The New York Times has a section in the most recent magazine called the Ninth Annual Year in Ideas. Divya Manian (@nimbupani) alerted me to the second idea in the business section: random promotions.
In 1969, the Canadian psychologist Laurence J. Peter posited the “Peter Principle”…Eventually the entire economy becomes like the paper company Dunder Mifflin in “The Office” — clogged with incompetence…Is there any way to avoid this trap? Yes, by promoting people at random.
It’s a short piece, and is based on organizational dynamics simulations by a trio of Italian scientists. Go check it out. It is an intriguing thought: that random promotions might break the Peter Principle. Do they break or validate the Gervais Principle?
The basic idea of using random processes to mitigate the pathological effects of deterministic models is extremely solid, and well known at least in engineering and biology. Randomized algorithms, as they are known, can get you better and/or faster answers to problems that are intractable in their deterministic form. I spent a good deal of time developing algorithms of this sort to solve team formation problems during my PhD (short version: I attempted to find “dream teams” by trying teams at random and gradually favoring the more successful ones). I am very partial to randomness as an attack on problems. Randomized methods tend to be inspired by physical processes such as evolution (genetic algorithms), annealing (cooling metals, which inspired “simulated annealing”) and spin-glasses (things like magnets). This appeals to my metaphoric instincts.
But simulation models are simulation models. Can this work in practice?
Randomess in Education
It just might. I’ve heard it suggested that the quality of outcomes in higher education could be vastly improved by setting a reasonable baseline (like a SAT threshold for colleges) and simply choosing candidates for top universities randomly above that threshold. So rather than Harvard and MIT skimming the supposed top-tier candidates based on all sorts of detailed criteria like essays, awards and social service, you’d just do a lottery.
I strongly believe that this is the direction education needs to go, and I am basing this on anecdotal evidence, including my own personal history. Over the years, I’ve met several people who toiled in obscurity in no-name universities, who have absolutely stunned me by their originality and creativity. The majority of graduates of elite universities on the other hand, all look indistinguishable. Pick-the-best models in education are not quite as bad as the sorts of over-performance that put you on the clueless track at work, but come close. Tests, as the cliche goes, largely test for your ability to pass them, and only incidentally for more fundamental things like your language or mathematical talents.
In my own history, I was first a beneficiary of a “cream” selection strategy (getting into Asok‘s alma mater, the IIT (Mumbai)). At IIT, I was mostly a disengaged and mediocre student, outside of a couple of courses that interested me. I spent the rest of my time on reading, writing, theater, goofing off, and swimming. Being on the swim team actually led to my “random” promotion to the University of Michigan for grad school, through a friend on the team who recommended me to his advisor for an assistantship. Based on my mediocre class ranking at IIT, I’d never have gotten into a grad program as highly ranked as Michigan. But I did pretty well there, after one false start, and my thesis adviser and committee at least, were very happy with my work. My technical work has never been called “technically brilliant” or “genius,” but what praise I’ve received tends to use the adjective “original.”
Randomness at Work
I think randomness is a brilliant strategy in organizations, but I don’t think you can operationalize it. That’s a contradiction in terms. The right way to increase randomness is by being lazy and sloppy in enforcing deterministic models. The minute you try to codify “randomness” you just create a whole new game.
By “deterministic” I mean the sort of talent-spotting algorithm proposed by Joel Spolsky in Smart and Gets Things Done. This sort of thing at once impresses and scares the hell out of me, because I know I’d never make it past such tough and “professional” gatekeepers anywhere (and I have the failures to prove it). Joel’s outfit, Fog Creek software, seems to have become a byword for the “hire only the best” philosophy. This reminds me of a Dilbert cartoon:
PHB: We hire only the best
Dilbert: Isn’t it also our policy to pay according to the industry average?
PHB: Yes
Dilbert: So we like ’em bright but clueless?
(I just remembered this cartoon by the way, so I am beginning to think there is more Gervais principle evidence in Dilbert than I originally thought).
