The Earth.Works

If AI can produce a correct answer instantly, testing for correct answers is pointless. What's left to assess is how someone got there and whether they'd know if they were wrong. To achieve this the process is to : 

​

1. Pose an ambiguous, real problem (not a textbook question)

2. Have the learner work through it with AI as a sparring partner, not an oracle

3. Make the learner defend their reasoning, choices, and sources

4. Log the process, not just the output, as the evidence of learning

​

What it actually assesses (replacing exams)

• Discernment — show them an AI-generated or flawed answer and ask them to find what's wrong with it. Grading the critique, not the recall.

• Synthesis — connect two unrelated domains to solve a problem neither covers alone.

• Making — actual artifacts: a composition, a performance recording, a piece of writing, a working prototype. Evaluated by process trail + peer/expert review, not multiple choice.

• Self-direction — did they set their own sub-goals, notice when they were stuck, and adapt? This is trackable behaviorally.

• Collaboration — real multiplayer tasks where the AI can observe how they negotiate, not just what they produce solo.

​

AI's role is to be adversarial/Socratic, not helpful-by-default

​

• Withhold the answer and ask "what would you check first?"

• Occasionally give a subtly wrong answer on purpose, to train skepticism

• Push back on weak reasoning instead of validating it

• Make the learner teach the concept back to the AI as a completion condition 

​

Accreditation issue

If there's no exam, something still has to signal competence to the world. The  app builds a longitudinal portfolio/reasoning-trail that becomes the credential itself — verifiable process, not a score.