How educators can make warranted decisions as research, tools, and practice continue to evolve

This brief is offered in the spirit of inquiry rather than prescription. It continues the Foundations Series by focusing on a practical reality in today’s schools: educators are increasingly exposed to a wider array of “evidence” (e.g., research summaries, dashboards, analytics, AI-generated recommendations) through everyday systems and workflows, but the availability of more evidence does not diminish the need for professional judgment and ethical responsibility (Dieterle et al., 2024; Holmes et al., 2019).

Evidence and context

Across education, the conditions under which decisions are made appear to be shifting. There are tools and programs that produce recommendations (e.g., suggested next steps, automated feedback, predicted risk, ranked options), and these outputs can shape what educators treat as “evidence” in day-to-day instructional and other professional decisions (Wang, 2024; Williamson & Piattoeva, 2019). As such outputs become embedded in everyday workflows, they may come to be treated as default “next steps,” even when they may represent only a subset of possible options, shaped by model design, data availability, and optimization choices, for example (Bauer & Gill, 2024; Romeo & Conti, 2025; Seaver, 2019)

At the same time, research on teachers’ AI use and professional learning emphasizes that classroom integration is uneven and requires more than technical knowledge — it requires ongoing sensemaking about purpose, pedagogy, and ethics (Tan, 2024).

This creates a central, practitioner-facing dilemma:

How do we use evidence — including AI-generated outputs — without letting it narrow our considerations or effectively assume the role of decision-maker?

Evidence can inform practice — without authorizing action

In education, it is common to talk about “data-informed” or “evidence-based” decisions. That orientation is valuable. However, evidence does not carry its own meaning or implications for action. Turning evidence into action necessarily involves judgment: determining what is relevant, attending to context, understanding assumptions and limitations, and considering potential consequences (Dieterle et al., 2024; Williamson & Piattoeva, 2019).

These interpretive and ethical demands become more pronounced when evidence is generated by algorithmic systems, because such systems filter, prioritize, and format information in ways that can shape what becomes visible and actionable to educators (Bauer & Gill, 2024; Bojic, 2024; Seaver, 2019). Algorithmic outputs often arrive as recommendations or classifications — which may be perceived as more authoritative than warranted, particularly when presented with numerical precision, rankings, or confident language (Dieterle et al., 2024; Wang, 2024). Education governance scholarship further cautions that such “algorithmic decisions” can function like policy decisions in practice, shaping choices while obscuring the assumptions, values, and processes through which they were produced (Wang, 2024; Williamson & Piattoeva, 2019). 

Takeaway: Evidence can inform practice, but it does not authorize action. Interpreting and using evidence, particularly algorithmically generated outputs, requires professional judgment and ethical responsibility.

AI changes decision-making by changing pace and presentation

Research and ethics scholarship on AI in education underscores that trustworthy use is not only a technical matter; it depends on how humans interpret outputs, manage uncertainty, and maintain accountability for decisions made in practice (Dieterle et al., 2024; Holmes et al., 2019). Rather than replacing judgment, AI systems tend to reshape the conditions under which judgment is exercised, affecting when decisions are made, what information is foregrounded, and how authoritative outputs appear. In practice, AI systems and analytics tools can influence educational decision-making in several interrelated ways, including:

1. Speed can compress opportunities for deliberation

When tools generate instant feedback, lesson materials, or predictions, they compress decision cycles. This acceleration can be beneficial, but it can also reduce opportunities and constrain time for reflection, prompting fewer pauses to ask: What is this based on? What does it miss? What are the consequences if it’s wrong? Ethics scholarship cautions that faster cycles of action can unintentionally bypass reflection and ethical review, especially when outputs are treated as ready-to-use inputs to instructional or other professional action, rather than as provisional information to be interpreted and weighted (Dieterle et al., 2024; Holmes et al., 2019).

