I think you're right about the clunkiness of the intelligence-to-political-success comparison. IQ is obviously good for solving problems, and there are indeed problems in politics - or you could consider "how best to be elected" as the problem to fix. But I think the reason that "the highest IQ person" doesn't always win is that who gets elected depends way more on the makeup of the electorate/country/events. Which sort of works into a potential Cowen-style subtext (maybe explicit, I haven't read the book) of maybe the key to getting better candidates (both types) lies in improving yourself/your company.
Myself, I'm unsure how much I, or anyone, should be trying to game systems versus, e.g. 'refuse to play' or find better systems to 'honestly' play in.
I definitely am inclined, by temperament I think, to refuse to play 'bad' games. I _think_ my primary justification is that I expect to be miserable 'playing along' in the obvious or 'accessible' ways I could have (or could in the future) do so, and miserable enough that doing so wouldn't even be net-positive were I to otherwise think playing along would be a reasonable 'risky bet' to make, e.g. to 'save the world' or 'be successful'.
But I _suspect_ that those thoughts might be self-serving (e.g. 'ego defense') and rationalizations to deflect from my petulance and resentment about what I think of as 'bullshit games' that I've been invited (or more or less coerced) into playing.
More concretely, I'm professionally a '(computer) programmer' and I've been invited several times to apply to work for, e.g. Amazon and Google. Just based on what little I know about how interviewing works (and, e.g. how long it usually takes) for those kinds of companies, and how uncertain I think I should be about specific expectations about the kind of work I'd end up doing (or with whom I'd be working with), even were I to make it thru the interview/hiring process, I've had approximately zero interest in pursuing those opportunities even tho they are, in a very real and meaningful sense, the obvious best avenues to achieve 'success' in my current career.
If a game is in some sense bad that is a strong signal that there are costs to playing it, and that you would not enjoy doing so or thrive by doing so. Yet that does not mean that it is not worth playing, depending on the payoffs and your other opportunities.
I don't know if you've read the Moral Mazes sequence (or Moral Mazes itself) but to me that would be the best reason not to apply to Google - you think that going to Google means more of such dynamics and don't want that. I don't know about Google in particular for this, but even at the starting levels the pay is very good and it's not obvious such dynamics much matter yet. Others can hopefully speak more to that. I do hear you need to be able to at least get along with the woke.
The other good reason not to apply is a better opportunity, especially with equity attached or better yet as a founder.
I realize the interview processes involved are not quick or free to do, and also that it's hard to know a lot about whether the job will be a fit until later, but my guess is that these are pretty bad reasons to not apply - if you think conditional on getting and taking the job you were right to take it, I'd apply. The potential upside to your entire job, career and life path are pretty huge. Even in failure you learn a lot of highly relevant information for you.
The good reason to not game the system to get success is if you don't want more 'success' with its air quotes. And that's a damn good reason, if it applies.
I'm pretty sure I've read all of your Moral Mazes posts, tho not the book itself. (I _think_ you might have _not_ recommended the book itself because the examples it uses aren't particularly central/prototypical of what you consider to be the core abstractions?)
I would tho expect Google (or similar) to consist-of/contain FAR more moral mazes than my current employer. I've mostly worked for very small companies (in terms of the number of employees). I do expect tho that working for Google would be much better than other company's of similar size. I'm inclined to continue that, for lots of reasons, tho I might seriously reconsider when I'm next ready (or 'forced to') look for a new job.
But, if I'm going to make an effort of the size I'd expect to need to make to be hired by one of these companies, I think I'd instead try to find a role doing AI alignment research (or research support).
What seems to me like perhaps the largest tradeoff is time. I really don't want to give up time with my child, let alone friends or family, and even much greater compensation just doesn't seem like a good trade to me. I couldn't buy _that much_ time with even a LOT more money.
(Another reason I'm skeptical that I would be satisfied working for Google specifically is that I'm still 'salty' about them killing Google Reader and it'd be very dispiriting to work on something like that only for them to decide it just wasn't worth them continuing it.)