Joel’s take apparently accounts for this in the gets things done part of the heuristic, but I think this is extremely hard to actually gauge a priori, so you effectively end up hiring on the “smart” part. I think pure paper-merit strategies (i.e. promoting/hiring potential clueless over-performers) will create short-term success, but long-term, cause organizations to lose their way due to insufficient creativity. The sorts of people who get past any codified strategy will likely be overperforming craftsmen and craftswomen with major blind-spots in other areas.
Long-term organizational success depends on randomness, and intellectual diversity is the type of randomness you need. Failing to pass traditional tests is a necessary, but not sufficient indicator. This means throwing in a few wildcards: people who don’t look good on paper, don’t interview well and seem generally like misfits. The more trouble your company is in, the more wildcards you need.
Fortunately, there are enough sociopaths around who ignore traditional hiring/promotion criteria beyond a point and go with their gut by placing random bets. My career has largely been driven by people placing such bets on me, and me placing such bets on others. So my preliminary conclusion: the results of the Italian research are entirely consistent with the Gervais Principle. Sociopaths promoting other sociopaths based on gut instincts might effectively be the random signal needed.
Postscript: Now that I think of it, the very beginning of my educational career started out randomly. I was an extremely shy kid, and though I knew the alphabet, I hid behind my mom and refused to recite it during my pre-school interview at a “good” school. Fortunately the headmistress was a personal friend of my parents, and I got into the school. Straight D’s in kindergarten. Things picked up after that.
About Venkatesh Rao
First I thought “he’s gotta be kidding” but instead it’s little more than insincerity in your use of language.
Example ” go with their gut by placing random bets”. ??? There is nothing less random, and more efficacious than a good gut feel. People telegraph their intent, the gut picks it up, presto-change0 we have success where normal metrics can’t determine the constraint of who sincerely wants to succeed.
I don’t disagree much with your conclusions then, perhaps your language is a bit slanted towards shock value where it could instead pick up nuances that empiricism doesn’t want to address.
Not quite. I genuinely believe that gut feelings inject randomness. Read my review of ‘Who: The A method for Hiring’ for instance. It is a very deliberate hiring process, and one of the first things the authors do is show just how arbitrary the ‘gut feel’ can be (based on pretty random things gut-interviewers pick up on). The result is high variance hiring outcomes (high risk/high returns… more chances of finding weird superstars, but also more chances of finding big wastes of money).
They conclude that because it is arbitrary it is bad. I accept their idea that it is arbitrary, and conclude that it is good.
“Gut feelings” aren’t simple. See Gerd Gigerenzer’s book “Gut Feelings”. They can be spectacularly on the money as you suggest, but they can go spectacularly wrong as well. The reason is that the gut picks up on and processes a lot more than conscious rational thought, but doesn’t always do relevance filtering rationally. You might take a violent dislike to a candidate simply because he is wearing a green shirt for instance, and you have some repressed childhood memory of some horrible associations with green shirts that your gut overweights.
Sometimes of course, the equation is simpler. Your gut simply tells you “this is extremely complex. pick randomly” and you do real coin-toss level random things. I’ve done that.
Venkat
I’m still back at the same point.
Agreed on the lack of efficacy on the gut, but that had nothing to do with the issue of the article’s semantics.
It was an example of “not random”. Just because any method is neither empirical or declared by name doesn’t mean it’s random. Rather, *random* in this usage comes across just a semantic for “not sure what to attribute this decision to, so I’ll call it random”.
Feels to me like I am splitting hairs for the sake of getting semantics right when there is no reason to do so. Really semantics isn’t that important because probably everyone already understands that you don’t really mean *random* when you are using it in this article. Kinda silly for me to even bring it up.
Well, it is not quite semantic hair-splitting. There’s an important point there. It is a common in modeling to represent unmodeled elements as effectively random. This is sometimes justified, sometimes not.
The entire behavioral economics revolution is based on the insight that in the case of the “rational man” model of economics, the unmodeled dynamics cannot be assumed to be random/mutually-canceling, but show up in systematic ways as biases.
So sociopaths promoting other sociopaths = effectively random should not be taken as an assumption, but needs testing. I believe at any rate, that it will create a lot more diversity in outcomes than deterministic processes.
