Video summary

Diseñar la enseñanza en contextos mediados por IA - Natalia Corvalán

Main summary

Key takeaways

Educational

Main ideas and lessons

  • Context/problem driving the talk:

    • Teachers and students increasingly use Generative AI to produce/submit assignments (“copied and pasted exactly”).
    • This creates tension: AI is perceived as helpful/innovative, but classroom use often becomes uncritical, product-focused, and can undermine teachers’ and students’ own voices.

  • Core shift proposed:

    • Move from AI as “mirage” (a promise that it will solve problems, save time, personalize learning automatically)
    • to AI as “mirror” (a tool/agent that reflects back what teachers are doing—sometimes uncomfortably—so they can revise their thinking and practice).

  • Critique of common AI integration patterns:

    • AI is often used to:
      • verify whether it gives “correct answers” (checking accuracy instead of learning/reflecting),
      • generate content quickly (efficiency → quality assumptions),
      • follow hype-driven “hackneyed” narratives (e.g., “AI will save time/personalize everything”),
      • replicate traditional classroom sequences using step-by-step instructions without true co-creation.
    • “Prompt engineering” alone is insufficient: pedagogical training is needed to know what counts as meaningful concepts/criteria, otherwise AI and the teacher define things differently.

  • Central methodology introduced: “Generative twins” (teacher-designed agents)

    • Instead of generic chatbots, design an AI system with a defined role to act as a pedagogical partner/tutor/counterpoint in a specific context and activity (not as a generic assistant).
    • Purpose is not to replace the teacher or automate correction/rubrics, but to:
      • support thinking,
      • reveal blind spots,
      • enable richer teacher/students’ dialogue with sources and criteria.

  • Emphasis on design over delegation:

    • There is no universal magic prompt.
    • The system must be iteratively tuned through teacher/designer experimentation.
    • A good design process requires:
      • selecting sources/knowledge base,
      • defining the agent’s role and constraints,
      • requiring the agent to behave in ways that foster reflection and critique rather than instant “answers.”

  • Digital/AI literacies and mediation:

    • Teachers should not only teach about AI, but also use AI as a mediated learning environment, modeling more than one type of AI use (e.g., beyond “generic GPT chat”).
    • Tools that connect to sources and validation (e.g., research assistants, citation tools) help move toward more critical thinking.

  • Bias and structural/epistemological concerns:

    • Even “source-based” systems embed bias:
      • in what the teacher selects/loads,
      • and in the model’s algorithmic selection/behavior (black-box criteria).
    • Instead of “fighting bias,” the talk suggests working with bias:
      • contrast outputs,
      • discuss why different parts are selected/understood,
      • use discrepancies as learning opportunities.

  • Boundaries and appropriate use:

    • AI shouldn’t be used all the time; sometimes activities need unplugged time and deeper human engagement.
    • “Agentic” AI use may not be appropriate for everyone at every stage—especially if learners are confused about what AI is and is not.

  • What to measure by outcomes:

    • The aim is not “better answers” alone; it’s to cultivate:
      • critical questioning,
      • metacognition,
      • reflective revision of prompts/categories,
      • movement toward more complex inquiry.

Methodology / instructions (detailed)

1) Diagnose classroom/teaching use cases (starting question)

  • Identify how AI is currently being used in practice, especially in:
    • assessment completion (solving exams/assignments),
    • copying/pasting finished outputs,
    • teacher/student reliance on AI correctness.
  • Ask: Is the AI being used to learn and reflect, or to replace the teacher/student’s own voice and thinking?

2) Decide whether an AI “generative twin” is needed (and for what)

  • Define a real pedagogical need (not “because hype” or “because it saves time”).
  • Decide:
    • whether you need a generic chatbot or a role-based pedagogical agent,
    • what the agent will do in a specific activity context.

3) Design a “generative twin” (teacher-designed pedagogical partner)

  • Use free/accessible tools (examples mentioned):
    • NotebookLM (first suggested; allows selecting a knowledge base),
    • Gemini (including its “GEMS” functionality at the time of experimentation),
    • PW (with colleague Nataline Calvo mentioned; also a possibility explored).
  • Build the twin as a non-human agent with:
    • a defined role (tutor, pedagogical partner, counterpoint, etc.),
    • constraints that prevent it from acting like a generic chatbot.
  • Provide the twin with:
    • sources (curriculum, program materials, selected texts/links/files),
    • a style/voice derived from the teacher’s positioning (example instructions about the teacher’s pedagogical stance).

