back
87 comments
Once upon a time, I had an employer who was looking for a machine learning expert. I mean, we had a brilliant mathematician who was breaking serious ground in the field, but we needed an expert, apparently. We were having trouble drawing experts out of Silicon Valley, and our resident genius said "Hey, I satisfy all of the requirements for this job you've posted, and I'm already doing the work. Can you promote me from this crap software-dev role you stuck me in?" and the company's response was "No, it clearly states in the posting that you'd need a PhD in machine learning." We promptly lost our genius. If only TPTB knew that it was a meritocracy!

In short: article's claims sound fishy; your lived experience may vary.

Meritocracy is a big word.

Chess is a meritocracy. So is tennis and sales. Anyone can play. The best player wins.

Chess coaching is not. Better coaches might win more or get better gigs, but stuff other than coaching "merit" can play a big role.

But rather than getting stuck in a semantic argument... What the article means is that it's a relatively accessible field, on the way that software engineering is. The formal pathways are not as locked in.

A motivated and talented individual can bootstrap a career without formal education, qualification or approval.

All this is relative to, for example, civil engineering, law, medicine... There you'll need qualifications earned via established formal channels, gatekeepers and such.

Could one say that meritocracy is not meritocracy? No matter what skill you measure, a population with adequate nutrition since conception is going to be far better than a population where malnutrition impacts both pregnancy and children.

So what is the take away, especially with applying this to all areas of our lives? We can do something to try to equalize the workplace, the chess tournament, or the tennis court. But what about equalizing the stage or the bar?

> Chess is a meritocracy. So is tennis and sales. Anyone can play. The best player wins.

I'm not sure you are having the correct definition of meritocracy in mind. A meritocracy is a system where the actors who provide the most benefit to the society, benefit the most. However, in chess, tennis, etc., there is no real benefit.

I wonder how normal stories like this are. Was this just one bad manager in one company, which you remember because of how surprising it was? If so, maybe the core claim that software is a meritocracy holds.

Or are stories like this common and everywhere?

Or are stories like this common and everywhere?

Very much so. Bad management is afraid of competence, because good performance cannot be argued away. I've seen this.

Not a genius but had something similar happen. Was literally doing the work already and successfully. My coworkers and boss all vouched for me but the company had a policy. There are probably other companies where a similar policy that effects single person that works there.
Super common. More common than the opposite.

In fact a version of the above story has happened in the last two companies I worked for.

It's not uncommon.

But there is sometimes more going on than meets the eye. If it's an A.I/M.L startup, the number of PhDs is a significant factor in fund raising. Being able to say "we have 6 PhDs on staff" will open doors.

Recruiters/managers are very picky about the Masters/PhD requirements, even if the corresponding day-to-day work does not necessarily warrant it.
It's normal in my experience. It's not necessarily that these managers are stupid: they're just working in what they perceive to be their own best interests.
In the general case it's impossible to objectively evaluate the merit of knowledge workers because so much of what they do can't be quantified. But a good manager can make reasonably accurate subjective rankings.
It is often the case, and very unfortunate.

Also besides a 'written job description' -- there are always unspoken 'tokens' for candidate's gender or, even, race, perhaps other attributes.

Studies on this type of subject are a taboo these days.

Promotions to different organizational levels in larger corps (with 20k employees or more, I would say) -- is like breaking through glass ceiling. You can see the other side, you can do the job on the other side -- but you cannot get through there...

Scott Adams, Dilbert author, had a personal experience about a promotion he was sharing. [1]

To me, it resonated, because it was closely matching my own at some point in my career.

[1] https://youtu.be/ddBRiUsL_5I?t=2041

As a machine learning freelancer, I can definitely understand the main argument of this article. When I started applying for contracts I always felt overwhelmed by the competition with math PhDs and other developers with brighter pedigrees. After closing my eyes to the competition, I started offering high quality communication, clear project specifications and complete honesty in terms of project feasibility, just as I would for any regular software project. After over 30 successful projects, I can't say I developed an Ai that can beat Starcraft, but I definitely brought a lot of value to all my clients. If you are passionate about machine learning and really like the grind, just go for it. Work finds a way
I'm one of those PhD types, and I would prefer to work with someone that has your skillset over someone who promises the moon and is solely excited about methodology. Based on your focus, I expect the projects you propose will have a measurable impact to the bottom line. And that is what builds reputation.
Hi!