The same sorts of randomness assumptions operate in crowdsourcing theories (where “private knowledge” makes individual behaviors effectively random in the macroeconomic sense)
> This sort of thing at once impresses and scares the hell out of me, because I know I’d never make it past such tough and “professional” gatekeepers anywhere (and I have the failures to prove it).
So do I. I have never made it past any such illustrious interview(except one). And I don’t think those are relevant either (maybe it is a case of sour grapes?).
While I do agree that ultimately people do place a random bet on you, but they do not think so, at that moment. There are always other reasons, like you mentioned about your parents “knowing the headmistress”, “gut feel”, “warm fuzzy feeling”.
So we arrive at the same conclusion, just a case of me not being able to get over myself.
“I don’t want to write another couple thousand lines of code to model this other behavior so I’ll just randomize it and it will probably work pretty close to the same in the model.” And it probably will.
So I’ve got some kind of perfectionism hangup with not writing the extra code, and your language works the same in the model either way. It just ain’t random, and I’m probably not going to get over my hangups because I believe we really can come up with better empirical models that reflect some kind of efficacy we can’t currently quantify. Just because we haven’t yet, I look like an idiot expressing the belief.
A weird example to analyze here is the Gladwell ‘Blink’ example of blind vs. open auditions for symphony orchestras. Without knowing it, and without overtly being sexist, male classical musicians were justifying all sorts of ‘gut feel’ ways of picking musicians, but were achieving a non-random effect (male dominated hiring).
Once screens were placed between the auditioners and juries, and the gut was blinded to everything but music, female hiring skyrocketed.
This example is weird because Western classical music is a weird domain, where most of the hiring you need to do is not of innovators but excellent craftsperson-performers. So you benefit by blinkering the gut so it only evaluates a complex but narrow area of competence.
In the opposite sort of domain, where technical skill is far less important than other, ambiguous variables, exposing the gut to more information probably increases, rather than decreases diversity (= randomness to me, not social justice. That is a side effect).
Conjecture: fast-growing companies probably need Spolsky type hiring (non-random) because every company is a Ponzi scheme, and those hired really early can break the “bright but clueless” tradeoff, since early-growth stock options can make these people rich. Later hires are bright but clueless, and at maturity, companies need wildcards to revitalize them (or do M&A etc., as per the MacLeod lifecycle)
Venkat
“The sorts of people who get past any codified strategy will likely be overperforming craftsmen and craftswomen with major blind-spots in other areas.”
I wonder what strategy Bell Labs used to assemble and run their research operation. For decades they had people who were really smart AND accomplished an astonishing variety of towering achievements. How did they do it?
Hamming’s talk goes into some of the details, but from the perspective of the researcher, not the corporate manager: http://www.paulgraham.com/hamming.html
Glad you brought up Hamming’s talk, because it illustrates some of the complexity. Particularly in his bit about Clogston, the guy he helped to do some math, which opened up his hands-on creativity and made him go on a stardom path.
There IS some managerial perspective there, from Bode, who was a major individual contributor too.
But I think Bell labs mythology should be taken with a grain of salt. It is JUST possible that they threw a lot of money at a lot of people, and there was enough in the portfolio that a few virtuous cycles took off.
But I think Bell labs mythology should be taken with a grain of salt.
What about Xerox Parc then? They weren’t that big.
Also I am glad someone takes Spolsky to task with his ridiculous “elitism”.
I don’t know enough about Bell Labs history to comment further, so I’ll leave it at that.
PARC was smaller, but had one interesting dynamic showing the same forces at work. Bob Taylor followed an elitist strategy, but when the work began to demand mortal labor, and Taylor wanted to stick to his un-scaleable “only the best” policies, some, like Simonyi, snuck in lesser mortals in the guise of testers. Simonyi’s large-scale s/w engineering models eventually became Microsoft :)
But these comparisons of free-rein R&D labs in the golden age of industrial R&D aren’t really useful for understanding line operations today. Very different dynamics.
I wouldn’t say I am taking Spolsky to task.. I’ve recommended his approach to others under some circumstances. Just pointing out that the model has its costs as well as benefits. And okay, throwing that Dilbert joke was too tempting to pass up.
