, New Zealand had been bringing in 3,500 to 4,500 highly skilled IT professionals from overseas, while only a few thousand New Zealanders graduated each year (NZTech, 2022, as cited in Clayton et al., 2023). Progress is slow, and the sector has continued to look offshore before it looks to Māori.
What Māori in the sector say
The 2023 Mapping the Māori Tech Sector drew on a survey of 84 respondents and interviews with 30 Māori technology practitioners (Clayton et al., 2023). It concluded that since the tech sector began and the Internet arrived in New Zealand, Māori have not been resourced or supported to take part as partners in the industry. Participants described tokenistic engagement, and found that Te Tiriti awareness and application were either missing or not working.
When respondents were asked why so many Māori graduates struggle to find internships or jobs, they gave a long list of reasons. These included institutional racism, unconscious bias, racism in interviews, biased CV evaluation, assumptions built into job design and recruitment, “cultural fit” expectations, and hiring practices that favour Pākehā or overseas workers (Clayton et al., 2023).
Dan Walker of Microsoft NZ explained that interviewers and hiring managers who lack cultural competency can misread Māori candidates’ behaviour, such as avoiding eye contact or not talking themselves up. Māori who do get in can find themselves stuck in entry-level roles and miss out on mentoring and networks because they do not fit the conventional mould of a successful professional (Making Everything Achievable [MEA], 2023).
Automated CV screening and hiring tools are trained on past hiring decisions. If those decisions marked down candidates who were modest, avoided eye contact or did not fit a Pākehā idea of “cultural fit”, the tool will learn that pattern and apply it to every Māori applicant, without anyone in the room to notice.
The more recent Ka Toro report, based on interviews with eleven tech education and training providers working with ākonga Māori, shows that these barriers have not gone away (MEA, 2025).
It found that many Māori feel they must suppress their cultural identity in mainstream tech environments. Brittany Teei of 3 Bags Full named unconscious bias and structural racism as continuing barriers. Jay Tihema of Toi Kai Rawa described interns and learners facing openly racist employers who say the right things in public but show no alignment in practice. Nikolasa Biasiny-Tule of Digital Natives Academy described creative tech work in Rotorua going to Pākehā firms, and having to fight for a 3D animation project that was about to go to a Pākehā contractor (MEA, 2025).
The report also notes that closures and job losses across the tech and Māori education sectors affected nearly half of the people interviewed (MEA, 2025). The Māori-led programmes that are building the pipeline are themselves fragile.
For wāhine Māori the gap is starker still. Shanon O’Connor, director of Tōnui Collab, has said that less than 2% of the tech workforce are wāhine Māori, and that this must change (Gisborne Herald, 2023). Mel Gollan, founder of Ripa Global, set up the Wāhine Toa Level Up programme so that every iwi could send two young wāhine to meet product owners, chief technology officers and architects in person. Her reasoning was simple: young wāhine need to see it to be it (MEA, 2023).
The barriers start well before anyone applies for a job. In 2026, Te Ara Hihiko, a Tech Users Association of New Zealand (TUANZ) member kaupapa, listened to rangatahi wāhine about their experiences of technology. Their findings are revealing (Tech Users Association of New Zealand [TUANZ], 2026):
- “Technology is still seen as a boys’ space.” Technology advertising and gaming culture remain heavily masculine. Engineering and STEM subjects are presented as male spaces, female technology teachers and role models are largely absent, and girls are more often steered towards teaching, healthcare and caring professions.
- Boys receive more attention, encouragement and praise, and more support when they struggle. Girls are often assumed to be fine because they are quieter. They worry about being judged or labelled for asking questions or getting answers wrong. They are also expected to perform well every time, while boys are praised for succeeding once.
- Being Māori or Pasifika adds another layer. Daughters are often expected to become the backbone of their families. Their career choices must support the whānau financially and fit around care giving, while male family members are given more freedom to try other paths or make mistakes.
- What young wāhine want us to know. They do not want to be a “diversity hire.” When Māori and Pasifika women appear in advertising, they are usually shown in family settings rather than as professionals in their own right. They are not necessarily lacking confidence. What they need is help to navigate environments that undermine it.
As Te Ara Hihiko puts it, gender stereotypes shape what girls feel comfortable trying long before careers are considered (TUANZ, 2026).
If wāhine Māori are not encouraged to experiment, make mistakes and be visible in technology classrooms, we cannot expect them to go on to design the AI systems that will shape Māori lives. Te Ara Hihiko’s next initiative will look at the lived experiences of Māori and Pacific professionals already working in technology, including systemic barriers, psychological safety and cultural burnout.
Marae, hapū and whānau have very little say in how AI is designed, bought or used, including the AI that is used on them.