2. Recommendations can quietly steer choices

Even when AI systems to not mandate a particular decision, recommendations can shape practice by influencing what counts as relevant evidence, which options are foregrounded, and the order in which possibilities are encountered. Rankings, alerts, risk flags, and “suggested next steps” structure attention by making some courses of action more salient than others, often without making the underlying assumptions or tradeoffs explicit (Wang, 2024; Williamson & Piattoeva, 2019; Seaver, 2019). 

Governance and human-AI interaction scholarship describes this as a form of soft influence: systems shape decisions not through formal authority, but by organizing visibility, priority, and perceived relevance within everyday workflows (Bojic, 2024; Bauer & Gill, 2024). Over time, such patterns can normalize certain responses while rendering alternatives less visible, thereby steering judgment even as professional discretion formally remains intact (Dieterle et al., 2024; Wang, 2024).

3. Confidence can be mistaken for validity

Scholarship across datafication, measurement, and AI ethics cautions against conflating clarity, precision, or automation with warranted conclusions (Alon-Barkat & Busuioc, 2023; Kucking et al., 2024; Lewis & Hartong, 2022). Standardization and quantification can produce outputs that appear objective and authoritative, even when they are based on selective representations, simplifying assumptions, or narrow operational definitions of complex educational phenomena (Kinnear et al., 2024; Williamson & Piattoeva, 2019).

In AI-mediated contexts, this risk is amplified. Algorithmic systems translate judgments about learning, risk, or success into scores, categories, or recommendations that can appear decisive because they are numerically precise, consistently formatted, or presented with confident language (Alon-Barkat & Busuioc, 2023; Bauer & Gill, 2024; Seaver, 2019). Yet these outputs reflect design choices about what to measure, how to model it, and which outcomes to prioritize, as well as the limitations and biases of the data on which they are trained (Dieterle et al., 2024; Holmes et al., 2019).

As a result, confidence in the presentation of outputs can exceed confidence in their validity, inviting deference to results that appear settled even when the underlying assumptions, uncertainties, and value judgments remain unexamined.

Takeaway: AI can support educational practice, but it reshapes the conditions under which judgment is exercised. When pace accelerates, attention is structured, and outputs appear authoritative, warranted use depends on disciplined professional interpretation and ethical responsibility.

Professional judgment is an essential condition for responsible use

Across AI ethics and education scholarship, a consistent finding is that responsible use of AI-enabled tools and decision-support systems in education depends not only on system performance, but on human judgment and accountability in how outputs are interpreted and acted upon (Dieterle et al., 2024; Holmes et al., 2019). These systems introduce new forms of evidence that must be evaluated, contextualized, and justified within practice — they require professional judgment.

In educational contexts, professional judgment can refer to the situated, interpretive work educators do to translate evidence, including algorithmically generated outputs, into defensible instructional and other professional actions (Lai et al., 2025). This conception of judgment includes:

  • interpreting evidence in relation to instructional context and students’ linguistic, cultural, and learning needs;
  • weighing competing goals and constraints (e.g., learning, accessibility, equity, relationships, time);
  • attending to what may be missing, oversimplified, or misrepresented in available data; and
  • selecting actions that remain explainable and justifiable to students, families, colleagues, and other interest holders.

Research on data-based decision making in education characterizes teachers’ data use as an interpretive process that integrates analysis with contextual and professional considerations, rather than a simple or automatic pathway from data to action. Data use is often socially situated and can benefit from collaboration and professional dialogue, particularly when evidence is ambiguous or consequential (Lai et al., 2025). This matters because AI-generated outputs are now entering the same interpretive ecosystem as other forms of educational data and evidence — an ecosystem in which information is typically negotiated, contextualized, and made actionable through professional judgment and accountability, rather than taken wholesale and simply acted upon.

Takeaway: Evidence informs practice; professional judgment makes evidence meaningful, appropriate, and defensible.

From “What works?” to “What’s warranted?” 