And yes, I'm mostly indifferent to 'success' – that's seemed to me, for decades, like an obvious 'trap'. I've for a long time explicitly come down on the side of 'satisficing' versus 'maximizing' in that regard. I'm fairly content with my accomplishments especially because I've deliberately avoided 'gaming the system' so far. But I think I'd be willing to trade against that for other sufficiently compelling or valuable benefits.
I don't understand Google engineer to be an especially time-demanding job, so I wouldn't be too worried about time with your kid, and if you can stay disciplined earning a lot more money can then give you flexibility to work less later or take more risk or do something in alignment or what not. Can totally see 'if serious could just do alignment directly now' as an argument too.
The thing in your model I'd most question is that you seem to think that getting hired at Google requires some huge effort, and I'm not convinced given you have the underlying actual coding skill?
I do think that maximizing your chances - if you don't want to blow your shot and what not - involves a big effort, but if you would otherwise never apply at all then I think taking the shot in a not-fully-maximized manner is probably a lot easier? And probably gets you hired a lot more often per hour of effort, and perhaps nothing is lost if it doesn't work. But you have to feel that way.
I'm really grateful for your (many) replies to this thread! But I'm also feeling a little guilty about maybe taking advantage of your generosity.
I was considering asking you to name a 'cheerful price' to talk to me about this further (and in a more private 'channel'), for say 30-60 minutes, as I think you might be able to give me sufficient 'activation energy' to possibly make an effort at all along these lines. I _think_ you can find my email via Substack. You're welcome to follow-up that way with me directly if you're interested.
I'll reply here anyways with what I'm comfortable sharing publicly.
My current job is particularly _un-demanding_ in terms of my time, and is extremely flexible otherwise. I agree that "earning a lot more money" is useful for the reasons you mention. I had recently decided to do at least another cursory investigation of 'doing alignment directly', but, like applying to work at 'Google', I have what are essentially a bunch of 'personal hangups' that make me think I wouldn't find a satisfactory opportunity.
I don't remember which of Google/Amazon/other I actually 'made an effort' for before, but it was at least several years ago. I think I'm a 'pretty good' coder. But I don't think I've ever had an interview that even remotely exercised anything like the 'mental motions' I use while coding. The 'computer science' questions, and even whiteboard 'problems', seem not entirely but still nearly unrelated to any professional work I've ever done (even those projects most superficially similar). It doesn't help that I've mostly only ever interviewed when I _wanted_ (or 'needed') a new job.
Most of the hiring interviews/meetings seemed to be 'filters' along the lines of 'has a CS degree' (which I don't – I have a math degree) or 'claims X years of experience with Y'. Despite suggestions by friends to do so, I'm not willing to lie on my resumé or in interviews about any of that. I've also long noted that a good number of 'poor fits' seem to make it thru those filters anyways. I do expect that, having several more years of 'experience' (really, in my mind, just more 'legible' titles for previous and current roles), it would be easier to pass the various filters.
It is interesting, I get the feeling that the authors are not very good interviewers, and that you are a bit negative on the book. Which is odd because you said the opposite, I think.
Do you think it is possible they manage to get good talent (which seems true) but for very different reasons than they think? I have at least one guess for why Cowan might, but I don’t know anything about the other guy.
Great review , by the way! I really enjoyed reading it and might read it again later, just in case I was too sleepy last night to have really understood it properly.
Interesting. My guess is that impression comes from the focus on the places I question or disagree with the book, since those seem like the places most worth discussing. If we agree, there isn't much to say.
Also, interviewing and finding talent are hard problems where in some sense everyone is terrible at them and the goal is to be less terrible than others.
My guess is that the authors are very good at interviewing compared to their peers/competition, especially those without the experience/focus on the problem, but that there's still a lot that could be improved. Tyler mentioned in one podcast he might do a sequel.
It would be interesting to study the interview habits of various people/companies and see if there really are strong differences in outcomes. I have been through an outsized number of industry and a good number of academic interviews, as well as being on the other side of the table in academia for a few, and often thought that the process was really random in outcome. That is to say, the interviews were probably all equally bad, and if they got one of the top 1-3 applicants it might be another aspect, or just be dumb luck.
Agreed though that it does make sense that you picked the things most likely to be disagreed with. I am wondering if the interviewing for founders/people to hand money to is also just very different from the normal "We have this job. We want to find out if you would be good at it," sort of interview.