In the 2023 Māori tech sector survey, 64% of respondents rated AI as beneficial or very beneficial, but they were clear that AI development must align with Māori values and tikanga (Clayton et al., 2023). One interviewee said they wanted Māori specialists building AI systems, not just commenting on the AI that others build.
When the people designing AI do not include Māori, the systems they build will not see Māori clearly, and they will not reflect Māori values, tikanga or needs.
Important terms
Synthetic data
Synthetic data is data made by computers, not collected from real people, and designed to look like real-world data. It is often used when real data is hard to get. Research already struggles to get Māori participation, so there is a real risk that the gaps will be filled with synthetic Māori data.
That data would be based on assumptions about Māori instead of on Māori themselves, and it can lock existing bias into the system. This is a particular concern in health, where synthetic data is increasingly used. Research on primary healthcare data shows that when certain patient groups are under-represented, synthetic data generated from that data carries the same gaps and distortions (Draghi et al., 2024).
Data bias
Data bias happens when data is limited in some way. It gives an inaccurate picture of the population or does not tell the full story.
Consider what many New Zealand government datasets contain about Māori.
- Māori make up about 17% of the population but more than half (around 52%) of the prison population.
- Māori make up 36% of all working-age people receiving a benefit as the primary recipient (Welfare Expert Advisory Group, 2019).
- Of Māori deaths from ischaemic heart disease in 2000–2004, 45% occurred before the age of 65, compared with 11% for non-Māori (Curtis et al., 2007).
- New Zealand Police’s own Understanding Policing Delivery research found that, with all other variables held constant, being Māori increased the likelihood of prosecution by 11% compared with NZ Europeans. It also found that bias and structural racism were part of the reason Māori men were more likely to be stopped and tasered (RNZ, 2024a).
- Māori made about 20% of all complaints to Police in the period studied, but 55% of complaints about use of force (RNZ, 2024b).
Any risk model trained on prosecution data would inherit that 11% and present it as a fact about Māori. In the health system, the Waitangi Tribunal’s Hauora report found that the primary health care system has failed to address persistent Māori health inequities, in breach of Te Tiriti (Waitangi Tribunal, 2019).
Data like this records what the system has done to Māori: decades of structural bias. It does not describe who Māori are. If an AI system is trained on this data without that context, it will learn that being Māori is a “risk factor.” It will then repeat that pattern at scale and with the false authority of a machine.
The data can also be wrong about who is Māori. A 2025 kaupapa Māori study found that Māori were undercounted in prison data by around 405 people, because Corrections was not following national ethnicity recording protocols (Te Ao Māori News, 2025). Any model trained on that data starts from an inaccurate picture before a single line of code is written.
Algorithmic bias
An algorithm is a process or set of rules followed in calculations or other problem-solving. A human writes the algorithm that the AI follows and learns from (machine learning), and the AI learns to make decisions from its training data. Flawed training data can lead algorithms to make the same errors over and over, produce unfair outcomes, or even amplify the bias already in the data.
Automation bias
Automation bias is when a person favours a suggestion from an AI system over their own judgement. A caseworker, clinician or police officer may know something is wrong but still defer to the machine. This is why accountability must always rest with a person, not just with a “human in the loop” who signs off on the system’s output. We saw this play out in a Rotorua supermarket when a wahine Māori was falsely accused of being a person of interest on the Facial recognition systems output but two human beings, despite her protests and providing identification.
Facial recognition
Facial recognition systems (FRS) show clearly how all of these forms of bias combine. People of colour, Māori and Pasifika are more likely to be misidentified because the training data lacks people who look like them. The landmark Gender Shades study found that two widely used facial analysis benchmark datasets were made up overwhelmingly of lighter-skinned people. It also found that commercial systems misclassified darker-skinned women at error rates of up to 34.7%, compared with a maximum of 0.8% for lighter-skinned men (Buolamwini & Gebru, 2018).
In 2015, Google Photos labelled photos of Black people as gorillas. Instead of fixing the model, Google removed the “gorilla” label altogether (MIT Technology Review, 2018). Some systems could not detect a dark-skinned face at all.
In Aotearoa, an independent review commissioned by New Zealand Police found that accuracy and bias are key concerns with facial recognition technology, and its authors warned that live facial recognition would affect Māori the most (Lynch & Chen, 2021; RNZ, 2021). For Māori, moko kanohi adds another layer of risk that most systems were never designed or tested for.
Solutions
Understand the AI life cycle
Bias can enter at every stage of the AI life cycle: planning and scoping, data collection, model design, training, testing, deployment and monitoring. This means it has to be addressed at every stage too. Māori need to be part of every step of the AI life cycle and to analyse it, not just be consulted at the end.
Ask for model cards
Model cards are short documents released with machine learning models. They explain the context in which a model is meant to be used, how its performance was evaluated, and other relevant information (Mitchell et al., 2019). For example, a model built to analyse voter demographics would come with a model card. It would report how well the model performs across culture, ethnicity, location, sex and intersectional groups.
