The Assessment Crisis

The Assessment Crisis

Traditional academic assessment assumes a simple model: students learn skills, demonstrate those skills in assessments, receive credit for demonstrated competence.

AI breaks this model. If a student can use AI to write an essay, solve a problem, or produce a project, what does the assessment measure?

The Immediate Problem

Educators face practical dilemmas:

  • Can’t tell if work is human or AI-produced
  • Banning AI is hard to enforce
  • Allowing AI makes assessment meaningless
  • Proctored exams don’t match real-world conditions
  • Oral exams don’t scale

Every assessment format has a failure mode in the AI era.

What Are We Measuring?

Traditional assessments assumed they measured:

  • Knowledge acquisition
  • Skill development
  • Ability to apply learning
  • Readiness for professional work

With AI, these same assessments might measure:

  • Ability to use AI effectively
  • Ability to prompt well
  • Willingness to follow rules
  • Luck in avoiding detection

These aren’t worthless, but they’re not what the assessment was designed to measure.

The Deeper Question

The crisis forces a question that was always present but could be avoided: what is the actual educational goal?

Possible answers:

  • Knowledge in heads: Students should personally know things. AI undermines this if students outsource.
  • Capability to produce: Students should be able to produce work. AI-assisted production might satisfy this.
  • Process and understanding: Students should understand how to approach problems. AI might short-circuit this.
  • Judgment and taste: Students should know quality when they see it. Maybe AI helps with this.
  • Credential signaling: The degree signals something to employers. What?

Different answers lead to different assessment approaches.

Possible Responses

Assess what AI can’t do: Design assessments around tasks AI fails at — but these change as AI improves.

Assess the process: Observe how students work, not just what they produce. But this is labor-intensive.

Assess AI use: Grade on how well students leverage AI. But this assumes AI use is the goal.

Assess in controlled environments: Proctored, no-device exams. But these don’t match real-world conditions.

Abandon assessment: Trust intrinsic motivation and self-directed learning. But this conflicts with credentialing functions.

Redefine competence: Decide that AI-augmented capability is competence. But what are students actually learning?

The Authenticity Problem

Any assessment asks students to demonstrate something. If that “something” can be delegated to AI, two possibilities:

  1. The skill being assessed isn’t valuable (if it can be delegated)
  2. The assessment isn’t measuring what we actually value

Both are uncomfortable. Either curricula need dramatic revision, or assessments need fundamental redesign.

Implications

  • Assessment practices need to evolve, probably fundamentally
  • The goals of education need explicit articulation
  • Different fields may have different answers
  • Accreditation and credential value are at stake

Open Questions

  • What should education measure if AI changes what humans need to know?
  • Can assessment keep up with AI capability growth?
  • Is there a stable foundation for assessment, or will it need constant revision?
  • How do employers adjust expectations for AI-era graduates?

See Also