Would you mind characterizing briefly, or giving examples of, the successful projects you've taken on as a freelancer?

I've thought before about trying to go that road, but the fact is, for almost every ml project opportunity I've come across, I would have had to say, "well, I can try out some stuff, and it might prove extremely valuable, but just as likely as not my POC will not perform well enough to use in production".

Not only is this the right attitude of what I'd want if I sought your services as a client, but this is heartening to hear as someone who's been thinking about getting more into this. Good on you!
This is a repetition of “tech is a meritocracy” that ignores all of the previous reasons that demonstrate why it isn’t.

There’s the nepotism portion (you need to get an interview first, before demonstrating merit, and nepotism bypasses a lot of that, and has the “they’re really good, just haven’t done much publicly/real world yet” backing).

Ignores that merit itself is not objective - just doing X doesn’t give an objective measure for how much merit X should get, and there’s a long history of evidence for non-merit/skill based bias in how much merit is “earned”.

Etc, etc

"Like the field of software development, the application of machine learning is a meritocracy. A meritocracy is a structure under which participants are valued based on their contributions or demonstrated achievement (merit)."

So far we've figured out that software development clearly isn't a meritocracy.

I think people aren't super clear on what they expect from meritocracy, this leads to statements like yours.

Speaking personally, I grew up in an incredibly deprived part of my nation, I have little in the way of formal education and I definitely do not have connections. However I have achieved incredible social mobility due to technology. Why? because there is _some_ level of meritocracy going on even if it's not entirely widespread through every single hiring manager or company and I have a deep unyielding passion for this field that drives me to continually educate myself which despite my poor connections and education is looked at favourably.

I still think of meritocracy as a noble goal, and I feel that merit is certainly definable if that's the main issue (IE; points based on time spent contributing to open source and things like invitations to speaking engagements and the like).

I genuinely do not buy the argument that having "merit" in other areas infers that you have license to be a toxic asshole; because at (least in my mind) part of merit is being able to work with others. You don't have "merit" if you can't communicate effectively.

In fact, there is an old saying about that: "Lots of MIPS but no I/O" which would make for a shitty CPU. Why would it make for a non-shitty developer?

As someone who was never considered prestigious enough for my first field (finance) who switched and has been successful in big tech, I can vouch it is way better than fields like finance and law
Compared to a lot of other fields, software development is such a screaming meritocracy, we're practically living in a real life utopia. Where else can you literally write your own ticket to a 200k+ salary without so much as a high school diploma?--IF you have the skills, that is. Compare medicine: you could walk on water and raise the dead, but good luck getting a position without certification!
I think the author is trying to lend support to people who feel excluded from the field due to a lack of credentials - great! You can, in fact, materially contribute and succeed without a PhD. Machine learning is a pretty big tent.

Claiming that a profession is meritocratic, though, strikes me as a harmful framing. From my vantage I don't see a) what an objective notion of merit would be and b) that all people who have the capability to contribute is strictly equal to the number of good jobs that are currently available.

I can definitely see when someone is already doing something useful well, but not if they might do that later or if the notion of useful will change.

Merit is relative. For example, it shows a lot of merit that you successfully completed a PhD program in machine learning. However, for some reason, there seems to be this constant need to discount people with academic qualifications when most of what machine learning is comes directly from there. For example, just because you write cool, useful applications in machine learning, doesn't mean you are breaking ground in the field. It means you are using other peoples knowledge effectively and there's absolutely nothing wrong with that and his highly necessary. But don't confuse it with making contributions to the theory itself.
Great point. If someone without credentials were truly doing ground-breaking stuff in the field, then getting a PhD should practically be a formality--schools should be banging the door down trying to recruit that person just for the status of having him or her (with full tuition waiver plus generous stipend being the bare starting ground).
I think what he meant by meritocracy is that the merits from doing projects hands on and having solid results, as opposed to the “scholastic” meritocracy which I had initially assumed he meant (and which may be confusing other readers too).
normal people call this an "empirical discipline"
Exactly
meritocracy is a structure under which participants are valued based on their contributions or demonstrated achievement (merit).

