I cannot fathom the number of times I have heard people go crazy with the thought that their degrees might be useless because hey,someone invented computers and now that's killing humanity.Stop! Just stop right there.Get all the science fiction out of your head and get down to real science.Okay?
I get the recent buzz of Indian IT giants citing 'automation' as the reason for job cuts,but Anyone who has been a part of these companies knows that it is nothing but a gimmick.The work in these companies involves clocking hours and quoting the number of trained and available resources at hand for offshore IT projects.I have friends who are working for top IT companies since last 1.5 years and all they do is watch educational videos all day and clock out to maintain those numbers (read manpower).So possibly,the 'job cuts' are a way to compensate for losses incurred during demonetization and GST restructuring.
To get into the details of it,Lets start by describing any basic Artificial Intelligence engine.It has the following components:
1: Agent
2: Sensor
3: Actuator
4: Environment.
For any purpose,an Agent perceives inputs from its environment using sensors and acts on the environment through actuators.A feedback loop is involved too.
It is very imperative to note here that machines do not learn the same way as humans do.
Learning happens in two modes,supervised( A kid being taught to write) and unsupervised( A kid learning to walk).
In both these modes,its impossible that the trained AI agent would surpass known human intelligence on all known and unknown levels simply because the environment and the sensors isint, and will never be,100% perfect.
Think of it this way,if you hire a language tutor and that person has got some grammar wrong,you would inherit the same by virtue of being trained by him/her.
All in all,an AI cannot be perfect till the environment is in a steady,non transient ,non-uncertain (See,this is not possible again due to Heisenberg's uncertainty principle).
Having said that, I am not denying that AI cannot pick up some undesired traits like biases if it exists in large volumes of data.In fact,it is a part of the training,fine tuning the model until its finally right involves letting it wander where it does and then applying corrective measures instead of hard coding the way early on. These traits,even if actionable,can be detected early on.It is exactly why the whole of medical science would not exist if our health was perfect perpetually.
I plan on writing one more post about this topic,drop in comments about what points would you like me to address.
I get the recent buzz of Indian IT giants citing 'automation' as the reason for job cuts,but Anyone who has been a part of these companies knows that it is nothing but a gimmick.The work in these companies involves clocking hours and quoting the number of trained and available resources at hand for offshore IT projects.I have friends who are working for top IT companies since last 1.5 years and all they do is watch educational videos all day and clock out to maintain those numbers (read manpower).So possibly,the 'job cuts' are a way to compensate for losses incurred during demonetization and GST restructuring.
To get into the details of it,Lets start by describing any basic Artificial Intelligence engine.It has the following components:
1: Agent
2: Sensor
3: Actuator
4: Environment.
For any purpose,an Agent perceives inputs from its environment using sensors and acts on the environment through actuators.A feedback loop is involved too.
It is very imperative to note here that machines do not learn the same way as humans do.
Learning happens in two modes,supervised( A kid being taught to write) and unsupervised( A kid learning to walk).
In both these modes,its impossible that the trained AI agent would surpass known human intelligence on all known and unknown levels simply because the environment and the sensors isint, and will never be,100% perfect.
Think of it this way,if you hire a language tutor and that person has got some grammar wrong,you would inherit the same by virtue of being trained by him/her.
All in all,an AI cannot be perfect till the environment is in a steady,non transient ,non-uncertain (See,this is not possible again due to Heisenberg's uncertainty principle).
Having said that, I am not denying that AI cannot pick up some undesired traits like biases if it exists in large volumes of data.In fact,it is a part of the training,fine tuning the model until its finally right involves letting it wander where it does and then applying corrective measures instead of hard coding the way early on. These traits,even if actionable,can be detected early on.It is exactly why the whole of medical science would not exist if our health was perfect perpetually.
I plan on writing one more post about this topic,drop in comments about what points would you like me to address.
/*not an engineer, not even in college, not an expert in literally anything. just a computer science (specially machine learning) enthusiast.. would be happy to get corrected*/
ReplyDeleteAn agent may never become 100% perfect.. but does it need to? Dont we just need an agent that is just more accurate than humans at only a specific task i. e. we just need a narrow AI (aka weak AI) that can give a better performance than people.. because people aren't perfect either..
Also robots may not completely replace people at their jobs but can surely reduce the no. of people involved massively.. An AI could check billions of possibilities within few seconds and provide dozens which are optimal.. A few experts could just remove the ones which overfit or are not feasible... Kinda when an AI designed a car when it was given big enough dataset of pressure point recordings in test runs by F1 racers...
We have robots creating an entire bridge from scratch using 3D printers themselves in Amsterdam.. Is it really accurate to say 'ever' with recent advances in quantum computing..
When we have strong supporters of singularity among us.. Is robots stealing jobs really that far-fetched?
Thanks for stopping by Siddharth.Exactly,a weak agent and a narrower one may become adept at a task and may also be able to perform it better than humans,but the question is can it do what humans cannot and get it right every single time? AI will definitely change the way people do their jobs (like it happened in industrial revolution),but it sure is not going to 'take over' those jobs. Overfitting is the bitter truth but we do not know which parts of the sample have caused overfitting. Even when we change the hyper parameter tuning and assume that a sample is a perfect representative of the population,what can be said that the population itself insist subject to changes with time and other things?
DeleteDo check out my other posts.Thanks :).
PS: Moving all my Tech posts to this,kindly follow here.
Deletehttps://wordpress.com/post/notabeautyblogindia.wordpress.com/32
The Citizen Titanium Dive Watch - AT the ITAC-BET
ReplyDeleteThe does titanium set off metal detectors Citizen Titanium fram titanium oil filter Dive Watch is ford fusion titanium 2019 a 2-in-1.5-channel digital video edc titanium display, featuring the ultralight titan metal and ultralight Vision in fine detail. The watch
v079r8ibgnv953 adult sex toys,vibrators,vibrators,dildos,sex doll,dildos,dildos,male sex doll,huge dildos b854h1ozzls510
ReplyDelete