There's no question that hiring in general is a skill and some people are vastly better at it than others. How much of that is interviews is less clear but I do know that some companies have a good process and others don't, and some VCs ask good questions and some don't - some VCs would have noticed whether our business was a good idea, others didn't bother asking. Some learned a lot about who I was, others didn't. Peter Thiel was likely the best VC at interviewing I've experienced, whether or not I agreed with what he did with that info.
> ...at some point you risk crossing from Goodhart’s Law problems to 2nd level Goodhart’s Law problems, where you are effectively optimizing a metric that was originally predicting who would be able to successfully game the original metric, and so on.
>You absolutely want extreme forms of ambition and self-improvement, but in a sufficiently broken and hyper-competitive world this starts to select against caring about anything else, especially if those around you are running forms of meta-Goodhart and punishing caring about anything else.
Could someone help me understand the second-level Goodharting? You want to hire someone who will be more productive than other applicants. To predict this, you measure the ambition they displayed before and during the interview(s). Expecting this, the applicants tend to optimize for displaying ambition. This comes at the expense of productivity, and other useful behavior. There's the Goodharting, right? If so, what is second-level Goodharting?
1st level: You have a metric for how ambitious and self improving they are, such as asking entry level employees "Where do you see yourself in 5 years" and rewarding answers like "Leading the functional team!" or asking "What do you do in your free time?" and rewarding for "I learn new programming languages!" (I once had a boss in supply chain who said his hobby was also supply chain... he was a great guy, but... I am afraid he was serious.)
The problem then is that applicants realize you are using those questions to measure, and so start answering "I will run the company in 5 years!" and "I study exactly what this company does and I have zero other hobbies." Now you have to discount all those answers, and anyone who answers honestly and reasonably looks worse.
Level 2: Now you have a problem where you want people to be honest-ish, because really ambitious people will learn to game the system, so you want to select for people gaming. The trouble is that now you have to look at every answer and guess "Is this person gaming just the right amount with their over zealous answer, and therefore desirable, or are they just batshit insane and honestly think they could be CEO in 5 years?"
What if they are smart enough to realize that is what you have to do, and just tell you "Look, I know what you are facing here, and what sort of information you are trying to glean. I can promise you that I am ambitious and serious about self improvement, but you can't really trust any answer I give you as being true, especially since sufficient ambition demands that I game the system and over state my ambition instead of being perfectly honest."
Worst case you over correct and end up hiring people who would murder their mother to feed their ambition, and wouldn't blink at betraying you or the company for a raise, and/or who is so single minded in their life that they have no peripheral life experiences or connections and end up committing suicide at their desk in 5 years when they get passed over the first time.
... I might be a overshooting here a little, but then I am disposed towards making it easier to fire people on the basis that hiring them with any sort of accuracy is really hard. Interviewing is pretty low value add for most every company, I think.
Thank you. I think I understand what I misunderstood. I thought that at level 2, the hiring manager would switch to a new metric (via a new question). And candidates, knowing this, would switch to gaming the new metric. But actually, level 2 means the hiring manager applies a different interpretation to the original metric (by trying to discount, to the appropriate degree, certain answers to the original question). And candidates, knowing this, switch to gaming the original metric in a new way.
> ...I am disposed towards making it easier to fire people on the basis that hiring them with any sort of accuracy is really hard.
I think easy firing is the preference of everyone I've ever heard or read on this subject. It makes me wonder what the steelman argument for making firing more difficult is, from a societal perspective if not a business perspective. Something about loss aversion and lawsuits? Perhaps the grief of the fired worker (& family), plus the risk of a lawsuit, outweighs the happiness of the newly hired replacement (& family), plus the firm's expected productivity gain (both directly and through happier coworkers)?
It seems like a difficult case to make. But I worry when everyone has the same perspective.
Wait to thank me till after Zvi confirms that is what he had in mind :D
I don't know that I can steelman the argument against easy firing, but I think the reasonable version is something like this:
Much of how employment related processes work assume you get a job and stay there for a really long time, or at least until you find another job; in other words, the system assumes overlapping, continuous employment.