Much educational improvement work has been organized around the question, “What works?” That question remains important; it reflects a commitment to evidence, effectiveness, and improvement. However, as AI becomes more integrated into educational systems and everyday practice, multiple strands of scholarship suggest that “what works” is no longer sufficient on its own as a guiding question, particularly when automated or algorithmically generated information influences decisions (Holmes et al., 2019; Williamson & Eynon, 2020).

In AI-mediated contexts, “what works” often defaults to optimization-oriented criteria, such as efficiency, prediction accuracy, or average gains, because these are the dimensions that technical systems are most readily designed to measure and improve (Wang, 2024; Williamson & Piattoeva, 2019). Ethics, governance, and sociotechnical scholarship caution that such framing can narrow attention to performance metrics while underexamining consequences related to, for example, equity, agency, trust, and responsibility (Dieterle et al., 2024; Holmes et al., 2019). As a result, decisions may appear successful by technical standards while remaining insufficiently examined in terms of whom they serve, what they amplify, and what risks or tradeoffs they may introduce (Wang, 2024; Williamson & Eynon, 2020).

The following can serve as a practitioner-friendly complement:

What’s warranted here?

A warranted decision is one that remains defensible when multiple considerations are taken into account, including:

  • Purpose: What am I trying to accomplish? Does this align with my instructional, developmental, or ethical aims?
  • Evidence quality: What evidence is being used; how was it gathered or produced; what assumptions, uncertainties, or limitations shape it; and what kinds of claims can — and cannot — be supported by it?
  • Context: What relevant knowledge about learners, communities, instructional conditions, and histories is not represented in the tool or model, and how might that matter for this decision?
  • Equity and ethics: Who is likely to benefit, who may bear risk, and what existing patterns or disparities might be reinforced or amplified?
  • Accountability: Who is responsible for the consequences of this decision, and to whom must it be explainable and justifiable?

This lens aligns with calls across AI ethics and education governance scholarship to treat transparency, responsibility, and human oversight not as add-ons or post-hoc safeguards, but as core conditions for the design, interpretation, and use of AI-enabled systems (Dieterle et al., 2024; Floridi et al., 2018; Holmes et al., 2019; Williamson & Piattoeva, 2019).

Takeaway: “Warranted” reflects a professional standard: not only whether a tool performs well, but whether the resulting action is defensible in context, given its purposes, evidence, and consequences.

Closing reflection

Educators are already doing complex interpretive work — balancing research, tools, and practice with care and professional responsibility. What is changing is not the need for judgment, but the volume, speed, and presentation of “evidence,” particularly as algorithmic systems increasingly package information as ready-to-use recommendations or suggested next steps (Dieterle et al., 2024; Wang, 2024).

The opportunity, then, is not simply to adopt tools or ideas, but to strengthen the conditions under which decisions are made, keeping professional judgment explicit, treating evidence as informative rather than authoritative, and creating space for reflection and shared sensemaking as practices and technologies evolve (Holmes et al., 2019; Tan, 2024). Framed this way, warranted decision-making recognizes guidance as informative while ensuring that relevant considerations of conditions, context, and consequences are weighed responsibly.

Reflection questions

  • What contextual factors most shape whether an evidence-based approach works in practice?
  • When does fidelity to research conflict with responsiveness to context?
  • How do you navigate decisions when evidence is incomplete, mixed, or evolving?

Join this post’s Discussion & Reflection below.

Continue the Inquiry and Reflection

Members receive resources related to each brief and podcast.

Learn more about membership.

About this series 

The edblogcast Foundations Series explores how evidence, ethics, and innovation can evolve together in education’s age of AI. Each brief invites reflection and dialogue across educators, researchers, and innovators — not to prescribe solutions, but to deepen shared understanding of how responsible innovation takes shape in practice.

Together, these explorations aim to bridge perspectives — from systems design to classroom realities — and to support a professional community that learns through inquiry, collaboration, and care.

