Twelve Provocations for the Future of AI and Assessment

Summary for Educators

By Allison Dulin Salisbury and Karen Alpuche | Humanist | July 8, 2026

Artificial intelligence is forcing educators to confront a fundamental question: If AI can produce the final product, what exactly should schools be assessing?

Salisbury and Alpuche argue that schools should resist making their primary AI-assessment goals the detection of cheating, the speeding up of grading, or the creation of assignments that students cannot complete with AI. Instead, educators have an opportunity to redesign assessment around the capabilities that remain distinctly important in an AI-rich world: judgment, reasoning, creativity, agency, collaboration, contextual intelligence, and the ability to use AI wisely.

THE BIG IDEA

For generations, schools have often treated assessment as something that occurs after learning: teach a unit, give a test, assign a grade, and move on.

AI opens the possibility of something very different.

Assessment could increasingly become part of the learning process itself—capturing how students formulate questions, revise ideas, solve problems, respond to feedback, explain their reasoning, collaborate with others, and decide when AI should or should not be used.

The authors describe a future in which assessment operates more like coaching than surveillance. Rather than concentrating primarily on whether a student's final answer is correct, schools could collect evidence about the intellectual process that produced it.

That distinction becomes critical when AI can generate an impressive essay, solve a mathematics problem, summarize a chapter, create computer code, or design a presentation in seconds.

The question shifts from “What did the student produce?” to “What can the student actually understand, explain, defend, apply, and do?”

KEY TAKEAWAYS

• Stop designing assessment primarily around catching AI use. Cheating and academic integrity remain legitimate concerns, but they are too narrow a starting point for redesigning assessment.

• Measure the process, not merely the product. Drafts, conversations, explanations, revisions, decision-making, reflection, and problem-solving may provide stronger evidence of learning than a polished final submission.

• Assessment should occur during learning. The report envisions assessment functioning more like a “heart-rate monitor” than an autopsy—giving teachers useful information while they can still change instruction.

• Knowledge and durable skills can be assessed together. Students should demonstrate foundational knowledge while also using communication, collaboration, reasoning, creativity, and problem-solving rather than taking separate tests for every competency.

• Standardization does not necessarily require identical tests. One of the report's provocations suggests that students may eventually complete different, contextualized tasks while being evaluated against common standards or rubrics.

• AI fluency itself will become something worth assessing. Students need to learn when AI adds value, when it introduces errors or bias, and when independent human thinking is essential.

WHY IT MATTERS

Perhaps the report's most important message for school leaders is that assessment drives behavior.

What schools measure tells teachers what to teach, students what to concentrate on, parents what schools value, and institutions what constitutes achievement. Salisbury argues that assessment therefore becomes a “linchpin” for redesigning education for an AI-shaped future.

If schools continue assessing primarily what AI can easily reproduce, they may unintentionally encourage students to outsource precisely the thinking educators want them to develop.

But eliminating AI from assessment entirely presents another problem: students will graduate into colleges and workplaces where AI will increasingly be part of everyday work.

Schools therefore need evidence of both independent human capability and effective human-AI collaboration.

LEADERSHIP ACTIONS

Audit current assessments. Ask teachers to identify assignments where a polished final product is being treated as evidence of learning even though AI could produce much of that product.

Increase “show your thinking” assessment. Include oral explanations, conferences, drafts, annotations, demonstrations, defenses, reflections, and questions about why students made particular choices.

Assess both with AI and without AI. Students should sometimes demonstrate what they know independently and at other times demonstrate how effectively they can extend their capabilities with technology.

Make formative assessment more immediate. Explore ways teachers can gather evidence from everyday student work rather than waiting for major tests.

Develop common expectations for AI fluency. Schools should define what competent, ethical, transparent AI use looks like at different grade levels.

Protect teacher judgment. AI may help collect and interpret evidence, but consequential educational decisions still require professional context and human judgment.

LEADER REFLECTION

If AI can increasingly produce the answers we traditionally ask students to produce, what evidence would convince us that a student genuinely understands something?

That may become one of the defining assessment questions of the next decade.

Original Article

“Twelve Provocations for the Future of AI and Assessment” by Allison Salisbury and Karen Alpuche, August 7, 2026 in Humanist

------------------------------

Prepared with the assistance of AI software OpenAI. (2026). ChatGPT (5.2) [Large language model]. https://chat.openai.com