A model card typically includes:
- Model details: who developed the model, the date, version, type and architecture, the training algorithms, parameters and fairness constraints, contact details, citation and licence information.
- Intended use: the use cases the model was built for, what is out of scope, and the intended users.
Performance metrics: the model’s real-world impact, including results across demographic groups, environmental conditions and technical attributes. - Training data: where the data came from and how different factors are distributed within it. Some organisations restrict this because the data is proprietary.
- Quantitative analysis: known biases and limitations, including how the model performs outside its intended use.
- Ethical considerations and recommendations: privacy, fairness, and individual and societal impacts, plus recommendations for further testing and monitoring.
Any organisation buying or deploying AI that affects Māori should ask: is there a model card, and does it report performance for Māori?
Map, plan and scope the AI project
Before any AI project starts, organisations should:
- Include Māori representation at all stages of the AI lifecycle;
- Identify what data is needed to train the algorithm;
- Find out where the data comes from and check that it is accurate;
- Check whether the data fully represents the people the AI will be used on;
- Ask whether the data is biased;
- Work out what happens if the AI performs poorly;
- Consider the impacts it could have on individuals, whānau and the organisation; and
assess the organisation’s risk tolerance.
AI and data governance
Organisations should have AI ethics expertise at board level. Iwi, hapū, community and marae should be engaged in government AI consultations and projects. International best practice is a useful starting point, but it must be adapted for Māori instead of adopted unchanged.
Organisations should also set up a data sovereignty team to:
- Uphold Te Tiriti o Waitangi and He Whakaputanga;
- Respect tikanga Māori;
- Evaluate vendor policies and practices;
- Make sure Māori data sovereignty is applied; and
- Make sure all New Zealand legislation is complied with.
Fix the workforce, not just the model
A model card can show that a system is biased, but it cannot change who builds the next one. The Mapping the Māori Tech Sector report made recommendations to the Crown that go to the source of the problem (Clayton et al., 2023). They include:
- Appointing a Government Māori Chief Data and Digital Steward to work as an equal alongside the Government Chief Data Steward;
- Including an enforceable and monitored Te Tiriti clause in all government tech contracts and partnerships, giving Māori rights to employment, training, governance and data sovereignty;
- Requiring all Crown-funded tech boards to have a Te Tiriti clause and seats for Māori tech representatives in proportion to the Māori population;
- Co-designing all future lead tech research with Māori, using kaupapa Māori research methodologies; and
- Applying progressive procurement so that at least 8% of tech contracts go to Māori-led tech companies.
For the tech sector itself, the report asks organisations to move beyond token gestures of partnership, to recruit Māori into governance and senior roles, and to make sure Māori staff do not have to leave their culture at the door (Clayton et al., 2023).
The Ka Toro report adds that hiring practices must be actively challenged, and that culturally safe workplaces are needed so that rangatahi are not undermined from the outset (MEA, 2025). An organisation that cannot recruit and keep Māori staff will struggle to build AI that sees Māori clearly.
Kaupapa Māori solutions
Kaupapa Māori tools already exist and can be built into how organisations govern AI:
- Te Tiriti and kaupapa Māori-centred algorithmic impact assessments;
- Te Tiriti and kaupapa Māori-centred AI health checks;
- Creating and adapting Te Tiriti and kaupapa Māori AI frameworks, such as the He Tangata, He Karetao, He Ātārangi framework;
- Te Tiriti AI audits;
- Privacy impact assessments with a te ao Māori and tikanga lens;
- Reviews of organisational policy.
Conclusion
AI and Machine bias is not a technical glitch that a better algorithm will fix. It reflects who builds technology, whose data is collected and how, and who has the authority to say no. Preparing safely for the future means treating Te Tiriti o Waitangi as the constitutional threshold for AI in Aotearoa, not a consultation checkbox. It also means making sure Māori are present and have authority at every stage of the AI life cycle.
For a powerful introduction to these issues, the documentary Coded Bias (2020), available on Netflix, follows Joy Buolamwini’s discovery of racial bias in facial recognition and is well worth watching.
