4 ms·
As a data scientist, I have the lingering fears of job automation. At least for applications-oriented work (as opposed to research). There's no reason Auto ML s
by wakkaflokka 9y ago
As a data scientist, I have the lingering fears of job automation. At least for applications-oriented work (as opposed to research). There's no reason Auto ML software couldn't take away most of the difficulty of model building in simpler scenarios. I mean, the software is already here. Just not widely used. And as computing power increases, no reason they can't tackle more of the complex problems where you might need to call in an expert. Feature generation based on the context of the industry, etc. Translation of findings into business-understandable language. One new algorithm or software package can automate much of what a data scientist might do. I mean, the unofficial slogan for some of the data science positions I've found were to "automate myself out of a job."
That's why, although I LOVE what I do, part of me thinks I need to make sure I'm in a position to move into management at some point - because I see a lot less ability for automation in areas that require "people" skills over the next few decades. On the other hand, I LOVE what I do. What's that French phrase? C'est la vie? I think it translates into "you're screwed, one way or another."
- amelius 9y agoMachine Learning today is where web-development was twenty years ago. As more tools become available, everybody will be able to do it.
- xiphias 9y agoIt's true, but web dev jobs didn't go away...quite the opposite.
- mgkimsal 9y agoand yet - most people still can't do it (webdev), or do it poorly, the expectations are higher, often the stakes are a bit higher, and many people try to skimp out on the lowest priced vendor (or DIY). Will we see those same trends in ML over the next 10-20 years?
- byte1918 9y agoI feel general ML is already extremely accesible. There are tons of frameworks that trivialize impleneting various types of classifiers. I am actually surprized it's not more popular. I think the only thing left is to make GPU processing more available or find a cheap alternative to expensive Titan cards (maybe Google's NPUs?) and not really a requirement but abstracting away some of the configuration parameters could help the general population that doesn't know or care about the math behind. I can already see a future, a dark future, where pre-trained models on pictures of people are sold on darkweb and you can just plug the model and hook it up to some live cam in your area and stalk anyone. This is almost if not already possible today and I can't imagine Facebook not already having something caple of this.
- rbanffy 9y agoI'd really love to see some GPU-like capacity integrated into the main CPU ISA so one could be certain to find those instructions and cores as you can be sure you'll find AVX.
- gaius 9y agoSuch a technology is the McGuffin in the last two Fast & Furious movies where it’s called God’s Eye (or The Devil’s Bumhole, by one character)