Sunday, December 6, 2015

The Scylla-Charybdis Heuristic: if more X is good, why not infinite X?

(Charybdis = care-ib-dis)

A Scott Alexander post really resonated with me, when he talked about one "development milestone" (#4): the ability to understand and speak in terms of tradeoffs, rather than stubbornly try to insist that there are no downsides to your preferred course of action. Such "development milestones" indicate a certain maturity of one's thought process, and greatly change how you discuss ideas.

When a Hacker News discussed Alexander's post, I remembered that I had since started checking for this milestone whenever I engaged with someone's advocacy. I named my spot-check the "Scylla-Charybdis Heuristic", from the metaphor of having to steer between two dangers, either of which have their downsides. There are several ways to phrase the core idea (beyond the one in the title):

Any model that implies X is too low should also be capable of detecting when X is too high.

Or, from the metaphor,

Don't steer further from Scylla unless you know where -- and how bad -- Charybdis is.

It is, in my opinion, a remarkably powerful challenge to whether you are thinking about an issue correctly: are you modeling the downsides of this course of action? Would you be capable of noticing them? Does your worldview have a method for weighing advantages against downsides? (Note: that's not just utilitarian cost/benefit analysis ups and downs, but relative moral weight of doing one bad thing vs another.) And it neatly extends to any issue:

- If raising the minimum wage to $15/hour is a good idea, why not $100?

- If lifting restrictions on immigration is good, why not allow the entire Chinese army to cross the border?

- If there's nothing wrong with ever-increased punishments for convicts, why not the death penalty for everything?

One indicator that you have not reached the "tradeoff milestone" is that you will focus on the absurdity of the counter-proposal, or how you didn't advocate for it: "Hey, no one's advocating that." "That's not what we're talking about." "Well, that just seems so extreme." (Extra penalty points for calling such a challenge a "straw man".)

On the other hand, if you have reached this milestone, then your response will look more like, "Well, any time you increase X, you also end up increasing Y. That Y has the effect/implication of Z. With enough Z, the supposed benefits of X disappear. I advocate that X be moved to 3.6 because it's enough to help with Q, but not so much that it forces the effects of Z." (I emphasize again that this method does not assume a material, utilitarian, "tally up the costs" approach; all of those "effects" can include "violates moral code" type effects that don't directly correspond to a material cost.)

I've been surprised to catch some very intelligent, respected speakers failing this on their core

What about you? Do you make it a habit to identify the tradeoffs of the decisions you make? Do you recognize the costs and downsides of the policies you advocate? Do you have a mechanism for weighing the ups and downs?


Michael said...

I think you're discounting balancing multiple models, or expecting a single model to do too much. To take your $15 vs $100 minimum wage example, the push for $15/hr is very focused on the microeconomic, that it's difficult to maintain a reasonable livelihood, much less a family , on $9/hr in many parts of the country. The models that say that a $100/hr minimum wage would be a problem are entirely different, and generally focus on a larger scale. Those models would likely advocate removing a minimum wage altogether, which could (possibly) result in a catastrophe for any number of people and families.

Admittedly, this is a part why we're often at such a stalemate... "High minimum wage is good for workers!", "High minimum wage is bad for business!" Honestly, they're both probably true, and I doubt there are any models sufficiently sophisticated to tell us that "The correct minimum wage to maximize quality of life is exactly $11.52." Much of the argument comes down to a gut feeling of where the balance of power currently lies, and which direction would be more "fair" to shift it.

(And yes, if a primary counter to an argument is to multiply it by 10 and ask "why not that?", that falls into the general understanding of "straw man", so thanks for the bonus points.)

Also, Hi! It's been a while, hope things are going well.

Silas Barta said...

But hold on Michael: The problem isn't simply about comparing "benefits to workers" and benefits to business". A sane model of this situation captures the fact that a high enough minimum wage should hurt workers too. Indeed, *any* increase should have some downside, even for workers, and the "tradeoff-milestone mind" should be able to model this.

The whole reason for the SX Heuristic is to detect flawed models like "a minimum wage accomplishes nothing except for transferring wealth from the rich to the workers, in the same jobs as they currently are".

And yeah, it has been a while :-/

Michael said...

Sure, but that's an oversimplification of the model being put forth by advocates for $15/hr, which is, let's say "Increasing minimum wage (up to a point) will transfer wealth to the workers most in need of, and deserving of it, without unreasonably harming business owners or investors." Of course, there's a lot of subjective pieces in that model, but just because there's not room for all those caveats and assumptions on a poster or a tweet doesn't mean that they aren't generally taken into account by advocates.

The SC Heuristic as stated seems to assume that the status quo is pointing down a moderately safe path. To perhaps overextend the metaphor, if you see your ship is heading straight for the whirlpool (let's say the whirlpool represents income inequality, or something), the only rational choice is to spin the wheel as hard as you can. It's not that they're aren't consequences to that action, it's that, at times, the consequences of under-action are worse than the consequences of over-action.

Silas Barta said...

@Michael: an implicit requirement of using the SXH[1] is that the downside model be a reasonable one. There's obviously going to be some disagreement of it, but your downside model of the minimum wage should be something more than "well gosh, it sure hurts those wealthy fatcats just a tiny bit, but overall, we make up for it by all the good we do for the less fortunate".

Hopefully you accept that a $100/hour minimum wage would be bad even for workers. Well, your model needs to incorporate that across the spectrum, so you don't have to treat the $100 as some special case "because it's absurd".

FWIW, I haven't seen an advocate actually able to model such a downside; the implicit assumption is that a higher minimum wage *only* increases wages, and has no downside for workers; or if there are, they can be ignored despite the extreme concentration of the costs among the most vulnerable.

You're right that in some circumstances, you don't want a full, fair model before steering father one way; but policy debates like the minimum wage are not conducted within such emergencies, and it's reasonable to expect advocates of such changes to have a model.

[1] I think the abbreviation should be X instead of C due to coming from the Greek chi (cf Xmas).

Michael said...

SXH it is.

So yes, I don't think supporters of a higher minimum wage have made a comprehensive model that captures how many jobs would be lost, sent overseas, or other economic consequences. Likewise, I don't think advocates of the current, or lower minimum wage have a model that meaningfully captures the quality of life for workers and their families, communities, and other social consequences.

It would be awesome if we could have super-comprehensive models like that, but given the massive disagreement on even fairly limited focus questions (eg "How much will tax package X increase or decrease revenue"), I don't think it's reasonable to expect advocates for a given cause to explicitly model other concerns, though I do think they have a responsibility to acknowledge the most plausible models in other areas.

I don't think there's any more obligation for one side to answer "Wouldn't $100/hr be a problem?" than there is for the other side to answer "Wouldn't a $0.01/hr decrease be trivial for a person to absorb?" In a good-faith discussion, side A says "We feel that $15/hr would dramatically increase happiness", and side B says "That would cost 100,000 jobs." A: "Hmm, $12.50/hr?" B: "10,000 jobs.", or something along those lines until a sweet spot is found.

Ultimately, I think we're more likely to manage a bunch of fairly simple models that can be compared and shuffled together as necessary over a single super-model that's capable of answering the meaning of life, the universe, and everything.