How to raise the bar for rigor

One of the most feared consequences in the age of AI is the proliferation of slop. AI is infinitely more “productive” than humans, needing no sleep, no breaks, can parallelize infinitely, and thus has the enormous capability to produce tremendous amounts of slop output. What can help us raise children who have the ability to approach AI with proper skepticism? What can help them understand when skepticism is needed, and when to forgo it in favor of speed instead?

The first skill is more fundamental and should be developed first. The second skill depends on judgment, experience, and intuition, which should come after the foundational skill is developed.

The answer to the first question comes back to verifiable learning. An important byproduct of verifiable learning is that there is a bar for rigor that the person has encountered and experienced before. Math, as we discussed in the previous article, has a hierarchy of difficulty – AMC, AIME, USAMO, MOP, IMO. One advantage of funding an IMO founder is that everyone knows they have encountered a level of difficulty few have before, and can set that bar again.

When an IMO person and a non-rigorous person ask a model the same query, the IMO person knows the first answer received is highly unlikely to be complete. Proof-based math is not probabilistic, it must be complete over all cases. Any counterexample invalidates the whole proof. The non-rigorous person has no idea of this concept, and doesn’t understand what completeness even looks like. For example, when it comes to a systems engineering question with distribution and retries and outages, the rigorous person will prove all corner cases will not produce worst-case outcomes, like double-counting or irreversible transactions. The non-rigorous person will trust AI output that covers 90% of cases, but produces catastrophic tail results.

Math and verifiable domains that can infinitely ratchet in difficulty help children understand what rigor is. Another method to instill rigor in children is to introduce early consequences. What better way is there to ensure children check all output than if they are fully responsible for all consequences coming from that output?

If an eight year old is responsible for writing a family’s holiday card, they better hone those writing skills. A ten year old with an online business will be responsible for real allowance money gained or lost. A twelve year old planning their own birthday party is responsible for whether the pizza is ordered ahead of time from a good pizzeria, or whether it’s Domino’s rush-delivered late. Real consequences, taught early, viscerally show children the importance of being fully liable for the things they produce. Lack of rigor, left too late, is what results in slop.