spongy floors

the perception of impending doom, and how to foster it

I promise, “Value at Risk: Its Part In My Downfall” is on the way, but I have been distracted by this lovely little rabbit hole … a few conversations over the last couple of weeks have got me thinking about regulatory craft again, and about Malcolm Sparrow’s argument that the management and control of complex organisations can’t be about eliminating discretion.  In caricatured, oversimplified form, he says that this is because it’s always the case that either:

a)      a rulebook is so complicated that there’s no possibility of having enough resources to enforce every rule at all times, or

b)     it’s simpler than that, and consequently there are loads of cases which aren’t covered (or are covered in a perverse or ridiculous way) by the rulebook.

In the first case, discretion is needed to decide which rules to prioritise; in the second it’s needed to make judgement calls in the unusual cases.  Readers will have noticed that these two cases are by no means mutually exclusive.

So, the problem of regulation and management is one of how to structure and support discretion, and to increase the proportion of good regulatory decisions.  In which context, I remember a whole load of books I read while researching “The Unaccountability Machine”, but which in the end didn’t find a place in the narrative.  These were on the research into decision making, and specifically “Recognition-Primed Decision Making”.

As I’ve said before, one of the purposes of this ‘stack is to try to be the cognitive equivalent of ambient music; not to always be trying to argue for conclusions, but to push out ideas that stick in the back of your mind.  One thing that really stuck with me from Gary Klein’s book on decision-making was the idea of “spongy floors”.

Klein’s research concentrated quite a lot on fire chiefs and the decisions that they made on when to bring a crew out of a building.  This is an unusual case – it’s a high-stakes decision that needs to be made under time pressure, by a single individual.  It turned out that one of the most important inputs to that decision was whether the floor of the building was reported by experienced firefighters to be “spongy”. 

Sponginess is a property of floors of burning buildings and describes how they feel when you step on them.  A floor becomes spongy when there is severe structural damage, and this indicates that the floor (and the whole building) is likely to collapse.  Klein found that sponginess had interesting conceptual features:

1)     Sponginess doesn’t seem to have a stable or predictable relationship to anything that might be usefully measurable for a fire chief – temperature, time of burn, etc.  Building collapses are themselves pretty unpredictable, and spongy-flooredness is just an initial stage of collapse.

2)     It has good intersubjective replication; experienced firefighters will almost always agree on whether a floor is spongy or not.

3)     Over the course of dozens of interviews, Klein didn’t find anyone who was able to describe the sensation of sponginess in any terms more precise than “it felt spongy”.

4)     To an inexperienced firefighter, all floors in burning buildings feel spongy.

There’s no need to posit any kind of sixth sense here; all that’s happening here is that there is a complex, high-variety physical property which human beings are capable of learning to detect.  Physically, sponginess presumably corresponds to a network of small cracks beginning to join up, and this affecting the elasticity and rigidity of the floor in a way than can be tactilely detected through the soles of firefighting boots.  It’s an impressive thing, but in a world where we know people can be taught to echolocate in ten weeks, it shouldn’t completely surprise us.

Klein argues that there are similar kinds of learned pattern recognition that people pick up for all sorts of other problems, not just learning to distinguish new physical sensations.  So you might say – this wasn’t in his book, but I definitely think I knew people who could do this – that an experienced bond trader might look at the order flow and think some equivalent of “the floor’s getting spongy” for a market crash.

The idea of regulatory craft might be that you need to create conditions in which people can learn these pattern-recognition skills, then demonstrate that they have the competence, and then be put in a position where their judgement is respected[1]. Of course, this raises the interesting question, which I shall not answer at this stage, of “could you train a neural network to do this?”.

---

[1] “Respected” does not mean the same as “blindly obeyed”! Another thing that Klein found was that although experienced firefighters know what a spongy floor is and what it means, very often the fire chief found themselves having to issue orders to pull out, to a team of firefighters who nevertheless wanted to stay on and keep fighting the fire for a few minutes more.