Seeking out fellow ML scientist bro/sis here on Blind. Having a mid-cliff crisis now. Punch line is that I think this role will be commoditized out of existence in at most 10 years. First, we need to get our definitions straight, because terminology varies. 1) Research Scientist (50% basic research, 50% applied research): These are the scientist working at Deepmind, OpenAI, FAIR. It's basically an academic research job that pays industry salary. PhD hard requirement. Publish papers or perish. Can only read/write research papers. But also have some encouragement to monetize their tech by introducing it to products. 2) ML / Applied Scientist (50% applied research, 50% engineering): This is my role, and we're halfway between Research Scientist and Engineer. PhD soft requirement. Not necessary to publish papers (but bonus points if you do). Primarily role is to translate ML basic research into some product. Comfortable in both reading research papers and implementation. Definitely does lots of proof-of-concept, some engineering work. But the maintenance engineering work is not done by them. No on calls. 3) AI / ML Engineer (25% applied research, 75% engineering): This is basically a SWE with ML flavor or specialty. Builds and maintains AI/ML products. PhD non requirement. No paper publication (and probably discouraged). Doesn't follow latest research and use plug-and-play methods or work with Scientist to incorporate latest tech into their product. Have on calls. 4) Data Scientist (75% statistician, 25% engineering): This seems to be the modern definition of Data Scientist, basically a re-titled statistician. PhD soft requirement, and is the statistician counterpart to the ML / Applied Scientist. Might do some ML, but mostly leans towards statistics side, so think A/B testing, multi-arm bandits, etc... every day. Like ML / Applied Scientist may do some proof-of-concept, some light engineering... but no large scale maintenance engineering work. No on calls. 5) Data Analyst: Most Data Scientist titles today are actually the Data Analyst role. This has usually the lowest TC. Most of them are SQL monkeys, and don't code. If they do ML work it's 100% commoditized at the GUI level. Now, the ML / Applied Scientist role used to be the "hot stuff" back in the days (before 2016-ish). That was because ML especially DL models were difficult and needed large expertise. You can see this in Amazon Applied Scientist role that pays higher than SWE, and Microsoft Applied Scientist role that pays equal to SWE. However, I see that this role will get commoditized out and in fact Google/Facebook is already showing us that. Why? 1) All big tech companies including Amazon and Microsoft are developing tech to "democratize" AI and ML, stuff like AutoML. This means that our job will eventually be pushed to the AI/ML engineer level and maybe even as far down to the Data Analyst level. 2) Research Scientist role will always remain in big tech, but even mid-size tech like Pinterest probably cannot sustain a large research workforce. Even Uber sold off Uber ATG which is a research organization. So while it is true Research Scientist will have highest demand and be on top of the food chain, but only the elite will make it and only in *some* not all FAANG level companies. Meaning you must be a student of Lecun, Hinton, Fei-Fi Li, etc. 3) Because of reason 2, that means that ML / Applied Scientist role will get commoditized out of big tech eventually. Why? Because for all standard applications, just push it down the AI/ML engineer. For non-standard applications, you could also get the Research Scientist to partner with the engineering team directly. Why is there a need for the ML / Applied Scientist? I'd argue in the past it was because there was lots of non-standard applications, but that's shrinking with commoditization. In fact, in both Google and Facebook, there is NO ML / Applied Scientist role. I believe this is going to the future. With the exception of Amazon and Microsoft... I have yet to find anyone who has a ML scientist role (as an official job family that is different from SWE). 4) For the not giant-tech companies, they will be increasingly reliant on AutoML and big tech commoditized tools. It is true they may have non-standard applications, but that will shrink over time. For example, I know Zillow has a small applied ML research team, but that's small and also TC is not good. ML bro/sis, do you agree with my analysis? I come from a research background with PhD (but not CS, physical sciences) and I like being the part-time scientist and part-time engineer, but I feel that I am living on borrowed time, as my role is doomed to extinction. What path forward do I have? 1) Upgrade to Research Scientist. Unless you *had* very good research pedigree this is not going to happen. Also, you have been years out of the publishing game. You can't go back. For my case, I could never had made it as a Research Scientist. Not because I lack talent (I believe in the right environment and ramp up I can make it), but because I don't have the right pedigree. Not whining about it, just laying out the facts. 2) Move laterally to AI/ML Engineer. I see this as our only option out. I will be sad as I get to do less research, but at least I will not be made irrelevant. 3) Keep trying to stay in the ML / Applied Scientist role, and keeping seeing the job market of this category shrink. Do I even want to stay on the sinking ship? 4) Help me figure out if there is something else besides these 3 options! #facebook #google #uber #lyft #microsoft #amazon #apple #netflix L64, $280k, PhD+5 yoe
Following
Great analysis
Yeah we're all fucked
a lot of stuff is getting commodotized even in software engineering...
Not at the rate of DS, it probably has something to do with quantifiable benefits for each role. SWEs bring more value to the table in spite of all the "no code"/ "low code" tools that have existed since 2001. (I'm a DS)
Probs will become a DS Manager and do lame business stuff
Sad but true 😭. Worse case scenario we can just become a mgr or move to SWE tho
I’m an SWE who wants to move to DS/AS lol
At Amazon, Research Scientists are Applied Scientists who did not pass coding bar of SDE 1 during interview. Tasks are the same :)
You can always go back to academia, earning a non-absurd amount of money.
Spot on mate!