Join the conversation at edblogcast.com
→ Subscribe, share your perspective, or submit a study.

© edblogcast.com. All rights reserved. 
Content provided for informational purposes and professional reflection. Materials are intended to support inquiry, dialogue, and thoughtful engagement, not to prescribe specific practices or policies.

References

Alon‑Barkat, S., & Busuioc, M. (2023). Human–AI interactions in public sector decision‑making: Automation bias and selective adherence to algorithmic advice. Journal of Public Administration Research and Theory, 33(1), 153–170.

Bauer, J. M., & Gill, J. (2024). Governing algorithmic decision-making: Accountability, transparency, and contestability in public-sector AI. Policy & Internet, 16(1), 3–22. https://doi.org/10.1002/poi3.360

Bojic, I. (2024). Human–AI interaction and soft governance: How algorithmic systems shape choice without coercion. AI & Society, 39(1), 45–58. https://doi.org/10.1007/s00146-023-01670-6

Dieterle, E., Dede, C., & Saxberg, B. (2024). AI-enabled decision-making in education: Ethics, evidence, and professional responsibility. Harvard Graduate School of Education.

Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., … Vayena, E. (2018). AI4People—An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations. Minds and Machines, 28(4), 689–707. https://doi.org/10.1007/s11023-018-9482-5

Holmes, W., Bialik, M., & Fadel, C. (2019). Artificial intelligence in education: Promises and implications for teaching and learning. Center for Curriculum Redesign.

Kinnear, G., Thompson, J., & Gray, C. (2024). Datafication, standardisation, and the limits of quantification in education. Educational Philosophy and Theory, 56(2), 135–148. https://doi.org/10.1080/00131857.2023.2250142

Kucking, D., Perez, C., & Reich, J. (2024). Measuring what matters? Validity risks in AI-supported educational decision systems. Educational Measurement: Issues and Practice, 43(1), 42–55. https://doi.org/10.1111/emip.12546

Lai, M. K., Schildkamp, K., & Poortman, C. L. (2025). Teachers’ professional judgment in data-informed decision-making: Interpretive practices and collaborative sensemaking. Teaching and Teacher Education, 124, 104048. https://doi.org/10.1016/j.tate.2024.104048

Lewis, S., & Hartong, S. (2022). From data to decisions? Exploring the datafication of education governance. Educational Policy, 36(1), 3–26. https://doi.org/10.1177/0895904820983037

Romeo, G., & Conti, A. (2025). Optimization, prediction, and the narrowing of educational choice in AI systems. Learning, Media and Technology, 50(1), 1–15. https://doi.org/10.1080/17439884.2024.2389123

Seaver, N. (2019). Knowing algorithms. Social Studies of Science, 49(3), 412–422. https://doi.org/10.1177/0306312719827617

Tan, E. (2024). Teachers’ sensemaking of artificial intelligence in classroom practice. Computers & Education, 204, 104889. https://doi.org/10.1016/j.compedu.2023.104889

Wang, J. (2024). Algorithmic governance in education: Decision-making, discretion, and responsibility. Educational Administration Quarterly, 60(1), 3–30. https://doi.org/10.1177/0013161X231201234

Williamson, B., & Eynon, R. (2020). Historical threads, missing links, and future directions in AI in education. Learning, Media and Technology, 45(3), 223–235. https://doi.org/10.1080/17439884.2020.1798995

Williamson, B., & Piattoeva, N. (2019). Objectivity as standardization in data-scientific education policy. Learning, Media and Technology, 44(1), 64–76. https://doi.org/10.1080/17439884.2018.1556215

Discussion & Reflection

This space is for shared reflection and inquiry. Rather than quick reactions, we invite thoughtful contributions that extend the questions raised in this brief.

This space is currently unavailable as updates are underway.
It will be restored soon. 
For now, you’re welcome to contact us at team@edblogcast.com