What comes to mind when you see the above image? It’s from a CS research paper in which a computer used millions of images to automatically remove the roof from the foreground and fill in the image so that you would never suspect there was a roof in the first place.
My initial reaction to seeing it during the first computer vision class of the year was of course ‘wow. That’s really freaking impressive. Go computer! (And the people who programmed it.)’
My attitude towards the first day of school has gone, over the course of 20 consecutive years, from “Oh man! Back to wonderland!” to “Oh, it’s Wednesday, guess I should go to class or something.” It’s a stark reminder of how familiarity can make even the most extraordinary experiences seem normal – always something to keep in mind and fight against.
The second thing that came to mind though, was ‘wow. Modern engineering is built only for mediocrastan. That’s terrifying. What can I do about that?’
Mediocrastan is a concept that captures how the world normally works the vast majority of the time. In this world, everything follows Gaussian distributions, so even the random things are predictable through certain tools. Most of our engineering practices and technology depend on that assumption, but that isn’t how the world always works.
Now, nobody gets hurt if instead of there being a bunch of boats behind that roof, there was actually Godzilla rising from the depths, or something equally unlikely. However, that’s not the case when a wave that, in mediocrastan, should only occur once every 10,000 years or so hits a bridge that was only designed for the 1,000-year waves.
I’m not advocating that we overengineer everything simply to avoid these rare events. It’s important that people keep in mind these assumptions that engineers make and that technology will fail in extreme circumstances (whatever extreme is for that technology.) I’m interested in how we can change engineering practices to better take into account these unknown unknowns and don’t have a particularly good solution right now.