Getting fired costs you your health insurance, which has issues if you have e.g. procedures scheduled before you can find a job and the new insurance starts up.
Most people do not have the savings to weather the time it takes to find a new job.
Unemployment insurance often takes a while to kick in after applying before getting your first check.
Outside of the coordination problems, if it were easier to fire people bosses would have more undue power over their employees, and bad things might happen.
Now, my response to that would be:
The current system is indeed bad, but partially predicated on the assumption it is harder to fire people. It would be much easier to get a job, and it would look less bad to lose one, if bosses could just say "Hey, it isn't working out" and fire people much easier.
People really should live more within their means, and have a few months savings available for situations exactly like "Lost my job." That's a bit normative, but I think the fragility of living paycheck to paycheck even for well employed/payed people is in part a result of how hard it is to get fired. Relying on state unemployment insurance is a really bad idea, in any case.
Insurance through employers is a really bad idea, but if you are going to insist on it one could make it a rule that employees get 2 months of continued insurance if fired. As it stands you can buy extra time through Cobra, though that is expensive. In any case, getting people insurance for a few months shouldn't be hard enough such that it makes a good argument against removing labor market matching inefficiencies.
As to the power imbalance, again I think this is largely corrected by how much easier it is to find a new job. The power to hold "I will fire you!" over an employee's head is entirely due to how hard it is to find a new job, and how hard it is to find a new job seems tightly bound to how easy it is to fire someone who isn't working out. (Assuming not all bosses are bastards, which from my experience is true.)
In general, I think people over estimate the difficulty of finding new jobs, or even new careers, and so are very worried about being fired. I expect that people who change jobs a lot on their own are not terribly worried compared to people who haven't looked for a new job outside their current company in years.
I would say you're not envisioning/explaining it quite the way I would but what you're pointing at is a thing very worth pointing at as well, and highly related, at a minimum. I think if you get it from that explanation, you mostly get it.
How much of the question about relevance of IQ relates to the 'Whole Package' Question, the people who are palpably bright and have high levels of energy are rare enough, that you get people who are smart and feel that they are owed something due to that intelligence, sort of a lazy or entitled smart, so on aggregate the downsides of grouping together the entitled smart and the smart who get things done induces a higher failure rate than you would want to deal with. As a manager having a sense of the reliability of a hire can be more important than the quality. For certain roles you definitely want someone who does a 80% quality job but does it all the time vs someone who 50% of the time knocks it out of the park, and 50% drops the ball, and I feel this happens more often with the very intelligent who may not be perfectly calibrated. So maybe the overvaluing of intelligence is not directly related to the value of intelligence but the fact that when it isn't part of the whole package you can get divergent results.
Great book review. I'm in the middle of it myself and found myself cringing at how I would feel if I were asked these questions or in these ways many times.
I've interviewed ~low hundreds of people over the past 10 years, mostly for straight out of undergrad level jobs in finance where there is a premium on problem solving.
I used to ask a "google" type question but phased them out as I found the same things as the authors (I was probably the last person on that train by 2014 anyway). However, one of the questions I asked (I made it up myself, it wasn't one of the famous ones) needed as an input the population of the world, and boy was that enlightening.
You would be absolutely shocked the variance in estimates for the world population by people holding undergrad degrees in finance in the US. Over time I figured out that whether they knew the world population was a much better indicator than anything else in the question.
I still usually find an excuse to ask people I interview what they think the population of the world is.
I would guess over 30%. I want to say as high as 50% but that has to be bias in remembering the wrong answers so clearly and I never kept track. Answers in the millions or trillions were not at all uncommon. "1 billion" was very common, too.
Pretty rare for someone to be within a factor of 2 either way and not just know the rough answer.
Guesses of the US population were not much better.
Something about not knowing how many other humans share the planet with you is crazy to me.
I think you're right about the clunkiness of the intelligence-to-political-success comparison. IQ is obviously good for solving problems, and there are indeed problems in politics - or you could consider "how best to be elected" as the problem to fix. But I think the reason that "the highest IQ person" doesn't always win is that who gets elected depends way more on the makeup of the electorate/country/events. Which sort of works into a potential Cowen-style subtext (maybe explicit, I haven't read the book) of maybe the key to getting better candidates (both types) lies in improving yourself/your company.
