The most prestigious law school admissions discussion board in the world.
|
 |
 
|
With machine learning, won't 95% of coders/programmers become obsolete?
Poast new message in this thread
Date: January 17th, 2017 5:17 PM Author: Black pea-brained dysfunction giraffe I think they would. How far away are we from that happening? (http://www.autoadmit.com/thread.php?thread_id=3494212&forum_id=2#32394225) |
 |
Date: January 17th, 2017 5:27 PM Author: low-t toaster preventive strike when machine learning advances to the point that an AI can write a broad range of programs, every other job can be automated away. you are basically asking how far away are we from artificial general intelligence (http://www.autoadmit.com/thread.php?thread_id=3494212&forum_id=2#32394266) |
|
Date: January 17th, 2017 5:29 PM Author: Pungent lime point Yea they will become obsolete the way printing press machinists became obsolete after that technology stabilized.
Already websites like facebook and twitter removed the need for tailored website building. |
|
Date: January 17th, 2017 5:36 PM Author: hairraiser gas station ape How long do you imagine? Should I try to get an MS in machine learning (http://www.autoadmit.com/thread.php?thread_id=3494212&forum_id=2#32394303) |
|
Date: January 17th, 2017 5:38 PM Author: impertinent mentally impaired boltzmann piazza these skills will still be useful for data analysis etc. though (http://www.autoadmit.com/thread.php?thread_id=3494212&forum_id=2#32394308) |
|
Date: January 17th, 2017 5:39 PM Author: boyish dog poop ticket booth Computers do data analysis better than anyone (http://www.autoadmit.com/thread.php?thread_id=3494212&forum_id=2#32394314) |
|
Date: January 17th, 2017 5:40 PM Author: impertinent mentally impaired boltzmann piazza you need to know what questions to ask etc
and getting the computer to tell you what you want, i assume, will require some knowledge of how the computer is giving you what you want
computers can't "just do data analysis" |
|
Date: January 17th, 2017 6:45 PM Author: boyish dog poop ticket booth They'll get stats PhDs who can code to do that stuff. 99% of the grunt work will be automated and presented to consumers in templates for their businesses (http://www.autoadmit.com/thread.php?thread_id=3494212&forum_id=2#32394673) |
|
Date: January 17th, 2017 6:53 PM Author: impertinent mentally impaired boltzmann piazza there is always opportunity for value-added work when the people doing the creative tasks are more knowledgeable about the nuts and bolts because they can find new ways to use the nuts and bolts, and to use them more efficiently.
and, in fact, this usually makes knowledge of the nuts and bolts a requirement over time. insofar as people know there's value in knowing the technical skills in industry x, people planning to go into industry x get those skills. many, many industries now see knowledge of at least basic data manipulation and analysis as a basic requirement and weren't requiring it only recently.
it's not so much as the creative workers just feeding the information to the stats phds, but whatever applicable skills the stats phds have becoming required.
regardless of who is touching the nuts and bolts the point is that there is a space in which you need a human to be manipulating things, and insofar as this is true, and insofar as the capabilities of the nuts and bolts are flexible, it's valuable to be able to have whatever knowledge is needed to work with them. |
Date: January 17th, 2017 5:38 PM Author: boyish dog poop ticket booth 5 years away tops. The rate with which this tech is improving is increasing (http://www.autoadmit.com/thread.php?thread_id=3494212&forum_id=2#32394311) |
|
Date: January 17th, 2017 5:52 PM Author: fishy aphrodisiac national security agency machine learning isn't magic.
right now the most advanced machine learning applications requires HUGE data sets and massive processing power to train an application to do a very specific task. a lot of machine learning requires intense human intervention. this is why it's taken years and years for ibm's jeopardy-winning watson program to do anything useful outside of winning jeopardy--there is no magic ai. ibm just handcrafted a really really good jeopardy-playing program; you can't just feed medical journals into "watson" to make it spit out the right answers. |
|
Date: January 17th, 2017 6:42 PM Author: low-t toaster preventive strike this is increasingly not true. there has been a great deal of progress recently in semi-supervised and one-shot learning. it isn't 2011 and we can do a lot better than simply feeding a few million labelled images into a convnet if we want it to do something useful. a lot of the promising work done on reinforcement learning requires no human input too.
neural language models have only recently become any good, so it shouldn't be a surprise that watson hasn't done much. the success of the neural translation models is a very encouraging sign. |
|
Date: February 22nd, 2018 12:59 PM Author: azure rough-skinned twinkling uncleanness national where do you see machine learning being in 10 years?
what applications do you think it will have for business problems? |