> ## Content Index
> Fetch the complete content index at: https://ai-public-health.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Superhuman vs. Non-Human
- URL: https://ai-public-health.com/forgetting-makes-the-robot/
- Published: 2026-09-21T12:39:00.000Z
- Updated: 2026-09-21T12:39:00.000Z
- Author: Min Wu
- Tags: AI in Healthcare, Machine Unlearning, Health Informatics, Robotics Ethics, Responsible AI

***Case of the Week, Min Wu, PhD · ai-public-health.com***

> **A note before we start.** Starting with this post, essays will move from weekly to **biweekly** through the fall and spring semesters — Mondays and Wednesdays are my teaching days, and the rest of the week fills in around them. Weekly returns each summer. See you every other Monday.

*Almost thirty years ago I started learning computer literacy, and by the time I graduated in the early 2000s, health informatics was living through what I still think of as the database era. My own research career started there, in biomedical database projects, where the entire point of the system was that it never lost anything. A record went in once and came back out exactly the same, indefinitely. A good memory meant a perfect memory — instant, exhaustive, lossless.*

*I still catch that standard running underneath my own instincts. But the AI systems I work with now don't earn trust by remembering everything. They earn it by knowing what to forget. Twenty-some years later, I find myself teaching the opposite of the lesson I learned as a graduate student, and I still haven't fully gotten used to it.*

**Key Takeaways**

- AI strengths split into two kinds: superhuman (faster, tireless, better at a familiar job) and non-human (capabilities no human role ever had).
- Min Wu's clearest example is selective forgetting: once treated as database corruption, now a design requirement for staying accurate rather than confidently wrong.
- The same robot built from superhuman strength is a tireless nurse; built from non-human strength, it notices what no caregiving script was designed to catch.

## Superhuman Still Reports to the Database

There's a category of AI strength that's genuinely impressive and, if you look closely, isn't actually new. Massive parallel processing. Memory with no gaps. Operation that never gets tired, never has an off day. Call it superhuman, because compared to a person doing the same job, it plainly is. But notice what job it's still doing: the database's job. Hold everything, retrieve on demand, never let anything slip. A robot built entirely out of superhuman strengths is a tireless, perfect-memory version of a caregiver — a faster librarian, not a different kind of thing. It looks like a leap forward because it's so far past what a human can do. It's still standing on the exact spot the database era staked out thirty years ago.

## Non-Human Breaks the Job Description

The strengths that don't have a human-role analog at all sit in a different category, and the clearest example is one my own field is still adjusting to: forgetting. In the database era, forgetting was corruption — a backup failure, data loss, the thing every system was built to prevent. In the AI I work with now, selective forgetting is a design requirement. A model that never forgets outdated transmission patterns, retracted safety reports, or the biases baked into whatever it was first trained on doesn't stay useful — it stays wrong, confidently. Teaching a system when to let go of what it knows isn't a workaround for a limitation. It's a capability the database era had no use for and no name for, because nothing in that era needed it. Add in the other non-human strengths — finding a correlation across sensor streams no human sense channel combines, noticing a pattern in a space too noisy or too large for any person to scan by hand — and you get capabilities that don't scale up a human role. They don't have one to scale up from.

## Same Robot, Two Starting Points

Take a home-care robot built around health tracking — the Physiologist persona, the one whose job is watching a person's condition over time. Built from superhuman strength alone, it does what a home health aide's daily check-in does, just tirelessly and with perfect recall: same questions, same schedule, no forgotten follow-up, ever. Genuinely useful. Still a nurse, scaled.

Built from non-human strength, the same robot does something a check-in was never designed to do: it continuously fuses gait, sleep, appetite, and voice tone against a rolling window it's actively allowed to forget the stale parts of, and it surfaces a correlation no caregiving script would have known to look for — because no human caregiving role was ever built to hold that many threads at once, or to let go of the ones that stopped mattering. Same hardware, same persona family. One version is a better version of something that already existed. The other is something that didn't.

## The Observer's Insight

I learned my trade in an era that measured a good memory by what it refused to lose. The robots worth building now are, in part, measured by what they're willing to let go of — and by what they can notice that no human role was ever asked to notice at all. Superhuman is safe to build because it's still explainable in old terms: a better version of a job we already understand. That's exactly why it isn't where the interesting work is. Non-human is where the real newness lives — the part with no job description to borrow instincts from, and the part I'm still teaching myself to trust.

***Forgetting makes it new.***

*Where have you caught an old-era instinct still running underneath a new-era tool, without noticing until later? Reply or leave a comment — I'd love to hear.*

*As a member of the Springer Nature Author Affiliate Program, I may earn a small commission from purchases made through this affiliated link to my book. Support the author by checking out my textbook, Artificial Intelligence in Public Health:* [https://tidd.ly/4mH9389](https://tidd.ly/4mH9389?ref=ai-public-health.com)