Of course what is considered meritorious or an achievement is almost entirely dependent on the stakeholders in power above you and rarely has any purely objective measure, and any objective measures are almost always project dependent, making it all a highly political game to figure out what's important to whatever system is judging you.

I would add that even some of the academic papers focus on the applied perspective of machine learning. Such examples can be found in papers issued by security research institutes/companies. And now that we have cloud computing to replicate/validate or even extend those research papers is really just the matter of determination even if the subject of the paper needs a whole cluster of distributed resources.
Says someone with a PhD. The advice in the text of just going into the field of ML blazing, is just weird. Every field is difficult, and you always have to see the light at the end of the tunnel. Personally I like to get the basics first, so yeah, I would read that required math first.

However, this just boils down to people's preference, are you a bottom up learner, or not. Both ways have yielded success for many different people. So what is the text actually saying?

Almost nobody will talk to you about ML positions unless you have at least MS in ML or have a large portfolio of recent SOTA implementations under your hood. More often a PhD in ML is a requirement for entry-level ML jobs with ridiculously low salaries (seeking enthusiastic doctorate holders etc.). Meritocracy is an illusion unless you do your own company and start competing in hottest areas - there your exit might be ridiculously big.
The author clearly abides by the old adage, “Tell them what you are going to tell them, tell them, then tell them what you told them.”
It's like a hamburger you see.
No it isn't. Applied machine learning is an arms race of who has the data.
There's a more interesting research race to figure out how to learn well with less data.
My question is if people are so good at math why are they doing machine learning, especially in practice?

Some very good people investigating it theoretically. But the best young math professors that I’m personally friends with actually do have something of the “elitist math attitude” and will tell me things like they think the ML field belongs in the 1600s while they successfully work in their obscure area of algebra applied to mathematical physics or something

I thought meritocracy was a discredited concept.
I’m always confused by this. Isn’t merit exactly what we want to be hiring for? I don’t understand why you wouldn’t want people who are the best at getting the job done. Or in opensource, I don’t understand why you would want to care about any attribute of a contributor other than the quality of their contributions. Anything else seems like discrimination.

And I acknowledge that we often don’t do a good job of this; but surely that doesn’t mean we want to move away from merit based decisions.

I understand why you would say that, but it's not an entirely discredited concept; except among a few people who seem to prefer identity politics.

I don't mean that to be inflammatory at all by the way; but there are people in the world who believe the circumstances of their place in society plays a significant role in how you should weigh them and their contributions.

So, I wouldn't jump to "discredited" just because there's one blog post from a surprisingly influential person. I would instead say "contested".

You can't discredit it universally, that'd be like disproving God.

More exactly, it is the psychological trait "Belief in a Just World" (BJW.)

One doesn't have to look far in our actual civilization to see innocent victims, injustice, but that still doesn't penetrate the existential strata of the universe eventually righting all wrongs.

Here's a good analysis of the psychological studies: https://www.youtube.com/watch?v=aQ0lR23T7FA

If you don't have time to watch the video, the most significant points are:

1) Believing in a Just World / Meritocracy, is an evolved emotional buffer to the chaos and reality of ever-present injustice.

2) BJW trait correlates with higher measures of happyness after traumatic events like natural disasters - and seems to be a trait that remains relatively static among persons regardless of the tradgedies they suffer or windfall they receive.

Well, if it's Open Source / free software, as the ML research field is, then it is a meritocracy.
Have seen jobs where PhD is "required". (Many data scientists tend to push this view as they feel it protects them and gives them elevated status).

Am also in the middle of training a group of future data scientists. The company is having issues hiring, and believe it is easier to train SME's in data science than the reverse.

Mmkay. It's an applause light. What are we even suppose to think/say?

https://www.greaterwrong.com/posts/dLbkrPu5STNCBLRjr/applaus...

Access to datasets makes machine learning not a meritocracy. Saw a recent talk at an insuretech conference where the speaker lamented that China will leap ahead in AI research since they gather more data about users than anyone else. From that perspective, things like GDPR hinder EU researchers.
Meritocracy all the things!
In this post I learned...