What you must know before AGI arrives | Carnegie Mellon University Po-Shen Loh
Carnegie Mellon math professor Po-Shen Loh argues that as AI surpasses human capability, the critical skills to cultivate are independent thinking, synthesis of novel ideas, language/logic, empathy, and world-simulation — not rote test preparation or AI-assisted shortcuts. He outlines an educational ecosystem designed to develop these skills at scale before AGI makes them obsolete. ---
Key Concepts
| Concept | Definition |
|---|---|
| Synthesis over memorization | The ability to combine new hints/inputs into an original solution is more valuable than recognizing previously seen problems |
| World simulation | Mentally modeling how different people think, what they need, and how situations will play out — the core superpower of entrepreneurs |
| Win-win-win ecosystem | Business/education structures where every participant (learner, expert, supporting professional) gains something meaningful simultaneously |
| Thoughtfulness | Po-Shen's core philosophy — building autonomous human thinking as a deliberate counterforce to AI dependence |
| Bias awareness | Recognizing that every AI system, news source, and person has a built-in agenda; critical thinking is the only defense |
| Customer discovery through immersion | Genuinely entering a sector (teaching sixth grade, riding overnight buses, giving park talks) to find real pain points rather than theorizing |
Notes
AGI and the Shift in Human Capability
- AI solved 4 of 6 International Math Olympiad problems — problems specifically designed to be globally novel
- IMO problems are vetted by national coaches to ensure nothing similar has appeared anywhere before
- AI surpassing this benchmark signals creativity, not just pattern recall, may no longer be uniquely human
- The last remaining uniquely human advantage: we care whether humans continue to exist
Why Schools Must Not Let AI Do Students' Thinking
- LLMs are powerful because they master *language* — the same skill students need to develop independently
- Using AI for writing homework = driving a car instead of running for exercise; zero mental fitness gained
- Students who outsource writing will grow up unable to think logically and will become dependent
- Distinction matters: adults using AI to do a *job task* is fine; students using it to skip *the learning process* is harmful
- Reading, writing, communication, and logic are the foundational skills that build good thinking
The Problem with Modern Test Prep
- Current industry response to math competitions: show students every possible question type until nothing surprises them
- Result: enormous time burden, removal of genuine problem-solving practice, and destruction of the chance to invent
- The world now needs people who can *grade* homework (evaluate, synthesize, judge) — not just do it
- Shift: school used to teach you how to answer; now everyone needs to learn how to ask the right questions
Po-Shen's Educational Ecosystem (EXPII / Live Program)
- Started with free math/science explanation website (XP) — no business model, unsustainable alone
- Moved to recorded video lessons (2019) — generated revenue but lacked human connection
- Key insight: people want live interaction with a knowledgeable *and friendly* expert — the least scalable thing in education
- **The three pain points unified:**
- Solution: high school math experts trained in improv/acting teach middle schoolers — classes look like Twitch streams, taught by smiling math geniuses
- Rule: no high school student is asked to do anything unless Po-Shen can personally explain to their parent why it's the best use of that student's time
- Curriculum subjects — algebra, geometry, combinatorics, number theory — chosen specifically because they train novel thinking, not school recall
- Goal: make students independent as fast as possible so they never need classes again
- Scale estimate: 100,000 US high school students (1%) teaching ~1 million middle schoolers
Philosophy: Contribution Over Competition
- Observed that highly competitive math Olympiad students often graduated depressed and directionless
- Root cause: life philosophy of "prove you're better than others" leads to permanent dissatisfaction
- Alternative: genuine delight in creating value for others — inherently addictive, compounds, and correlates with conventional success
- Money matters not as an end but as a requirement for impact and influence at scale
Skills Needed in an AGI World
- **Empathy and world-modeling**: You cannot solve a problem you cannot visualize through another person's eyes
- **Authentic value creation**: The only reason anyone will want to work with you when AI can do most tasks
- **Critical thinking / bias detection**: AI will sound so complete and reasonable that people will mistake partial stories for full ones
- **Language and logic**: These are how thinking is built; they must not be outsourced
On AI and Bias
- A handful of AI providers (OpenAI, Claude, Gemini, DeepSeek) representing effectively 4–5 worldviews vs. 7.5 billion human perspectives
- Po-Shen's personal practice: follows right-leaning content on X, left-leaning on Facebook — reads both daily to triangulate
- AI's fluency and apparent completeness makes it *more* dangerous as a source of one-sided narratives, not less
- Everyone has an agenda (including Po-Shen — he states his openly: build a more thoughtful world)
Customer Discovery and Rural America
- Traveled city to city giving free math talks in public parks; 50–100 attendees regularly
- Direct interaction with thousands of parents and students revealed real pain points that led to the ecosystem idea
- Goes into schools personally — including sixth-grade classrooms in high-poverty areas — to stay grounded
- Observed a rural, low-income fourth-grade classroom: kids had no phones or internet, invented their own games for fun → demonstrated exceptional curiosity, creativity, and collaborative thinking
- Conclusion: rural America holds enormous untapped intellectual potential waiting for access
Actionable Takeaways
- **Do your own writing** — use AI to research and inform your thinking, not to produce the output; your ability to reason depends on it
- **Practice novel problems** — seek questions you've never seen before and sit with discomfort before looking for hints; that discomfort is the workout
- **Consume multiple biased sources intentionally** — follow at least two ideologically opposed feeds and treat both as incomplete rather than either as authoritative
- **Use AI to build your mental model, not replace it** — query AI for context and links, then draw your own conclusions (as Po-Shen did with the Nashville singer)
- **Reframe your life philosophy** from outcompeting others to creating genuine value — it compounds, it's sustainable, and it makes you a desirable collaborator in an AGI economy
- **Immerse yourself in the sector you want to impact** — ride the bus, go into classrooms, attend the parks; pain points only reveal themselves through direct contact
Quotes Worth Keeping
The only unique thing about human intelligence is that we hopefully care that humans still exist.
Using AI to do your writing homework in school is like saying, 'I'm not going to run a mile for exercise. I'm going to drive my car one mile for exercise.'
People used to go to school to learn how to do the homework and do the exams. Today, everyone needs to learn how to grade the homework.
Our goal is to make it so that as fast as possible, you don't need any classes from anyone ever again.
You cannot create value if you don't interact with people. You cannot just theoretically think about the value.
The AI is going to be so good at looking complete that you may think you have the entire story on a controversial situation, but you don't.
I wasn't using the AI to write the report for me. I was using AI to make myself better at that particular goal.