I'm (somewhat) joking, but what do you think the 'Straussian reading' is of the book?
Always a good question given the author. Perhaps that you are not gaming the system enough if what you want is success.
That's a good answer; thanks!
Myself, I'm unsure how much I, or anyone, should be trying to game systems versus, e.g. 'refuse to play' or find better systems to 'honestly' play in.
I definitely am inclined, by temperament I think, to refuse to play 'bad' games. I _think_ my primary justification is that I expect to be miserable 'playing along' in the obvious or 'accessible' ways I could have (or could in the future) do so, and miserable enough that doing so wouldn't even be net-positive were I to otherwise think playing along would be a reasonable 'risky bet' to make, e.g. to 'save the world' or 'be successful'.
But I _suspect_ that those thoughts might be self-serving (e.g. 'ego defense') and rationalizations to deflect from my petulance and resentment about what I think of as 'bullshit games' that I've been invited (or more or less coerced) into playing.
More concretely, I'm professionally a '(computer) programmer' and I've been invited several times to apply to work for, e.g. Amazon and Google. Just based on what little I know about how interviewing works (and, e.g. how long it usually takes) for those kinds of companies, and how uncertain I think I should be about specific expectations about the kind of work I'd end up doing (or with whom I'd be working with), even were I to make it thru the interview/hiring process, I've had approximately zero interest in pursuing those opportunities even tho they are, in a very real and meaningful sense, the obvious best avenues to achieve 'success' in my current career.
If a game is in some sense bad that is a strong signal that there are costs to playing it, and that you would not enjoy doing so or thrive by doing so. Yet that does not mean that it is not worth playing, depending on the payoffs and your other opportunities.
I don't know if you've read the Moral Mazes sequence (or Moral Mazes itself) but to me that would be the best reason not to apply to Google - you think that going to Google means more of such dynamics and don't want that. I don't know about Google in particular for this, but even at the starting levels the pay is very good and it's not obvious such dynamics much matter yet. Others can hopefully speak more to that. I do hear you need to be able to at least get along with the woke.
The other good reason not to apply is a better opportunity, especially with equity attached or better yet as a founder.
I realize the interview processes involved are not quick or free to do, and also that it's hard to know a lot about whether the job will be a fit until later, but my guess is that these are pretty bad reasons to not apply - if you think conditional on getting and taking the job you were right to take it, I'd apply. The potential upside to your entire job, career and life path are pretty huge. Even in failure you learn a lot of highly relevant information for you.
The good reason to not game the system to get success is if you don't want more 'success' with its air quotes. And that's a damn good reason, if it applies.
I'm pretty sure I've read all of your Moral Mazes posts, tho not the book itself. (I _think_ you might have _not_ recommended the book itself because the examples it uses aren't particularly central/prototypical of what you consider to be the core abstractions?)
I would tho expect Google (or similar) to consist-of/contain FAR more moral mazes than my current employer. I've mostly worked for very small companies (in terms of the number of employees). I do expect tho that working for Google would be much better than other company's of similar size. I'm inclined to continue that, for lots of reasons, tho I might seriously reconsider when I'm next ready (or 'forced to') look for a new job.
But, if I'm going to make an effort of the size I'd expect to need to make to be hired by one of these companies, I think I'd instead try to find a role doing AI alignment research (or research support).
What seems to me like perhaps the largest tradeoff is time. I really don't want to give up time with my child, let alone friends or family, and even much greater compensation just doesn't seem like a good trade to me. I couldn't buy _that much_ time with even a LOT more money.
(Another reason I'm skeptical that I would be satisfied working for Google specifically is that I'm still 'salty' about them killing Google Reader and it'd be very dispiriting to work on something like that only for them to decide it just wasn't worth them continuing it.)
And yes, I'm mostly indifferent to 'success' – that's seemed to me, for decades, like an obvious 'trap'. I've for a long time explicitly come down on the side of 'satisficing' versus 'maximizing' in that regard. I'm fairly content with my accomplishments especially because I've deliberately avoided 'gaming the system' so far. But I think I'd be willing to trade against that for other sufficiently compelling or valuable benefits.
