Or: Mind the Cliff

Suppose you are a clumsy fellow, liable to knock things over. You've got to write up your thoughts on AI grant today for a job application. Suddenly, there's a knock at the door. You get up, and bump into your side table on the way out, knocking over your glass of 80% alcohol beer, ruining your heirloom Turkish carpet. Cursing yourself, you vow to never place your drinks near so sharp a drop again. 

In other words, you choose to change your failure mode from sometimes knocking your drink to the floor to sometimes failing to reach your drink. 

But what failure modes should you choose? 

As I go about my life, I often need to leave objects on their own for a while. Sometimes it's for a moment, like when I halt my bike and search through my backpack. Other times it's for a bit longer, like when I put a cup of water on my desk while I work. 

But in each case, I find it helpful to place the objects in such a way that there are few ways for it to drop far. The bike has it's kickstand out, and hugs a wall. The cup is far from an edge. 

You can view this as choosing the potential energy landscape about your object. You want to place your object somewhere s.t. there are no nearby plunges in potential energy that it could be knocked down. So even if your object gets kicked out of equilibrium, it doesn't gain too much energy and get messed up as a result. The bike doesn't drop onto someone, the glass doesn't shatter on the floor etc. 

This generalized beyond gravitational potential, of course. It also applies to e.g. drops in chemical potential energy, or in free energy. 

Let me give you a grim example. I knew someone who bought a free-standing heater for their room. One day, they were busy, and their kid was playing around with their heater. They were old enough, of course, to know not to stick their fingers into it. But they weren't old enough to know not to climb on it. Predictably, the heater fell on the kid, and burned their entire back. They've still got the scars today.

In general, you want to avoid failure modes with large downside risks, like dancing at the edge of a cliff, cycling on the pavement where children are playing or gambling with your life savings. Instead, go dancing at the nightclub, cycling on the road away from the kids or gambling with sub Kelly-sized stakes. 

In some cases, you can even choose activities where even if you fail, you're better off than you were before. Say, you try to solve a problem beyond your current capabilities, like writing a ray tracer for black holes. Even if you fail, you'll have learnt something. 

But the main point is that you can choose your failure modes to mind the cliff.