I tend to think the “random” successes come from inadvertently doing something unusual: getting a good mix of specialists and “big picture” types together. most companies over invest in specialists and act surprised when they wind up with short sighted but technically well crafted plans. increasing the aerodynamics of the car by 2% won’t get you anywhere faster if you’re driving in a circle.
Consider also the other side of coin – random firings could be beneficial simply because they open up space for new players and new ideas.
Josef Stalin practiced this in an extreme way – lots of people were executed or gulaged for reasons not related to their performance in their jobs. The purges stopped with Stalin’s death in 1952. Soviet Union as a country and a world power was doing fairly well (it was agile and able to respond to changing situations), all the way into 1950-s despite the huge damage inflicted by WW2, but it started stagnating in the 1970s. The effect was very visible at the top of the communist party – overwhelming majority of Politburo members (the ruling council of the country) by 1980s were very old and unable to consider new problems faced by the country or new solutions needed to solve these problems. In the next 10 years they were still officially in power but have pretty much lost touch with reality and ended up losing control over the armed forces and the rest of the executive branch in 1991.
Modern democracies sometimes practice purges in a milder form known as “term limits” – political leaders are not allowed to stick around beyond, say, 10 years. I am not aware of any similar practices in corporate world – things have to become really bad before purges begin, which is really counterproductive.
Random firings… definitely like that idea. Valuable even to the randomly fired in some sense… too many people think of the world as a very deterministic place, and bouts of bad luck can help you wise up real quick about what you do and do not control.
Yes, the father of 3 who works tirelessly to keep his family fed and clothed and works hard during the week to spent time with his wife and kids on the weekend, he deserves to be randomly fired so he can “wise up real quick about what he doesn’t control.”
YOU can be the one to explain to him how valuable a lesson you’ve taught him.
To me, this is flatly unethical.
Not suggesting that this be arbitrary for the hell of it. What if you know you HAVE to let 10% of the workforce go, and there are no good criteria on which to base your decision. Would you rather make up ‘performance’ criteria, use a last-in-first-out rule, or use a random heuristic?
I am not particularly open to “father of 3” type arguments. Kids are certainly necessary to the economy as a whole, and childless people certainly need the willing parents to have kids to sustain things like social security. But people who have kids choose to have them, and using an unqualified “father, husband” type argument in this type of context is disingenuous.
A business is not a purely humanitarian enterprise. Humanitarian concerns have to be weighed against survivability of the business. In trouble, would you rather fire the hotshot young kid who you know is one of the few remaining hopes for turning the company around, but could easily find another job, or the 55 year old with 3 kids who is known to be deadwood?
Tough call, and you need to be as humane as you can with whatever flexibility you have beyond retaining the ‘necessary for company survival’ people. But if randomness is a good strategy for business survivability, it needs to be considered.
Venkat
This is a great series, thank you for that.
Could you point me to some other related material?
Ive read How to win friends and influence people, and 48 laws of power, wich seem somehow related.
This is a brilliant idea, and I don’t think you’ll have to wait too long before some eccentric companies in the tech industry start experimenting with it. Your academic/work experiences are a perfect example of Aaron Brown’s theories that economic activity as gambling (despite misguided millennia of moral philosophy trying to disentangle the two). Many professors and managers know (but might not admit) how often their good hires were gambles (and their bad hires too, of course, a gamble means you lose a lot), and, as I said above, I don’t think we’ll have to wait too long before they let that instinct escape the Platonic fold and enter empirical reality.
Venkat: congratulations also on withholding judgement on the lessons (if any) of Bell Labs or PARC or individual other cases, when so many would have made an easy and smart-sounding inductive leap. Your intellectual humility is refreshing on this blogosphere.
I didn’t see much mention of the Peter Principle, so here’s an unsolicited link to John Gall’s “Systemantics” which you may have heard of: http://tinyurl.com/2zhm7c
Venkat,
It sounds like the effects of incentives and risk lurk closely behind your thinking on the categorisation of workers whether using your Gervais Principle terms the above.