I don't understand Google engineer to be an especially time-demanding job, so I wouldn't be too worried about time with your kid, and if you can stay disciplined earning a lot more money can then give you flexibility to work less later or take more risk or do something in alignment or what not. Can totally see 'if serious could just do alignment directly now' as an argument too.
The thing in your model I'd most question is that you seem to think that getting hired at Google requires some huge effort, and I'm not convinced given you have the underlying actual coding skill?
I do think that maximizing your chances - if you don't want to blow your shot and what not - involves a big effort, but if you would otherwise never apply at all then I think taking the shot in a not-fully-maximized manner is probably a lot easier? And probably gets you hired a lot more often per hour of effort, and perhaps nothing is lost if it doesn't work. But you have to feel that way.
I'm really grateful for your (many) replies to this thread! But I'm also feeling a little guilty about maybe taking advantage of your generosity.
I was considering asking you to name a 'cheerful price' to talk to me about this further (and in a more private 'channel'), for say 30-60 minutes, as I think you might be able to give me sufficient 'activation energy' to possibly make an effort at all along these lines. I _think_ you can find my email via Substack. You're welcome to follow-up that way with me directly if you're interested.
I'll reply here anyways with what I'm comfortable sharing publicly.
My current job is particularly _un-demanding_ in terms of my time, and is extremely flexible otherwise. I agree that "earning a lot more money" is useful for the reasons you mention. I had recently decided to do at least another cursory investigation of 'doing alignment directly', but, like applying to work at 'Google', I have what are essentially a bunch of 'personal hangups' that make me think I wouldn't find a satisfactory opportunity.
I don't remember which of Google/Amazon/other I actually 'made an effort' for before, but it was at least several years ago. I think I'm a 'pretty good' coder. But I don't think I've ever had an interview that even remotely exercised anything like the 'mental motions' I use while coding. The 'computer science' questions, and even whiteboard 'problems', seem not entirely but still nearly unrelated to any professional work I've ever done (even those projects most superficially similar). It doesn't help that I've mostly only ever interviewed when I _wanted_ (or 'needed') a new job.
Most of the hiring interviews/meetings seemed to be 'filters' along the lines of 'has a CS degree' (which I don't – I have a math degree) or 'claims X years of experience with Y'. Despite suggestions by friends to do so, I'm not willing to lie on my resumé or in interviews about any of that. I've also long noted that a good number of 'poor fits' seem to make it thru those filters anyways. I do expect that, having several more years of 'experience' (really, in my mind, just more 'legible' titles for previous and current roles), it would be easier to pass the various filters.
It is interesting, I get the feeling that the authors are not very good interviewers, and that you are a bit negative on the book. Which is odd because you said the opposite, I think.
Do you think it is possible they manage to get good talent (which seems true) but for very different reasons than they think? I have at least one guess for why Cowan might, but I don’t know anything about the other guy.
Great review , by the way! I really enjoyed reading it and might read it again later, just in case I was too sleepy last night to have really understood it properly.
Interesting. My guess is that impression comes from the focus on the places I question or disagree with the book, since those seem like the places most worth discussing. If we agree, there isn't much to say.
Also, interviewing and finding talent are hard problems where in some sense everyone is terrible at them and the goal is to be less terrible than others.
My guess is that the authors are very good at interviewing compared to their peers/competition, especially those without the experience/focus on the problem, but that there's still a lot that could be improved. Tyler mentioned in one podcast he might do a sequel.
It would be interesting to study the interview habits of various people/companies and see if there really are strong differences in outcomes. I have been through an outsized number of industry and a good number of academic interviews, as well as being on the other side of the table in academia for a few, and often thought that the process was really random in outcome. That is to say, the interviews were probably all equally bad, and if they got one of the top 1-3 applicants it might be another aspect, or just be dumb luck.
Agreed though that it does make sense that you picked the things most likely to be disagreed with. I am wondering if the interviewing for founders/people to hand money to is also just very different from the normal "We have this job. We want to find out if you would be good at it," sort of interview.
