Addressing Bias in AI: What Adult Educators Should Consider

 

By Lilian H. Hill

 

Artificial intelligence (AI) is increasingly shaping how people learn, work, and access information. From adaptive learning platforms to automated feedback tools, adult educators are finding themselves navigating opportunities and challenges that come with these technologies. One of the most pressing concerns is bias in AI systems, a complex issue that raises questions of fairness, equity, and responsibility in teaching and learning.

 

Concerns about biased algorithms predate the current popularity of artificial intelligence (Jennings, 2023). As early as the mid-1980s, a British medical school faced legal repercussions for discrimination after using a computer system to evaluate applicants. Although the system’s decisions mirrored those of human reviewers, it consistently favored men and those with European-sounding names. Decades later, Amazon attempted to streamline hiring with a similar AI tool, only to find it was disadvantaging women —an outcome rooted in biased training data from a male-dominated tech workforce.

 

OpenAI, the creator of ChatGPT and the DALL-E image generator, has been at the center of debates over bias since ChatGPT launched publicly in November 2022 (Jennings, 2023). The company has actively worked to correct emerging issues, as users flagged examples ranging from political slants to racial stereotypes. In February 2023, OpenAI took a proactive step by publishing a clear explanation of ChatGPT’s behavior, providing valuable insight into how the model functions and how future improvements are being shaped.

 

Understanding Bias in AI

Bias in AI occurs when algorithms produce outcomes that are systematically unfair or unbalanced, often due to the data used to train these systems. When the data reflects historical inequities, stereotypes, or informational gaps, AI may unintentionally reproduce or amplify those patterns (Mehrabi et al., 2022). For instance, résumé screening tools trained on past hiring data may undervalue applications from women or people of color (Dastin, 2018). Similarly, language models can generate content that perpetuates cultural stereotypes (Bender et al., 2021), and facial recognition systems may be less accurate for specific demographic groups, particularly individuals with dark skin (Buolamwini & Gebru, 2018). Understanding that AI bias often mirrors societal biases enables adult educators to engage with AI tools more critically and thoughtfully.

There are three primary sources of biased data: 1) use of biased training data, 2) human influence on training AI systems, and 3) lack of a shared understanding of bias.

 

1.    Biased Training Data

AI models learn from vast datasets that reflect the world as it is, including its prejudices. Just as humans are shaped by their environments, AI is shaped by the data it consumes, much of which comes from a biased internet. For instance, Amazon’s hiring algorithm penalized women because it was trained on historical data that was male-dominated. When datasets disproportionately represent particular groups or viewpoints, the model’s outputs reflect that imbalance. In short, there’s no such thing as a perfectly unbiased dataset.

 

2.     Human Influence in Training

After initial training, AI outputs are refined through Reinforcement Learning with Human Feedback (RLHF), in which human reviewers judge and rank responses. While this helps shape AI into behaving more like a “responsible” human, it also introduces personal and cultural biases. If all reviewers share similar backgrounds, their preferences will influence how the model responds, making complete neutrality impossible.

 

3.    No Shared Definition of Bias

Even if we could remove all data that reflects human bias, we would still face one unsolvable problem: people disagree on what bias means. While most can agree that discrimination is harmful, opinions vary widely on how AI should navigate complex social, political, or moral issues. Over-filtering risks producing a model that is so neutral it becomes unhelpful, stripped of nuance and unable to take a stand on anything meaningful.

 

Why This Matters for Adult Education

Adult learners bring diverse backgrounds, identities, and experiences into the classroom. AI tools built on non-representative data can worsen existing inequalities in education unless developers improve their training methods and educators use the technology thoughtfully (Klein, 2024). When AI tools are introduced without awareness of bias, the risk is that inequities become amplified rather than reduced (Holmes et al., 2022). For instance:

 

• Learners from marginalized groups may encounter materials or assessments that do not accurately represent their knowledge or potential.
• Automated tutoring or feedback systems may respond differently depending on dialects, accents, or language use.
• Predictive analytics used to flag “at-risk” learners could disproportionately affect specific student populations (Slade & Prinsloo, 2013).

 

Educators play a pivotal role in mediating these risks, ensuring that AI supports equity rather than undermining it.

 

What Adult Educators Should Consider

1. Critical Evaluation of Tools
• Ask: How was this AI system trained? What kinds of data were used?
• Explore whether the developers have published documentation about bias testing (Mitchell et al., 2019).
2. Transparency with Learners
• Explain how AI is being used in the classroom and its potential limitations.
• Encourage learners to evaluate outputs critically rather than accepting them at face value.
3. Centering Equity and Inclusion
• Select tools that offer options for cultural and linguistic diversity.
• Advocate for systems that are designed with universal access in mind (Holmes et al., 2022).
4. Ongoing Reflection and Adaptation
• Keep a reflective journal or log of how AI tools perform with different groups of learners.
• Adjust teaching strategies when inequities appear.
5. Collaborative Dialogue
• Create opportunities for learners to share their experiences with AI.
• Engage in professional learning communities where educators discuss emerging issues and solutions.

 

Moving Forward

AI literacy is more crucial than ever. When talking about AI with your adult learners, ensure they understand that these models are not flawless, their responses shouldn't be accepted as the absolute truth, and that primary sources remain the most reliable. Until better regulations are in place for this technology, the best approach is to "trust but verify." AI technologies are not neutral—they mirror the values, assumptions, and imperfections of the societies that create them. For adult educators, the challenge is not to reject AI outright but to engage with it thoughtfully, critically, and ethically. By proactively recognizing and addressing bias, educators can help ensure that AI contributes to inclusive, empowering learning environments.

 

References

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–623. https://doi.org/10.1145/3442188.3445922

Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research, 81, 1–15. http://proceedings.mlr.press/v81/buolamwini18a.html

Dastin, J. (2018, October 10). Amazon scraps secret AI recruiting tool that showed bias against women. Reuters. https://www.reuters.com/article/idUSKCN1MK08G

Holmes, W., Porayska-Pomsta, K., Holstein, K., Sutherland, E., Baker, T., Shum, S. B., Santos, O. C., & Koedinger, K. R. (2022). Ethics of AI in education: Towards a community-wide framework. International Journal of Artificial Intelligence in Education, 32(4), 731–761. https://doi.org/10.1007/s40593-021-00239-0

Jennings, J. (2023, August 8). AI in education: The bias dilemma. EdTech Insights. https://www.esparklearning.com/blog/get-to-know-ai-the-bias-dilemma/#:~:text=Some%20things%20teachers%20can%20do%20to%20help,use%20primary%20sources%20as%20the%20best%20sources

Klein, A. (2024, June 24). AI's potential for bias puts onus on educators, Developers. Center for Education Technology. https://www.govtech.com/education/k-12/ais-potential-for-bias-puts-onus-on-educators-developers#:~:text=Schools%20should%20be%20wary%20if,'%22

Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2022). A survey on bias and fairness in machine learning. ACM Computing Surveys, 55(6), 1–35. https://doi.org/10.1145/3457607

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. Proceedings of the Conference on Fairness, Accountability, and Transparency, 220–229. https://doi.org/10.1145/3287560.3287596

Slade, S., & Prinsloo, P. (2013). Learning analytics: Ethical issues and dilemmas. American Behavioral Scientist, 57(10), 1510–1529. https://doi.org/10.1177/0002764213479366