The End of Assessment? Disruption and Transformation in the Age of AI

Synthesis: Generative AI is collapsing the long-standing boundaries between learner, test, and assessor — an epistemic disruption in what counts as evidence of learning. These systems can be test-takers, test-makers, test-raters, and analysts/reporters simultaneously. Rather than merely adding a tool, AI dissolves the field's traditional representational-measurement paradigm. The authors argue assessment must transform from static measurement into an evidence ecosystem centered on meaning-making, justice, and AI literacy, led by a "cyborg" assessment professional fluent in both disciplinary assessment expertise and AI.

Key Findings

• AI as test-taker, test-maker, test-rater, and analyst. LLMs (ChatGPT, Claude, Gemini) can complete the same tasks used to evaluate students (essays, quantitative problems, code, data analysis) and have excelled on the bar, medical licensing, and SAT. They can also generate items, rubrics, and validity documentation in seconds; score written work with speed and consistency; and synthesize institutional findings for accreditation — redistributing the interpretive authority once reserved for assessment professionals.

• Validity is thrown into question. The authors anchor on Messick's (1989) integrative-judgment definition of validity. When AI designs the very measures from which evidence is derived, the interpretive chain becomes opaque: are we assessing a construct grounded in human theory, or one generated from statistical patterns in training data?

• Justice is a validity question. Validity is inseparable from value implications. AI can encode dominant linguistic patterns and rhetorical styles as quality markers — e.g., privileging standardized academic English and misclassifying equally sophisticated reasoning from multilingual or culturally marginalized learners — and AI image generators represent White individuals more accurately than people of color (Yang, 2025). These biases are not just ethical concerns but direct threats to validity, connecting to Bias Mitigation and Equity.

• AI literacy becomes assessment literacy. Where traditional assessment literacy involved rubrics, reliability, and validity, AI literacy extends this to prompting, data interpretation, and understanding how generative systems produce evidence. Without it, practitioners, faculty, and students risk becoming passive consumers of algorithmic judgments (AI Literacy).

• The "cyborg" assessment professional. The field's future is a hybrid worker who combines high AI literacy with high disciplinary/assessment expertise — designing prompts with intention, creating human-machine partnerships, auditing outputs for bias, shaping data pipelines, and stewarding institutional meaning-making. A two-axis framework (AI literacy × disciplinary/content expertise) yields four quadrants, and the authors outline a pathway from traditional assessment specialist (low AI literacy) through "AI Explorer" to the cyborg/Futurist Assessment Professional.

• A national Community of Practice as a case in transition. The Generative AI in Assessment CoP, launched by the Assessment Institute in 2025 with ~80 members (plus ~130 affiliates), reflects themes of urgency, desire, variation, constraint, and community — evidence that the field is actively negotiating new professional identities around AI-mediated evidence.

Assessment as meaning-making rather than measurement

A central transformation claim is that assessment shifts from extracting scores to constructing meaning from complex, hybridized systems. AI can surface relationships too complex for humans to detect alone, but it cannot generate the "why" — the moral and contextual lens through which meaning is co-constructed. The assessment professional becomes "sensemaker-in-chief," a translator between algorithmic data and human understanding, and dashboards become evolving evidence narratives rather than snapshots.

What this means for practice

• Instructors. Redesign what is assessed, not just how: shift from testing isolated knowledge to assessing how effectively students work with AI to interpret, apply, and evaluate knowledge, and treat AI Literacy as a foundational learning outcome.
• Instructors. Reconceive your own role as auditor and translator of algorithmic evidence — verifying, interpreting, and explaining AI-augmented results and teaching students to question model reasoning — rather than guarding the purity of measurement.
• Administrators. Treat AI equity audits as validity work: audit training data, prompts, and results for representational bias instead of treating dominant norms such as standardized academic English as neutral quality standards.
• Administrators. Build the "cyborg" pathway deliberately through structured professional development in prompting, algorithmic auditing, and data stewardship, because the two-axis AI-literacy x disciplinary-expertise framework places most current staff below the fluent quadrant.
• Administrators. Pair automated scoring with human judgment, keeping people in the interpretive loop over the unresolved questions of bias, fairness, and transparency.

Limitations

• This is a conceptual stance, not an empirical study: the authors state they do not offer settled accounts of AI's impact on assessment and do not presume a single trajectory, so the transformation claims are argued rather than tested.
• The AI-literacy x expertise framework and the "cyborg" pathway are presented as a conceptual space rather than a developmental sequence, with positions labeled illustrative and non-hierarchical; movement through them is not demonstrated.
• The one concrete case, the Generative AI in Assessment Community of Practice, is a self-selected closed cohort of roughly 80 members (plus about 130 affiliates) in its first year, which the authors describe as "still evolving" — it illustrates a field in transition rather than reporting outcomes.
• The bias argument rests on cited examples rather than the authors' own data, such as the finding they attribute to Yang (2025) that AI image generators represent White individuals more accurately than people of color.

Citation

Hathcoat, J. D., Slotnick, R., & Miller, W. (2026). The End of Assessment? Disruption and Transformation in the Age of AI. Research & Practice in Assessment, 21(3).