# DeepMind CEO Demis Hassabis says learning how to learn is the next generation's most needed skill

> With AI changing work, schools must teach learning how to learn: metacognition, retrieval, spacing, AI as partner. Update assessment, build teacher workflows, and show equity gains.

- Type: AI news
- Page: https://completeaitraining.com/news/deepmind-ceo-demis-hassabis-says-learning-how-to-learn-is/
- Published: 2025-09-15
- Updated: 2025-09-15
- Author: Jeroen Erne
- Jobs: Education
- Reading time: 3 min

### Learning How to Learn: What Educators Need to Prioritize Next

At the Odeon of Herodes Atticus in Athens, Demis Hassabis, CEO of Google DeepMind and 2024 Nobel laureate, made a clear call to education leaders: the core skill students need is learning how to learn. With AI changing tools and job demands week by week, content expires quickly. Meta-skills endure.

Hassabis suggested artificial general intelligence could arrive within a decade. Big gains are likely-and so are risks. Greek Prime Minister Kyriakos Mitsotakis added a warning: if people don't see personal benefit and wealth pools around a few companies, expect backlash.

#### What "learning how to learn" looks like in practice

- **Metacognition:** Teach students to plan, monitor, and adjust how they study. Short weekly reflections: What worked? What didn't? What's the next move?
- **Retrieval over review:** Use low-stakes quizzes, oral checks, and one-page recalls. Retrieval strengthens memory far more than re-reading.
- **Spaced and interleaved practice:** Spread practice and mix problem types. Build long-term retention and transfer.
- **Deliberate practice:** Break skills into sub-skills. Give specific feedback loops with clear criteria of success.
- **Learning with AI:** Treat AI as a thinking partner, not an answer machine. Require students to show prompts, iterations, and reasoning traces.
- **Transfer across domains:** Have students apply one method (e.g., Fermi estimates, causal maps) across science, history, and design.

#### Program moves you can implement this year

- **AI across the curriculum:** Add one authentic AI use case per course (summarization critique in literature, code review in CS, data cleanup in science).
- **Prompting and critique:** Teach prompt patterns (role, constraints, examples) and require students to compare outputs, cite sources, and flag hallucinations.
- **Assessment upgrade:** More oral defenses, live problem-solving, version histories, and portfolios. Grade process and evidence, not just final answers.
- **Micro-credentials:** Offer short badges in data literacy, AI safety, prompt design, and automation. Stack them into course credit.
- **Teacher PD sprints:** Monthly 60-minute labs: one tool, one workflow, one shared template. Produce a classroom-ready artifact every session.
- **Academic integrity policy:** Define permitted use, required disclosures, and consequences. Provide examples of good and poor AI use.
- **Tool audits:** Review privacy, bias, data retention, and cost before adoption. Prefer tools with clear FERPA/GDPR language.

#### Equity and trust: show personal benefit

Mitsotakis cautioned that concentrated gains fuel inequality. Schools can counter that by showing concrete wins for every learner. Track and publish outcomes: time saved, feedback quality, and student growth across groups-not just top performers.

- Offer device access and offline options. Pair AI use with foundational skill building.
- Run family workshops on safe, effective AI use. Explain data practices in plain language.
- Use AI to extend feedback and tutoring time, then monitor impact on struggling students.

#### Why this matters now

Hassabis' point was blunt: the only certainty is change. Curriculum locked to a static content list falls behind. Programs built on meta-skills, authentic tasks, and continuous upskilling stay relevant.

#### Context on AI and science progress

Hassabis' team helped crack protein folding prediction, accelerating biomedical research. For background on that scientific step, see DeepMind's overview of AlphaFold [here](https://deepmind.google/research/highlighted-research/alphafold?ref=completeaitraining.com). For policy framing, the OECD's work on AI and education offers a useful reference [here](https://www.oecd.org/education/ai-in-education/?ref=completeaitraining.com).

#### Quick start: templates and courses

- Create a "learning audit" template: goals, strategies tried, retrieval schedule, AI prompts used, and evidence of improvement. Use it weekly.
- Build a shared prompt library with examples for lesson planning, rubric drafting, and feedback scripts.

If you need structured upskilling for faculty or by role, explore [courses by job](https://completeaitraining.com/courses-by-job/) and the [latest AI courses](https://completeaitraining.com/latest-ai-courses/).

#### Key takeaways for education leaders

- Prioritize meta-skills over static content: metacognition, retrieval, spacing, transfer.
- Normalize AI as a partner with documented process, citations, and critique.
- Shift assessment to demonstrations of thinking and authentic work.
- Invest in teacher workflows that save time and improve feedback quality.
- Measure equity impacts and communicate clear benefits to students and families.

Change is coming either way. Build learners who can keep learning.

---

This is the Markdown form of https://completeaitraining.com/news/deepmind-ceo-demis-hassabis-says-learning-how-to-learn-is/. The whole site is mapped at https://completeaitraining.com/llms.txt.
