Information is a Relation, Not a Commodity

Understanding

Algorithmic Bias and Its Impact on Indigenous Communities

1. What Causes Algorithmic Bias?
2. The Risks of Algorthmic Bias

3. Real-World Examples of Alogorithmics Bais

4. AI and the Unique Threats to Save American Spirituality and Language

02

Agenda

What Causes Algorithmic Bias?

03

Biases in Training Data:

• The data used to teach algorithms often carry historical injustices or incomplete representation. When flawed data—non-representative, historically biased, or insufficient

—are fed into AI models, the resulting decisions tend to reinforce and amplify those very biases. For example, an AI trained on arrest records reflecting racial disparities will perpetuate those disparities in predictive policing.

​

Biases in Algorithm Design:

• Developers may unintentionally encode their own conscious or unconscious biases when designing AI logic. The process of weighting variables or selecting features might reflect assumptions that introduce unfair prejudices into outputs.

04

What Causes Algorithmic Bias?

Unpacking Algorithmic Bias: What You Need to Know

Biases in Evaluation:

• Even neutral algorithms can produce skewed outcomes depending on how humans interpret and implement the results. Preconceptions influence decision-makers and can cause unfair actions post-analysis.

Biases in Proxy Data:

• Sometimes AI cannot use sensitive attributes like race or gender directly and resorts to proxies— indirect indicators such as postal codes or income levels. These proxies can correlate inaccurately with protected traits and lead to discrimination.

05

What Causes Algorithmic Bias?

Unpacking Algorithmic Bias: What You Need to Know

Biases in Data Labeling and Correlation vs. Causation:

• During training, data may be mislabeled or the AI might mistake correlation for causation, leading to flawed conclusions. An infamous example relates to an algorithm wrongly associating ice cream sales with shark attacks, when both happen to increase in summer but are unrelated.

06

What Causes Algorithmic Bias?

Unpacking Algorithmic Bias: What You Need to Know

The Risks of Algorithmic Bias

07

Unveiling the Hidden Risks of Algorithmic Bias

Discrimination and Inequality:

• When AI replicates historic injustices present in training data, it can make decisions that unjustly target certain racial or ethnic groups. For Native peoples, this may exacerbate existing inequalities in law enforcement, access to medical care, job opportunities, and loan approvals.

Legal and Reputational Damage:

​

• Organizations employing biased AI face potential lawsuits, financial penalties, and tarnished reputations. The concept of disparate impact holds that even neutral-seeming algorithms can cause disproportionate harm to protected classes, triggering legal scrutiny.

The Risks of Algorithmic Bias

08

Real-World Examples of Algorithmic Bias

09

Algorithmic Bias: Real-World Impacts and Insights

Criminal Justice:

• The COMPAS risk assessment tool in the U.S. was found to disproportionately classify Black defendants as higher risk compared to white counterparts, raising concerns about racial prejudice baked into automated sentencing aids.

Predictive Policing:

​

• Algorithms trained on police report data in predominantly Black U.S. cities, over-predicted crime in neighborhoods with high reports involving Black residents, demonstrating how social biases in data distort AI forecasts.

​

Healthcare:

• Clinical AI systems sometimes yield less accurate diagnostics for minority patients due to underrepresentation in medical data sets, potentially leading to inferior treatment outcomes.

Cascading effects of health inequality and discrimination manifest in the design and use of artificial intelligence (AI) systems

Real-World Examples of Algorithmic Bias

10

Algorithmic Bias: Real-World Impacts and Insights

Recruitment:

• Amazon scrapped an AI hiring tool that showed systemic bias against women because it was trained primarily on resumes from male candidates.

Financial Services:

• Research revealed that mortgage algorithms charged minority borrowers higher interest rates than white borrowers with the same credit profiles.

​

Facial Recognition:

• Studies at MIT exposed significant failures in recognizing darker-skinned individuals, particularly women, owing to skewed training datasets.

​

Pricing Algorithms:

• Ride-hailing apps were found to charge higher fares in predominantly non-white neighborhoods, exemplifying economic bias embedded in AI pricing strategies.

Real-World Examples of Algorithmic Bias

11

AI and the Unique Threats to Native American Spirituality and Language

12

Language Is Living

Cultural Erasure and Linguistic Homogenization:

• Most AI is trained on massive datasets dominated by English and other major languages. When applied to Native languages or dialects, AI frequently strips away rich nuances, slang, and contextual meanings, forcing local expressions into standardized, generic forms. This sanitization risks eroding the living character of Indigenous speech.

​

Unconsented Data Harvesting:

​

• Indigenous communities have voiced concern about "linguistic colonization," where tech companies scrape Indigenous texts and recordings without free, prior, and informed consent. This unauthorized collection threatens control over sacred cultural knowledge and violates digital sovereignty..

AI and the Unique Threats to Native American Spirituality and Language

13

Cultural Distortion Risks

Hallucinations and Factual Inaccuracies:

• AI chatbots often invent words or blend dialects inaccurately, undermining their usefulness in preserving endangered tongues. Such hallucinations can mislead learners and distort cultural transmission.

AI and the Unique Threats to Native American Spirituality and Language

14

Protecting Indigenous Languages and Knowledge in the AI Era

15

Stewardship in

Practice

Community-Led Models:

• Rather than relying on corporate AI, some groups develop their own private, culturally accurate language tools that prioritize authenticity and data protection.

​

Mukwa doesn’t advise this as it still consumes extracted minerals predominantly found on Indigenous lands.

Protecting Indigenous Languages and Knowledge in the AI Era

16

Stewardship in

Practice

• The United Nations Declaration on the Rights of Indigenous Peoples (UNDRIP) framework emphasizes Indigenous control over cultural data. Rights groups urge international forums to enforce protections against unauthorized data harvesting.

​

Digital Sovereignty:

​

• Empowering Native peoples to govern their digital resources ensures AI applications respect cultural heritage and linguistic diversity.

Advocacy and Policy:

Protecting Indigenous Languages and Knowledge in the AI Era

17

Information is a Relation, Not a Commodity

Understanding

Algorithmic Bias and Its Impact on Indigenous Communities