Imposter Syndrome (in Data Science)(brohrer.github.io)
brohrer.github.io
Imposter Syndrome (in Data Science)
https://brohrer.github.io/imposter_syndrome.html
12 comments
Data science is one of the most imposter-filled "professions". It's a recently-established category of worker that falls across multiple disciplines and is very effected by technological progress. I have met "data scientists" who aren't really good at any aspect of it, but they still get by because of the supply/demand and lack of any existing expertise to say "hey, you know, this person we hired is barely competent and just googles everything we ask of them"
All this means is that we will inevitably reach a point where separate titles are used, and 'data scientist' will means about as much as 'engineer'.
We will distinguish 'machine learning specialist data scientist' from 'database specialist data scientist' just like we distinguish 'electrical engineer' from 'lab systems engineer', for example.
Then, we might have a term for generalists like 'data science technician'. And by then, the people who 'aren't really good at any aspect of it' and can't really function as generalists will be naturally sorted out because they can't really fit into any of those titles
We will distinguish 'machine learning specialist data scientist' from 'database specialist data scientist' just like we distinguish 'electrical engineer' from 'lab systems engineer', for example.
Then, we might have a term for generalists like 'data science technician'. And by then, the people who 'aren't really good at any aspect of it' and can't really function as generalists will be naturally sorted out because they can't really fit into any of those titles
The thing is that we already have/had that. Data science is really just a blend of statistics, machine learning, data engineering, and software development. I think the data science explosion is the result of more people/companies wanting to hire a single person to do all of these jobs rather than individuals for each.
If we ever go back to a world where we distinguish people by these specialties we would basically just be going back a decade to where we had statisticians focusing on statistics, data engineers maintaining databases, software developers creating the prodect, etc. (which isn't necessarily a bad thing).
If we ever go back to a world where we distinguish people by these specialties we would basically just be going back a decade to where we had statisticians focusing on statistics, data engineers maintaining databases, software developers creating the prodect, etc. (which isn't necessarily a bad thing).
You hit me right in the feels, bud.
This describes my situation to the point. I have worked in big unicorns and have deployed many ml based models in production which had moved the numbers significantly while many a data scientists in our team just kept cribbing about errors in data or scarcity of it.
I have no DS background, am a humble engineer but believe it's 10x better to just work with whatever you have available and get sit done.
I have no DS background, am a humble engineer but believe it's 10x better to just work with whatever you have available and get sit done.
Entirely depends. "Moving the numbers significantly" doesn't mean much if the numbers aren't moving in the correct direction. Errors and scarcity in data are real, significant problems. I'm a statistician working in data science, and I can't count the number of times people complain about "why can't you just work with what is available" while failing to understand that what is available is total garbage.
You can dress up bad data in any number of ways to get results that sound and look pretty. I see this all the time. Sometimes you get lucky and the model is ok regardless. Lots of times the model performance isn't great, and it is later assumed there are other outside issues to blame, or the project is redone for the umpteenth time. Ocassionally you will have colossal failures that do real damage.
Keep in mind that when a poorly designed machine fails and kills dozens, or the financial system of the world crumbles under the weight of terrible loans and convoluted financial instruments, or millions of people's personal info gets hacked due to terrible, antiquated security systems, and everyone starts asking "How could people be so stupid to let something like this happen?", the answer is almost always executives, management, or "humble engineers" sweeping the problems they don't like under the rug because they believe "it's 10x better to just work with whatever you have available and get shit done".
You can dress up bad data in any number of ways to get results that sound and look pretty. I see this all the time. Sometimes you get lucky and the model is ok regardless. Lots of times the model performance isn't great, and it is later assumed there are other outside issues to blame, or the project is redone for the umpteenth time. Ocassionally you will have colossal failures that do real damage.
Keep in mind that when a poorly designed machine fails and kills dozens, or the financial system of the world crumbles under the weight of terrible loans and convoluted financial instruments, or millions of people's personal info gets hacked due to terrible, antiquated security systems, and everyone starts asking "How could people be so stupid to let something like this happen?", the answer is almost always executives, management, or "humble engineers" sweeping the problems they don't like under the rug because they believe "it's 10x better to just work with whatever you have available and get shit done".
same here, I haven't explored all nooks and crannies of the "data science map", but all the ml/dl I deployed were a success. I still feel like a major imposter though.
As a data scientist I have wondered if this field is particularly suited to imposter syndrome. My formal background is economics, and every once in a while I become terrified at how little formal statistics I've studied, or large gaps in data structures etc. although I'm similarly surprised at how far I've gone by just going home and studying the basics when I run into something I don't know, and the gaps in knowledge some coworkers have in areas where I know more.
...although I have met a few genius data scientists who seemingly really can do everything. Although I'm pretty sure they are paid upwards of 300k.
...although I have met a few genius data scientists who seemingly really can do everything. Although I'm pretty sure they are paid upwards of 300k.
I'm pretty certain data science is worse than normal fields. It's probably due to the fact that huge proportion do have PhDs, but that a PhD is not required for the vast vast vast majority of what we do.
So for instance, I totally feel like a data science imposter, but in the last year have done the following: - Pushed a custom deep learning NLP model to production - Created and maintain company's ETL and data warehouse mechanisms - Performed statistical analyses to find ways to better target and increase customer engagement. - Implemented event tracking and performance metrics across products - A sales prediction product that has contributed to $~5M in incremental revenue
Somebody obviously believes in me since I've grown the team from just myself to ~6, but I also know that I've had dozens of past colleagues that would instantly disqualify me since I 1) don't have a PhD and/or 2) can't/don't read statistics/machine learning papers
So for instance, I totally feel like a data science imposter, but in the last year have done the following: - Pushed a custom deep learning NLP model to production - Created and maintain company's ETL and data warehouse mechanisms - Performed statistical analyses to find ways to better target and increase customer engagement. - Implemented event tracking and performance metrics across products - A sales prediction product that has contributed to $~5M in incremental revenue
Somebody obviously believes in me since I've grown the team from just myself to ~6, but I also know that I've had dozens of past colleagues that would instantly disqualify me since I 1) don't have a PhD and/or 2) can't/don't read statistics/machine learning papers
I tend to agree. I think it's because the field uses statistics, which most people have decided are incomprehensible and don't even try to understand. Combined with how bad our brains are at thinking statistically, and you have a powerful desire-for-avoidance by non statisticians.
Combine with the value that good ones can provide, and you have a perfect situation where a boss or peer just thinks thar be dragons in the work sphere of the statistician.
Combine with the value that good ones can provide, and you have a perfect situation where a boss or peer just thinks thar be dragons in the work sphere of the statistician.
I think the issue lies in the fact that data science is a massive umbrella term which really covers multiple large fields, namely statistics and computer science, both of which by themselves are massive. Its very difficult if not impossible for people to feel like they have enough expertise in all of the subjects that fall under the umbrella of "data science". Which typically leads to lots of self-teaching, learning on the fly, and hacking solutions together, which is often the source of all impostor syndrome feelings.
I enjoyed this Partially Derivative podcast where two of the podcasters discuss their experience with imposter syndrome
http://partiallyderivative.com/podcast/2017/03/06/badasses-f...
http://partiallyderivative.com/podcast/2017/03/06/badasses-f...