4) Iteratively prompt and tune (no one-shot solution)

  • Use an iterative process:
    • test how it responds,
    • identify shortcomings,
    • refine role definitions and requirements.
  • Apply expansion beyond basic prompting, e.g.:
    • “Act like a good tutor/person”
    • then continue specifying: “act as a pedagogical partner”
    • and define:
      • what “tutor” means,
      • the scope of support (what it can/can’t do),
      • the target learner level and learning context.

5) Avoid replacing the teacher’s function

  • Do not aim for:
    • “teacher replacement,”
    • automated emailing,
    • fully delegating instruction sequencing,
    • automated correction/rubric automation.
  • Use the twin to help think rather than produce ready-made “final answers.”

6) Ensure it’s source-grounded and not “generic”

  • Distinguish:
    • generic corporate chatbots (large, pre-designed; limited traceability; reactive behavior; no defined pedagogical role),
    • from teacher-designed twins (defined role; teacher-selected sources; more traceability via provided materials).
  • In NotebookLM specifically:
    • load content you want it to use (files/links/videos/documents),
    • so outputs are grounded in your chosen materials.

7) Run classroom/teacher-training activities that model alternative AI use

  • Model “mirror” behavior:
    • the twin returns counterpoints, questions, and alternative interpretations,
    • students/teachers compare responses against the teacher’s criteria and sources.
  • Turn discrepancies into learning:
    • If two students use similar tools/prompts but get different answers, investigate:
      • why outputs differ (digital footprint/cookies/subscriptions were hypothesized),
      • and use that to stop encouraging copy-paste reliance.

8) Use the “mirror” outcome loop (metacognition + revision)

  • After interaction:
    • revisit the defined categories/prompts,
    • refine the definitions based on what the twin reveals,
    • optionally add new sources/authors that better fit the teacher’s pedagogical intent.
  • Evaluate success by whether it:
    • stimulates reflection,
    • uncovers blind spots,
    • strengthens critical questioning and source validation,
    • preserves the teacher’s identity/voice rather than replacing it.

Tools / frameworks mentioned (as examples)

Generative twins environment examples

  • NotebookLM
  • Gemini (GEMS mentioned)
  • PW

Other AI-assisted research/thinking tools referenced

  • Elicit
  • Consensus
  • ISIT (mentioned in relation to source validation/comparison, as described in subtitles)

Institutional/resource documents referenced

  • UNESCO guide: “Use of Generative Artificial Intelligence in Education and Research” (recommended read)
  • Spanish Ministry of Education (AI teaching/use dimension)
  • INTEF document with recommendations on AI use in education

Speakers / sources featured

Speakers (people)

  • Natalia Corbalán / Natalia Corbalán (primary presenter)
  • Andrea (moderator/interviewer; asks questions and introduces speaker)
  • Mariana Ferrar / Mariana Serrareli / Mariana Serrareli (collaborator mentioned; co-author of works referenced; initially proposed the “generative twin” framing in the team’s work—names appear with some subtitle variation)
  • Karina Leon (co-authors a referenced text with Mariana)
  • Valeria Odeti (mentioned as an investigation contributor)
  • Nataline Calvo (co-mentioned colleague for experimenting with tools)
  • Flavia Costa (mentioned as reference for socio-technical literacy framing)
  • Miriam Cap (mentioned as supporting “focus on questions” idea)
  • Tomás Balmaceda (mentioned regarding bias: “don’t fight bias; fill it with more biases”)
  • Molic Molic (mentioned as author of an inspiring 2023 text about language models fulfilling different roles)
  • Emily Vender (mentioned in analogy: “stochastic parrots” framing of LLM behavior)

Institutions / organizations (sources)

  • UNESCO
  • Spanish Ministry of Education
  • INTEF
  • Aula Abierta (context/host of the open class)

Original video