There's no question that hiring in general is a skill and some people are vastly better at it than others. How much of that is interviews is less clear but I do know that some companies have a good process and others don't, and some VCs ask good questions and some don't - some VCs would have noticed whether our business was a good idea, others didn't bother asking. Some learned a lot about who I was, others didn't. Peter Thiel was likely the best VC at interviewing I've experienced, whether or not I agreed with what he did with that info.
This seems to broadly agree with Joel Splotsky, who siad you want to hire for two traits:
1. Smart
2. Gets things done
> ...at some point you risk crossing from Goodhart’s Law problems to 2nd level Goodhart’s Law problems, where you are effectively optimizing a metric that was originally predicting who would be able to successfully game the original metric, and so on.
>You absolutely want extreme forms of ambition and self-improvement, but in a sufficiently broken and hyper-competitive world this starts to select against caring about anything else, especially if those around you are running forms of meta-Goodhart and punishing caring about anything else.
Could someone help me understand the second-level Goodharting? You want to hire someone who will be more productive than other applicants. To predict this, you measure the ambition they displayed before and during the interview(s). Expecting this, the applicants tend to optimize for displaying ambition. This comes at the expense of productivity, and other useful behavior. There's the Goodharting, right? If so, what is second-level Goodharting?
I'm going to run an experiment and see how other people explain this...
I think the answer is this:
1st level: You have a metric for how ambitious and self improving they are, such as asking entry level employees "Where do you see yourself in 5 years" and rewarding answers like "Leading the functional team!" or asking "What do you do in your free time?" and rewarding for "I learn new programming languages!" (I once had a boss in supply chain who said his hobby was also supply chain... he was a great guy, but... I am afraid he was serious.)
The problem then is that applicants realize you are using those questions to measure, and so start answering "I will run the company in 5 years!" and "I study exactly what this company does and I have zero other hobbies." Now you have to discount all those answers, and anyone who answers honestly and reasonably looks worse.
Level 2: Now you have a problem where you want people to be honest-ish, because really ambitious people will learn to game the system, so you want to select for people gaming. The trouble is that now you have to look at every answer and guess "Is this person gaming just the right amount with their over zealous answer, and therefore desirable, or are they just batshit insane and honestly think they could be CEO in 5 years?"
What if they are smart enough to realize that is what you have to do, and just tell you "Look, I know what you are facing here, and what sort of information you are trying to glean. I can promise you that I am ambitious and serious about self improvement, but you can't really trust any answer I give you as being true, especially since sufficient ambition demands that I game the system and over state my ambition instead of being perfectly honest."
Worst case you over correct and end up hiring people who would murder their mother to feed their ambition, and wouldn't blink at betraying you or the company for a raise, and/or who is so single minded in their life that they have no peripheral life experiences or connections and end up committing suicide at their desk in 5 years when they get passed over the first time.
... I might be a overshooting here a little, but then I am disposed towards making it easier to fire people on the basis that hiring them with any sort of accuracy is really hard. Interviewing is pretty low value add for most every company, I think.
Thank you. I think I understand what I misunderstood. I thought that at level 2, the hiring manager would switch to a new metric (via a new question). And candidates, knowing this, would switch to gaming the new metric. But actually, level 2 means the hiring manager applies a different interpretation to the original metric (by trying to discount, to the appropriate degree, certain answers to the original question). And candidates, knowing this, switch to gaming the original metric in a new way.
> ...I am disposed towards making it easier to fire people on the basis that hiring them with any sort of accuracy is really hard.
I think easy firing is the preference of everyone I've ever heard or read on this subject. It makes me wonder what the steelman argument for making firing more difficult is, from a societal perspective if not a business perspective. Something about loss aversion and lawsuits? Perhaps the grief of the fired worker (& family), plus the risk of a lawsuit, outweighs the happiness of the newly hired replacement (& family), plus the firm's expected productivity gain (both directly and through happier coworkers)?
It seems like a difficult case to make. But I worry when everyone has the same perspective.
Wait to thank me till after Zvi confirms that is what he had in mind :D
I don't know that I can steelman the argument against easy firing, but I think the reasonable version is something like this:
Much of how employment related processes work assume you get a job and stay there for a really long time, or at least until you find another job; in other words, the system assumes overlapping, continuous employment.