References
Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research, 81, 77–91. https://proceedings.mlr.press/v81/buolamwini18a.html
CIO. (2025). Women in tech statistics: The hard truths of an uphill battle. https://www.cio.com/article/201905/women-in-tech-statistics-the-hard-truths-of-an-uphill-battle.html
Clayton, J., Eruera, H., Taiuru, K., & Dunn, K.-M. (2023). Mapping the Māori tech sector: 2023 and beyond. Making Everything Achievable; Te Whare Wānanga o Awanuiārangi; NZTech. https://www.mea.nz
Curtis, E., Harwood, M., & Riddell, T. (2007). Cardiovascular disease. In B. Robson & R. Harris (Eds.), Hauora: Māori standards of health IV. A study of the years 2000–2005. Te Rōpū Rangahau Hauora a Eru Pōmare. https://www.otago.ac.nz/__data/assets/pdf_file/0024/303864/chapter-8-cardiovascular-disease-067747.pdf
Draghi, B., Wang, Z., Myles, P., & Tucker, A. (2024). Identifying and handling data bias within primary healthcare data using synthetic data generators. Heliyon, 10(2), e24164. https://doi.org/10.1016/j.heliyon.2024.e24164
Gee, B., & Peck, D. (2017). The illusion of Asian success: Scant progress for minorities in cracking the glass ceiling from 2007–2015. Ascend Foundation. https://www.ascendleadership.org/research
Gisborne Herald. (2023, June 21). Kōhine relish STEMM challenge in programme run by Tōnui Collab. NZ Herald. https://www.nzherald.co.nz/gisborne-herald/news/kohine-relish-stemm-challenge-in-programme-run-by-tonui-collab/AOBR72ZCTBAQHG7FFVNEHCHAGY/
Lynch, N., & Chen, A. (2021). Facial recognition technology: Considerations for use in policing. New Zealand Police. https://www.police.govt.nz/node/64936
Making Everything Achievable. (2023). Mapping the Māori tech sector interviews: 2023 and beyond. https://www.mea.nz
Making Everything Achievable. (2025). Reimagining a strong Māori tech workforce: Ka Toro [Final report]. https://www.mea.nz
MIT Technology Review. (2018, January 11). Google Photos still has a problem with gorillas. https://www.technologyreview.com/2018/01/11/146257/google-photos-still-has-a-problem-with-gorillas/
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019). Model cards for model reporting. In Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT* ’19) (pp. 220–229). ACM. https://doi.org/10.1145/3287560.3287596
National Academies of Sciences, Engineering, and Medicine. (2021, December 9). Actions needed by higher education institutions, technology companies, federal agencies to increase representation of women of color in tech [News release]. https://www.nationalacademies.org/news/2021/12/actions-needed-by-higher-education-institutions-technology-companies-federal-agencies-to-increase-representation-of-women-of-color-in-tech
NZTech. (2024). Digital Skills Aotearoa. https://nztech.org.nz/
Reveal from The Center for Investigative Reporting. (2018, June 25). Here’s the clearest picture of Silicon Valley’s diversity yet: It’s bad. But some companies are doing less bad. https://revealnews.org/
RNZ. (2021, December 9). Review prompts police to halt plans to use facial recognition technology. https://www.rnz.co.nz/news/national/457588/review-prompts-police-to-halt-plans-to-use-facial-recognition-technology
RNZ. (2024a, August 21). Māori men more likely to be stopped, tasered, prosecuted by police due to ‘bias’ and ‘structural racism’. https://www.rnz.co.nz/news/national/525785/maori-men-more-likely-to-be-stopped-tasered-prosecuted-by-police-due-to-bias-and-structural-racism
RNZ. (2024b, August 27). Report reveals why half of complaints about police use of force come from Māori. https://www.rnz.co.nz/news/top/526321/report-reveals-why-half-of-complaints-about-police-use-of-force-come-from-maori
Te Ao Māori News. (2025, December 4). Kaupapa Māori study exposes gaps in prison data and support for Māori. https://www.teaonews.co.nz/2025/12/04/kaupapa-maori-study-exposes-gaps-in-prison-data-and-support-for-maori/
Tech Users Association of New Zealand. (2026, November 19). Te Ara Hihiko hui November 2026 [Presentation slides]. TUANZ.
U.S. Bureau of Labor Statistics. (2026). Employed persons by detailed occupation, sex, race, and Hispanic or Latino ethnicity [Table 11, Labor force statistics from the Current Population Survey, 2025 annual averages]. https://www.bls.gov/cps/cpsaat11.htm
Waitangi Tribunal. (2019). Hauora: Report on stage one of the Health Services and Outcomes Kaupapa Inquiry (Wai 2575). Legislation Direct.
Welfare Expert Advisory Group. (2019). Whakamana tāngata: Restoring dignity to social security in New Zealand. https://www.weag.govt.nz/weag-report/whakamana-tangata/
Licence
© 2026 Karaitiana Taiuru. This work is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) licence.
Scope: This licence applies only to Dr Taiuru’s original text. It gives no rights over mātauranga Māori, whakapapa or other taonga referred to in this article, which remain under the tino rangatiratanga of the iwi and hapū they belong to.
How to cite this page
Taiuru, K.N. (2026, November 21). Māori Bias – AI and Emerging Technologies. Taiuru & Associates Ltd. https://www.taiuru.co.nz/bias-ai-and-emerging-technologies/