Tech firm creates robot that bests humans in reading comprehension

Scope News ^ | 1.15.18 | Chris Agee

[Lazamataz]

3) there are so many perverts in the world that the second you take your eyes off those poor robots someone will be molesting them. The first law of robotics stating that “A robot may not injure a human being or, through inaction, allow a human being to come to harm.” will have to be updated so the robots can do a little pervert ass-kicking.

I'm a robosexual.

DON'T YOU JUDGE ME.

[Fedora]

What human is going to still be awake after 100,000 questions?

[Myrddin]

The new deep learning paradigms can do this easily. The simplest training (supervised learning) is good when you fully understand the in/out relationship. The next step is unsupervised learning where the neural net and models work out the mapping. The more interesting case in reinforcement learning. You feed data in, but only give the neural net "praise" for correct answers. It derives all the model and algorithms on its own. It is free to execute millions of trial/error attempts seeking reinforcement.

I'm emusing myself building deep learning models using python3, keras framework, TensorFlow and nVidia GPU processing. Easily accessible resources on a modern laptop with an nVidia graphics card. Manning Publications has a nice early access book detailing how to build a deep learning implementation to play the "Go" game. They also offer a video tutorial "Keras in Motion" to bring you up to speed on practical application.

My co-workers were doing good work with hardware neural networks in the early 90s. That hardware has been gathering dust for years. The brave new world is big data under the aegis of Google and Amazon for training input and big data centers with racks of GPU hardware to crunch the neural network data. Expert systems and more will become very affordable.