Getting fired costs you your health insurance, which has issues if you have e.g. procedures scheduled before you can find a job and the new insurance starts up.
Most people do not have the savings to weather the time it takes to find a new job.
Unemployment insurance often takes a while to kick in after applying before getting your first check.
Outside of the coordination problems, if it were easier to fire people bosses would have more undue power over their employees, and bad things might happen.
Now, my response to that would be:
The current system is indeed bad, but partially predicated on the assumption it is harder to fire people. It would be much easier to get a job, and it would look less bad to lose one, if bosses could just say "Hey, it isn't working out" and fire people much easier.
People really should live more within their means, and have a few months savings available for situations exactly like "Lost my job." That's a bit normative, but I think the fragility of living paycheck to paycheck even for well employed/payed people is in part a result of how hard it is to get fired. Relying on state unemployment insurance is a really bad idea, in any case.
Insurance through employers is a really bad idea, but if you are going to insist on it one could make it a rule that employees get 2 months of continued insurance if fired. As it stands you can buy extra time through Cobra, though that is expensive. In any case, getting people insurance for a few months shouldn't be hard enough such that it makes a good argument against removing labor market matching inefficiencies.
As to the power imbalance, again I think this is largely corrected by how much easier it is to find a new job. The power to hold "I will fire you!" over an employee's head is entirely due to how hard it is to find a new job, and how hard it is to find a new job seems tightly bound to how easy it is to fire someone who isn't working out. (Assuming not all bosses are bastards, which from my experience is true.)
In general, I think people over estimate the difficulty of finding new jobs, or even new careers, and so are very worried about being fired. I expect that people who change jobs a lot on their own are not terribly worried compared to people who haven't looked for a new job outside their current company in years.
I would say you're not envisioning/explaining it quite the way I would but what you're pointing at is a thing very worth pointing at as well, and highly related, at a minimum. I think if you get it from that explanation, you mostly get it.
I presume the spelling of “But you can prey they don’t alter it any further” was deliberate at least on the part of your subconscious.
“What Did You Like to Do as a Child?” is a good question because it differentiates high agency kids.
High agency kids get around the bossing.
How much of the question about relevance of IQ relates to the 'Whole Package' Question, the people who are palpably bright and have high levels of energy are rare enough, that you get people who are smart and feel that they are owed something due to that intelligence, sort of a lazy or entitled smart, so on aggregate the downsides of grouping together the entitled smart and the smart who get things done induces a higher failure rate than you would want to deal with. As a manager having a sense of the reliability of a hire can be more important than the quality. For certain roles you definitely want someone who does a 80% quality job but does it all the time vs someone who 50% of the time knocks it out of the park, and 50% drops the ball, and I feel this happens more often with the very intelligent who may not be perfectly calibrated. So maybe the overvaluing of intelligence is not directly related to the value of intelligence but the fact that when it isn't part of the whole package you can get divergent results.
Great book review. I'm in the middle of it myself and found myself cringing at how I would feel if I were asked these questions or in these ways many times.
I've interviewed ~low hundreds of people over the past 10 years, mostly for straight out of undergrad level jobs in finance where there is a premium on problem solving.
I used to ask a "google" type question but phased them out as I found the same things as the authors (I was probably the last person on that train by 2014 anyway). However, one of the questions I asked (I made it up myself, it wasn't one of the famous ones) needed as an input the population of the world, and boy was that enlightening.
You would be absolutely shocked the variance in estimates for the world population by people holding undergrad degrees in finance in the US. Over time I figured out that whether they knew the world population was a much better indicator than anything else in the question.
I still usually find an excuse to ask people I interview what they think the population of the world is.
Wow. That's really something. So what percentage of them had the wrong number of digits in their answer?
I would guess over 30%. I want to say as high as 50% but that has to be bias in remembering the wrong answers so clearly and I never kept track. Answers in the millions or trillions were not at all uncommon. "1 billion" was very common, too.
Pretty rare for someone to be within a factor of 2 either way and not just know the rough answer.
Guesses of the US population were not much better.
Something about not knowing how many other humans share the planet with you is crazy to me.