AI is Starting to Create it's own Machine Learning Programs
Google's Artificial Intelligence research group used human created AI design it's own machine learning software. This newly created machine learning software even beat out the original program on a language processing benchmark by using 800 graphics cards.
The concept of AI making AI deeply scares invokes images of Skynet for some with others having concerns of machine learning professionals creating themselves out of a job. To try to prevent things from running wild, the Google team discourages it from disabling their own kill-switches during training. Many of these software professionals aren't concerned with job security, viewing this as a way to free themselves up to work on more interesting tasks while the AI finds the most efficient and accurate designs.
The hope is that this will help complete projects and bring them to market in a much more timely fashion. The benefit over human designs is that the AI creator programs are able to search across multiple potential designs at the same time to find the best ones.
According to MIT Tech Review, Jeff Dean, the leader of Google Brain research group, such efforts can increase the pace of the implementation of the machine-learning software in various fields of economy. It should be noted that companies pay a premium salary to the machine learning experts–a class of experts that are in short supply.
-Source
We may end up seeing the trend of creating the better AI designer programs instead of the end products themselves in the future.
Additional Reading:
Articles
Google Is Making AI That Can Make More AI
Google Is Working on Plans to Prevent a Skynet Situation
AI Software Learns to Make AI Software
Publications
Safely Interruptible Agents
Neural Architecture Search with Reinforcement Learning
Designing Neural Network Architectures Using Reinforcement Learning
Are you new to Steemit and Looking for Answers? - Try https://www.steemithelp.net.
Image Sources:
Robot Typing on Keyboard
Technology always get an increase every year very well, @sykochica thank you for sharing
This post has been ranked within the top 25 most undervalued posts in the first half of Jan 30. We estimate that this post is undervalued by $18.91 as compared to a scenario in which every voter had an equal say.
See the full rankings and details in The Daily Tribune: Jan 30 - Part I. You can also read about some of our methodology, data analysis and technical details in our initial post.
If you are the author and would prefer not to receive these comments, simply reply "Stop" to this comment.
I think I read about something similar relating to translation algorithms at the end of last year. The AI had created it's own base language that was a way for it to translate between any two unrelated languages if I remember correctly. The engineers weren't exactly sure how it had done it as they had not programmed the AI to actually do that specifically.
I think that is one step away from AI creating new AIs!
https://twitter.com/Soul_Eater_43/status/825882279209742337
That's pretty cool! I don't think I caught that one, but definitely going to do some reading up on that! It's a pretty good idea for it to make it's own base language, especially without the humans telling/making it do so.
TY!
Yes:)
Hallo my mentor, great post !
Thank you! This one was a bit shorter than usual, but I've got some more in depth ones in the works and hopefully posted in the next day or so.