A potentially intersting question is what type of person different organisations should hire. I suspect that asking this question you might find that the lifecycle (https://ribbonfarm.wpenginepowered.com/wp-content/uploads/2009/10/compLifeCycle.JPG) is rational.
Small growth organisations (Giants in their infancy) can mostly keep moral hazards under control. This means that the risk takers can place the big leveraged bets that make million dollar companies into billion dollar companies. In between, there is some low risk – medium reward work to do milking. This is essentially collecting the winnings from earlier bets.
As organisations grow, incentives become impossible to keep straight. Moral hazards dominate, especially in the high risk/reward arena. To keep workers in any kind of agreement with company incentives requires risk averse people and delusions that keep people from acting on their own personal incentives (the former assists the latter). You need people who will allow a 50% chance at doubling their salary with a 10% of being fired to go by. The rationalize that the organization will eventually reward their hard work & above average performance.
A big bank, for example, cannot harness the performance potential of really big risk takers in the same way a startup can. There is to much room to maneuver within the organisation and too little room to maneuver organization itself. To think that some executive at a multinational bank charged with ‘customer satisfaction’ is actually incentivized to improve customer satisfaction seems laughable. Such people are highly aware of personal incentives & these do not correlate woth the organisations incentives. That environment, however, can put the people with predictable performance to good use. Overachievers to particularly good use.
The details have to be worked out, but I definitely agree that the lifecycle is rational in a macroeconomic sense, but possibly not for an individual firm (microeconomic sense), unless you count organizational suicide as a rational act (I am on the fence on that one).
Life-cycle stage hiring/promotions is conceptually clean, but the details are not yet clear to me. On a startup team, your first hires can’t be too random. Only when you get to 20-25 headcount do you have the portfolio diversity to sustain a few bets, for instance.
I think we are looking at this from different angles. I suggesting the life cycle is rational from a microeconomic POV, rational for the organisations themselves. I guess I was using an implicit assumption that those taking big, risky, rational bets are very aware of their own incentive structure.
Employee number 5 has the ability to influence the company & the incentive to do so. Employee number 500 doesn’t really. If she is risk seeking, employee number 5 place risky bets that benefit herself and the organisation. A risk seeking employee number 500 is still going to place large bets, but these will be of the kind that benefit her, not the organisation. What sort of risks would a bank teller or branch manager take? Would the bank benefit.
Dilbert’s bright but clueless become more relatively beneficial as organisations grow. What can a bright young risk taker possibly do as a bank branch manager? Anything noteworthy would be a major coup. I suppose these black swan (is this term becoming a cliche?) events are what you are trying to optimise for with your randomness.
You may be right, but even optimally this would still represent a minority of hires. It wouldn’t effect the proportion. Most bank managers would still need to be risk averse & predictably productive.
If this concern is real, is the solution simply—have small companies? Paul Graham (the startup guy) has argued as much numerous times. (Not that I know…)
Simple for who?
Your example of university lottery dovetails neatly with a chapter in Gladwell’s “Outliers” which looks at the extreme upper end of the IQ scale and the various trials which have apparently shown very little correlation between “success” and IQ, once a certain minimum standard is reached.
Gladwell goes on to suggest that above this minimum IQ standard, “creative/open” testing is a far better indicator of flexible thinking than “narrow/closed” testing.
As an example of open testing, when they were still running with their exam entry system, both Oxford and Cambridge were renowned for their “open” exam questions. In a 3 hr exam you were supposed to answer 6 questions along the lines of ‘how long is a piece of string?’, ‘what is courage?’ etc.
Apart from supplying endless schoolboy anecdotes about the various answers (the most memorable one for ‘what is courage?’ being: ‘this is!’) these sorts of questions are far harder to prepare for and allow less structured but more flexible thinkers greater scope to shine.
I’d love to start using them in graduate job interviews – just think of the panic that would ensue!
New Scientist has a piece covering the same research. Some interesting discussion.
So privilege equals randomness?
You’re contradicting yourself when you speak of your own success based on privileged access while dismissing other people who got the same access in the same manner. In fact, you’re not even so much as saying your extra-ordinary now are you? Do you think you might in fact be a part of the problem?